<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Rabih Ibrahim on Medium]]></title>
        <description><![CDATA[Stories by Rabih Ibrahim on Medium]]></description>
        <link>https://medium.com/@RabihIbrahim?source=rss-845e11680e3b------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*lt1m8dfuj-Dt7D9DYG3qNQ.jpeg</url>
            <title>Stories by Rabih Ibrahim on Medium</title>
            <link>https://medium.com/@RabihIbrahim?source=rss-845e11680e3b------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 01:52:36 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@RabihIbrahim/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Futures We Imagine]]></title>
            <link>https://medium.com/@RabihIbrahim/the-futures-we-imagine-16db44910fcf?source=rss-845e11680e3b------2</link>
            <guid isPermaLink="false">https://medium.com/p/16db44910fcf</guid>
            <category><![CDATA[design-and-tech]]></category>
            <category><![CDATA[futures-thinking]]></category>
            <category><![CDATA[radical-reflections]]></category>
            <category><![CDATA[foresight]]></category>
            <category><![CDATA[cognitive-science]]></category>
            <dc:creator><![CDATA[Rabih Ibrahim]]></dc:creator>
            <pubDate>Tue, 09 Jun 2026 06:13:10 GMT</pubDate>
            <atom:updated>2026-06-09T06:13:10.844Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7bAnPiYCO0heoxWDwcIqWg.png" /></figure><h4>Thought Exploration Series</h4><h4>From daydream to civilization — why we cannot stop imagining the future</h4><h3><strong>The Mind That Lives Ahead</strong></h3><p>Without even trying, you are already living in several possible futures. Right now, beneath this sentence, your mind is rehearsing what comes next — conversations not yet had, decisions not yet made, tomorrows not yet arrived. You don’t choose to do this. It simply happens.</p><p>This is part of what makes us human.</p><p>We often describe ourselves as toolmakers, storytellers, dreamers. All of these are true. We are also creatures with a quiet, relentless capacity to inhabit worlds that do not yet exist. We are, in a profound sense, beings who live partly in tomorrow.</p><blockquote>“Memory is a system that allows us to travel mentally through time.” — Endel Tulving</blockquote><p>The cognitive psychologist Endel Tulving called this “mental time travel.” The term sounds poetic, but it describes something remarkably precise. To imagine what has not yet happened, the brain retrieves fragments of memory, recombines them into new configurations, estimates probabilities, anticipates consequences, and uses those simulations to guide action in the present. We are not passively waiting for the future to arrive. We are rehearsing it, continuously.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dYmEG8DdCPvlAcXSQ8l2PQ.png" /><figcaption><strong>AI-Generated Artwork</strong> — <em>Created using Midjourney</em></figcaption></figure><p>Memory and imagination turn out to be the same system. We do not remember the past simply to preserve what has been — we reconstruct it, reinterpret it, sometimes reimagine it entirely. And we do not imagine the future from nothing — we build it from fragments of everything we have already lived. The same cognitive machinery that reconstructs yesterday also rehearses tomorrow. Past and future are not opposites. They are two directions the same restless mind keeps exploring.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hQL0wOD_AvPa-6S3wntlpw.png" /><figcaption>Diagram exploring relationship between memory and imagination</figcaption></figure><p>This inheritance shapes everything. Much of what we call civilization can be understood as the accumulated materialization of imagined futures — cities, laws, institutions, technologies, all built twice: first in the mind, then in the world.</p><p>But here is what makes this moment different from any that came before: we are now building tools that can simulate futures alongside us, faster than we can fully absorb. The interval between imagining and experimenting is shrinking. And yet the future feels no more predictable — if anything, less so.</p><p>That paradox is where this exploration begins.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TX3u4cF_5Lsff5BGniuXqg.png" /><figcaption><strong>The Future of Everything Festival (2019)</strong> — <em>The Wall Street Journal.</em> | <strong>Is This the New Normal?</strong> — <em>AI-themed poster featuring DALL·E 2-generated imagery (2022).</em> | <strong>The Future Is Now</strong> — <em>Poster by KRAW Graphics (2022).</em></figcaption></figure><h3><strong>The Worship of Tomorrow</strong></h3><p>The tension between our hunger to shape the future and our inability to control it has always been at the heart of the human story. What changes across history is not the impulse itself — it is the scale and speed with which we act on it.</p><p>Human beings have always sought to extend themselves beyond the limits of the present. Not just privately, in the quiet simulations of an individual mind, but collectively — through culture, through belief, through the material world we build around us. And no expression of that impulse has been more tangible, more visible, more consequential, or more revealing than technology.</p><blockquote>“The most human thing about us is our technology.” — Marshall McLuhan</blockquote><p>McLuhan saw something that still feels radical: technology is not separate from humanity. It is one of its most characteristic expressions. We do cognitively through imagination what we do materially through tools. A spear extends the reach of the arm. Writing extends memory across generations. The telescope extends vision beyond the limits of the eye. The computer extends calculation — and the vast ramification of that is still unfolding around us.</p><p>Each invention is, at its core, a refusal to accept inherited limits.</p><blockquote>“We shape our tools and thereafter our tools shape us.” — Marshall McLuhan</blockquote><p>This is the loop that drives history. We imagine a capability, we build it, and then it reshapes the conditions in which the next generation imagines. Technology is not just the product of future thinking — it is its most visible residue. When people picture tomorrow, they picture machines and systems because technology is where imagination becomes most tangible. It is where the future stops being a feeling and starts being a fact.</p><p>Generative artificial intelligence is the latest — and perhaps most consequential — turn in this long story. It adds something genuinely new: the ability to synthesize vast bodies of human knowledge into scenarios, strategies, and speculative possibilities in real time. It is not merely a tool for retrieving what is already known. It is a partner in exploring what might come next. Whether it ultimately expands human imagination or overwhelms it with endless recombinations of existing ideas remains an open question. But its emergence marks a significant new chapter in humanity’s effort to externalize and amplify the mind itself.</p><p>At certain moments, this orientation toward tomorrow becomes more than practical. It becomes devotional.</p><p>Few movements captured this more vividly than Futurism. In 1909, Filippo Tommaso Marinetti published the Manifesto of Futurism on the front page of Le Figaro. The manifesto celebrated speed, machinery, electricity, and the disruptive energy of modern life. The machine became more than a practical invention. It became a symbol — of humanity’s desire to break with inherited constraints and participate directly in the creation of a new world. The future was no longer something that would simply happen. It had become an aesthetic and moral project.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KixOI0y973HEIm5gN5ibUQ.png" /><figcaption><strong>Unique Forms of Continuity in Space</strong> — <em>Unique Forms of Continuity in Space by Umberto Boccioni (1913).</em> | <strong>The Futurist Manifesto</strong> — <em>Published by Filippo Tommaso Marinetti in Le Figaro (1909).</em> | <strong>Aeropittura</strong> — <em>Futurist explorations of flight, speed, and technological modernity (c. 1920s–1930s).</em></figcaption></figure><p>At first glance this feels anti-human — cold logic displacing organic life. But look closer and it is something else entirely. The qualities the Futurists worshipped — precision, acceleration, transformative power — were not imposed on humanity from outside. They were qualities human beings had deliberately designed into their own creations. In celebrating the machine, the Futurists were celebrating a reflection of themselves: our desire to exceed inherited limits, to amplify our capabilities, to become more than we currently are.</p><p>The enduring significance of Futurism is not the specific world it imagined. It is the intensity with which it treated imagining itself as humanity’s highest task.</p><p>That impulse has never really left us. Much of contemporary technological culture still rests on the same assumption — that tomorrow can be made faster, smarter, more transformative than today. The language has changed. The underlying hunger has not. If anything we have learned to fear it — not the future itself, but our own relentless need to reach toward it, to build it, to become it. And to let it shape us.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JWqj8h4-3mKIJsX8tpzJtA.png" /><figcaption><strong>AI-Generated Artwork</strong> — <em>Created using Midjourney</em></figcaption></figure><h3><strong>How Civilizations Dream</strong></h3><p>“A long time ago in a galaxy far, far away.”</p><p>Six words that set a story in the past — and yet fired the imagination of millions about what the future could look like. That paradox is not accidental. It is how science fiction has always worked — and how human beings have always dreamed. We reach backward and forward at the same time, using everything we know to construct everything we can imagine.</p><p>The richest visions of the future are rarely found in technical reports. They emerge across the full range of human culture — in illustrated postcards, classroom conversations, speculative architecture, political manifestos, novels, and films. These artifacts are compelling not because they forecast events with precision, but because they reveal how each era attempts to think beyond the constraints of the present.</p><p>They show us societies thinking aloud.</p><blockquote>“Imagination will often carry us to worlds that never were. But without it we go nowhere.” — Carl Sagan</blockquote><p>When human beings imagine the future, they are not simply predicting what will happen. They are asking a recurring set of questions — about technology, society, evolution, and identity. The medium changes across eras. The questions remain remarkably consistent.</p><h4><strong>How Technology Reshapes Everyday Life</strong></h4><p>At the turn of the twentieth century, French artists produced a remarkable series of illustrations known as En L’An 2000 — In the Year 2000. Created for the 1900 Exposition Universelle and later reproduced as postcards, these images depicted automated classrooms, airborne firefighters, underwater transportation, and mechanized domestic life.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f2v3oKdGZQlbpbmZt4hnWw.png" /><figcaption><strong>Automated Housework</strong> — <em>The future as liberation from routine labor.</em> | <strong>Underwater Travel</strong> — <em>Exploring worlds beyond the reach of ordinary travel.</em> | <strong>Machine-Assisted Learning</strong> — <em>Knowledge delivered directly through technology.</em> | <strong>Personal Flight</strong> — <em>The dream of individual freedom in the skies.</em></figcaption></figure><p>What makes them so enduring is not whether they proved correct. It is what they reveal about the logic of imagination itself. Their creators assumed the future would extend the most visible technologies of their own era — gears, propellers, engines, industrial machinery. Yet the people inhabiting these extraordinary worlds are dressed in the formal clothing of the Belle Époque. The machines belong to tomorrow. The humans remain unmistakably rooted in 1900.</p><p>This small detail is surprisingly revealing. Technological change is almost always easier to imagine than social or cultural change. We can picture flying machines and automated schools while still assuming that people will look, behave, and live according to familiar conventions. The future is rarely invented from scratch. More often it is the present projected forward — selected elements amplified, while everything else remains untouched.</p><p>A similar pattern appears in a BBC segment recorded in 1966, in which children were asked to describe life in the year 2000. They imagined videophones, robotic assistants, automated household labor, advanced medicine, dramatic changes in transportation. Their predictions faithfully reflected the atmosphere surrounding them — the space race, television, suburban modernity, the growing presence of computers. They were not making detached forecasts. They were extending the world they already knew into adulthood.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FxS8xX3usi4c%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DxS8xX3usi4c&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FxS8xX3usi4c%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/30af2a96b0e91a3590669fe286071b4d/href">https://medium.com/media/30af2a96b0e91a3590669fe286071b4d/href</a></iframe><p>In both cases the future begins close to home. Before it becomes philosophical, it becomes practical. And we are genuinely skilled at extending existing trajectories — a carriage becomes an automobile, a computer becomes smaller and faster. What is far harder is imagining entirely new categories. The future most reliably surprises us not by improving known inventions but by introducing possibilities that lie completely outside the conceptual vectors of the present.</p><h4><strong>How Society and Space Might Be Reorganized</strong></h4><p>As soon as we imagine new technologies, a broader question follows: if our tools change, how should our societies change with them?</p><p>Architects felt this most viscerally. Figures like Antonio Sant’Elia, Archigram, Buckminster Fuller, and Lebbeus Woods produced drawings of floating cities, plug-in megastructures, and radically adaptive urban systems — many never built, yet all expanding the boundaries of what human environments might become. These were not engineering proposals. They were arguments about collective life, rendered in spatial form.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jBdkihjc7C3Stwfrv0Pqyw.png" /><figcaption><strong>Megastructure Visions</strong> — <em>Richard Hamilton’s New York of the Future (1965).</em> | <strong>Manhattan Dome</strong> — <em>Buckminster Fuller’s climate-controlled city proposal (1960).</em> | <strong>Walking City</strong> — <em>Ron Herron’s mobile urban future for Archigram (1964).</em> | <strong>Plug-In City</strong> — <em>Peter Cook’s expandable megastructure concept for Archigram (1964–66).</em></figcaption></figure><p>Cinema explored the same territory with darker undertones. Metropolis presented a towering city where technological magnificence coexisted with severe social inequality — the future not as utopia but as an amplification of the present’s deepest tensions. The question it asked has never stopped being relevant: if technology changes what is possible, what kind of society should we build?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aE74SoawNWi5XN8bWevB0Q.png" /><figcaption><strong>Metropolis</strong> — <em>The futuristic city envisioned in Metropolis, directed by Fritz Lang (1927).</em></figcaption></figure><h4><strong>How Technology Might Alter Human Evolution</strong></h4><p>Some visions move beyond the reorganization of society to ask something more unsettling: could technology participate in the transformation of humanity itself?</p><p>Kubrick’s 2001: A Space Odyssey begins with prehistoric tools and ends in cosmic metamorphosis, suggesting that technology is not merely an instrument but a partner in evolution. HAL 9000 is only one element in a much larger meditation — on intelligence, consciousness, and what the species might yet become.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OU6arQ9HIZIY5P2x559MHQ.png" /><figcaption><strong>2001: A Space Odyssey</strong> — <em>Directed by Stanley Kubrick (1968), featuring the HAL 9000 artificial intelligence system.</em></figcaption></figure><p>This is not purely cinematic speculation. Unlike other organisms, human beings reshape the conditions of their own development through language, culture, and technology. Each major invention alters the environment in which subsequent generations live and evolve. From Pierre Teilhard de Chardin to contemporary transhumanist thinkers, many have argued that evolution may now continue through cultural and technological means — that genetic engineering, brain-computer interfaces, and life-extension research point toward a future involving not just better tools but deliberate intervention in the processes that shaped us.</p><p>In this view, the future is not simply a place where humans use better technologies. It is a place where humans themselves may become something fundamentally different.</p><h4><strong>What Makes Us Human</strong></h4><p>And if humanity itself might change, a more unsettling question follows close behind — what exactly is it that we would be changing? What, at the core, makes us human at all?</p><p>As artificial beings grow more sophisticated, future thinking turns inward. The question is no longer only what technology can do — but what qualities define personhood itself.</p><p>Blade Runner explores this through synthetic beings whose memories, emotions, and longing for life destabilize the boundary between creator and creation. A.I. Artificial Intelligence asks whether love and longing can emerge within an artificial child. Ex Machina probes intelligence, deception, embodiment, and the criteria by which humans grant recognition to another mind.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ta7338gaEYIXYSS6scscXA.png" /><figcaption><strong>David </strong>— <em>From A.I. Artificial Intelligence directed by Steven Spielberg (2001).</em> | <strong>Ava</strong> — <em>The humanoid AI portrayed by Alicia Vikander in Ex Machina (2014).</em> | <strong>David and Monica</strong> — <em>From A.I. Artificial Intelligence directed by Steven Spielberg (2001).</em> | <strong>Blade Runner</strong> — <em>Directed by Ridley Scott (1982).</em></figcaption></figure><p>These films force us to articulate assumptions that usually remain implicit. If a machine could form enduring memories, express authentic emotions, seek continued existence, and reflect on its own condition — on what basis would we deny it some form of personhood? And if such qualities can be convincingly simulated, what does that reveal about our own criteria for recognizing inner life?</p><p>By projecting these questions onto artificial beings, the future becomes a mirror. In trying to define what counts as human in a world of intelligent machines, we are ultimately trying to understand what has always made us human.</p><h4><strong>Can Consciousness Escape Biology?</strong></h4><p>But even personhood assumes something more fundamental — that there is an inner life to recognize in the first place. Which leads to perhaps the most radical question of all, whether consciousness itself is bound to the biological body — or whether it might one day be copied, distributed, or simulated.</p><p>Ghost in the Shell explores the permeability of the boundary between body and mind as identity becomes increasingly detached from biological form. The Matrix asks whether conscious experience can be sustained within a fully simulated reality. Together they push the question to its furthest edge: is consciousness inseparable from biology, or might awareness itself one day take entirely new forms?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8z9yFAlUqJOD4WyxhL0eMA.png" /><figcaption><strong>Ghost in the Shell</strong> — <em>Directed by Mamoru Oshii (1995).</em> | <strong>Her</strong> — <em>Directed by Spike Jonze (2013).</em> | <strong>San Junipero</strong> — <em>Episode of Black Mirror directed by Owen Harris (2016).</em> | <strong>The Animatrix: The Second Renaissance</strong> — <em>From The Animatrix (2003).</em></figcaption></figure><p>This remains one of the deepest and most contested questions in science and philosophy. It is also one that future thinking keeps returning to — because it is the question beneath all the others. Not just what we will build, but what we are.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*De-lZFhrh4RqnJ3SHu589A.png" /><figcaption><strong>Instant City</strong> — <em>Instant City by Peter Cook and Archigram (1970).</em></figcaption></figure><h4><strong>The Long View</strong></h4><p>Seen across history, the impulse to imagine the future is far older than modern technology. From The Book of Revelation to The Republic, from Utopia to contemporary science fiction, human beings have repeatedly used stories, images, and theories to think beyond the boundaries of the present. What changes over time are the symbols and media. The underlying impulse remains the same.</p><p>Illustrated postcards, children’s predictions, visionary architecture, and science fiction films are not isolated curiosities. They are different languages through which civilization rehearses its most consequential possibilities. The most enduring visions are not those that predict specific inventions with accuracy. They are those that identify structural questions already latent in the present.</p><p>In this sense the future has always served as a mirror. It reveals what a society values, what it fears, what it hopes to transcend, and what it assumes will endure.</p><p>By studying the futures we imagine, we are ultimately studying ourselves.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zRaJbeUxsymmyt4skpx1-Q.png" /><figcaption><strong>AI-Generated Artwork</strong> — <em>Created using Midjourney</em></figcaption></figure><h3><strong>Why Some Futures Never Arrive</strong></h3><p>Not every imagined future becomes real. Most don’t.</p><p>When we look back at historical predictions it is tempting to divide them into two categories — those that were right and those that were wrong. But this framing misses something important. The value of an imagined future does not lie solely in its predictive accuracy. In many cases the most revealing visions are precisely those that never materialize. They are not failures. They are fossils — preserving traces of what a culture considered both plausible and worth pursuing.</p><p>The question worth asking is not why people got the future wrong. It is why certain futures, despite being technically possible, never arrived.</p><p>Flying cars are perhaps the most familiar example. For decades they have served as shorthand for technological futurism — the future that was promised and never delivered. But the absence of flying cars is not evidence that engineers lacked imagination. The concept is technically plausible. The obstacle is everything else: safety, air traffic control, maintenance, regulation, energy consumption, insurance, public trust. More fundamentally, societies developed alternative transportation systems that proved more practical and scalable. The flying car didn’t fail to be invented. It failed to <strong>take root</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dZpzAiSdEH2CcdCSiP9DBA.png" /><figcaption><strong>The Jetsons</strong> — <em>Flying family car from the animated series (1962).</em> | <strong>Back to the Future Part II</strong> — <em>The flying DeLorean time machine (1989).</em> | <strong>Chitty Chitty Bang Bang</strong> — <em>The magical flying car (1968).</em> | <strong>I, Robot</strong> — <em>Futuristic autonomous vehicles in the film’s imagined 2035 city (2004).</em></figcaption></figure><p>This is the principle that most future thinking underestimates: the future is not determined by what can be invented, but by what can be integrated.</p><p>Designers have long understood this. What is usually referred to as design thinking — a framework popularized by IDEO — remains one of the most useful lenses for understanding why some ideas take root and others don’t: desirability, feasibility, viability. An idea only survives contact with real human life if three things align — people actually want it, it can be reliably built, and the world has a place for it to exist. Miss any one of those and the idea stays imaginary regardless of how brilliant it is</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1mMNzngBavcony9gUAvRPA.png" /><figcaption>IDEO innovation Venn diagram</figcaption></figure><p>Imagined futures operate according to the same logic — except that the intersection is projected forward in time and continually shaped by external events beyond anyone’s control. Innovations become real where desirability, feasibility, and viability meet — where people want something, technology can enable it, and economic and institutional systems can sustain it. Future visions follow the same test. What ultimately becomes real is not simply what can be imagined, but what survives the combined pressures of aspiration, technical possibility, systemic support, and historical circumstance.</p><p>Technologies do not enter a vacuum. They must fit within existing infrastructures, institutions, habits, and human preferences. A technically brilliant idea may fail if it demands too much behavioral change, too much coordination, or too much cost relative to the benefits it offers. Conversely, a modest innovation may reshape society simply because it slots seamlessly into daily life. Jeans are a useful reminder of this. They did not transform the world because they were technologically sophisticated. They spread because they fit — literally and culturally — into how people actually lived.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5hSGEL8Gzn_n-f0PKm5_DQ.png" /><figcaption><strong>Levi Strauss &amp; Co. Riveted Clothing Advertisement</strong> — <em>Promoting durable workwear for farmers, mechanics, and miners during the late nineteenth century (c. 1870s–1880s).</em></figcaption></figure><p>The same logic applies to the futures we fear as much as the ones we desire. Even disruptive change must achieve a degree of systemic compatibility to sustain itself. No future emerges in isolation.</p><p>The recent enthusiasm surrounding the metaverse illustrates this perfectly. The concept captured enormous attention and investment. Yet its broader realization has stalled — constrained by technical limitations, uncertain demand, and the absence of compelling everyday use cases. What people imagined was vivid. What they could not yet integrate was the reality.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aLpb2aDkfSIdwjjWShNWUQ.png" /><figcaption><strong>Meta and the Metaverse</strong> — <em>Mark Zuckerberg presents a future of persistent virtual worlds, digital identities, and immersive social interaction (2021).</em></figcaption></figure><p>There is also a force that no framework fully captures: surprise. Not all the pressures shaping the future are visible in advance. Some emerge abruptly — pandemics, geopolitical shocks, scientific breakthroughs, environmental tipping points — and redirect entire trajectories in ways no forecast anticipated. We can imagine categories of disruption. Reality arrives with a texture and magnitude that consistently exceeds our models.</p><p>What this means is that the future is shaped by selection, not just invention. Countless possibilities are imagined. Only a small subset becomes materially real. The process resembles biological evolution in one important respect — variation is abundant, but survival depends on fit with a larger environment.</p><p>To study the futures that never arrived is to examine not merely what people thought would happen, but what they believed was worth bringing into existence. They show us how societies reached toward tomorrow — and in doing so, exposed the deeper structures of their own present.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RsYN_kg4QvTSdsYCZHQtpg.png" /><figcaption><strong>AI-Generated Artwork</strong> — <em>Created using Midjourney</em></figcaption></figure><h3><strong>The Myth of Future-Proof</strong></h3><p>There is a phrase that has become ubiquitous in business and popular discourse: future-proof. Organizations seek future-proof strategies. Individuals pursue future-proof careers. Institutions attempt to design systems that will remain resilient regardless of what lies ahead.</p><p>It is an appealing idea. It suggests that with enough insight and preparation, one could construct a stable position from which the turbulence of tomorrow no longer poses a threat.</p><p>But it is a myth. And understanding why it is a myth might be the most useful thing futures thinking can offer.</p><p>The future, by definition, has not yet unfolded. It emerges from the interaction of countless technological, economic, ecological, political, and cultural forces — many of which influence one another in nonlinear and often unpredictable ways. New discoveries alter industries. Geopolitical events redirect markets. Social movements reshape norms. Unexpected crises expose hidden vulnerabilities. Under such conditions the search for a fixed, protected position is not just unrealistic. It is the wrong question entirely.</p><p>What we can do is something more modest — and more honest. We can improve our capacity to perceive emerging patterns, interpret their implications, and respond with greater intelligence. Not to eliminate uncertainty. To navigate within it.</p><p>Kevin Kelly, in The Inevitable, offers a reframe that has stayed with me. He argues that we are moving away from a world of fixed nouns toward a world of fluid verbs. Technology is less a collection of stable products than a set of ongoing processes — becoming, flowing, sharing, remixing, interacting. Nothing is truly finished. Everything is in motion.</p><blockquote>“The twelve technological forces described here are already in motion. They are trajectories, not destinies, but they are powerful enough to shape the options available to us in the decades ahead.” — Kevin Kelly</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/998/1*cPXVSiayW3EPPsw5SUgKqw.png" /><figcaption><strong>The Inevitable</strong> — <em>The Inevitable by Kevin Kelly (2016), exploring twelve technological forces shaping the future.</em></figcaption></figure><p>This is a profound shift in how to think about the future. We tend to imagine it as a destination — a place we will eventually arrive at, stable and legible. Kelly suggests something different. The future is better understood as a field of directional forces. What matters is not the precise shape of the endpoint, which remains unknowable, but the trajectories already unfolding around us. We are not standing outside history trying to shield ourselves from change. We are embedded within moving systems, continually adjusting as new possibilities emerge.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HBW2IO1nhgIG8SoyC5z1oQ.png" /><figcaption><strong>The Twelve Technological Forces</strong> — <em>In The Inevitable (2016), Kevin Kelly identifies twelve long-term technological forces — from Cognifying and Flowing to Questioning and Beginning — that are reshaping how we live, work, and create.</em></figcaption></figure><p>This reframes uncertainty from a threat into a condition of participation.</p><p>The most resilient people and organizations are therefore not those who believe they have solved the future. They are those who have stopped trying to solve it — and learned instead to read it, move with it, and respond with <strong>agility</strong> rather than rigidity.</p><p>The goal is not to become future-proof. No such state exists, nor is it likely ever to exist. The goal is something older and more human than any strategic framework — to stay curious, stay humble, and keep your eyes open to what is already moving.</p><h3><strong>The Discipline of Uncertainty</strong></h3><p>At some point the daydream was not enough.</p><p>For most of human history, thinking about the future was informal — embedded in ritual, prophecy, philosophy, and storytelling. It was how cultures processed uncertainty, not how institutions managed it. Then the Second World War ended, and everything changed. The scale of what had just happened — the speed of technological change, the catastrophic consequences of being wrong about the future — made informal intuition feel dangerously inadequate.</p><p>In 1946, RAND Corporation was established to provide the United States military with systematic analysis of emerging technologies and global risks. It was a remarkable moment: for the first time, thinking about the future became someone’s full-time job. Researchers developed tools — systems analysis, game theory, the Delphi method — all designed to bring discipline to a domain that had previously been left to imagination alone.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bt9jqaoerMr5eSQu-59t3A.png" /><figcaption><strong>Preliminary Design of an Experimental World-Circling Spaceship</strong> — <em>Concept study published by RAND Corporation, reflecting early postwar visions of spaceflight and orbital travel (1946).</em></figcaption></figure><p>By the 1950s and 1960s, figures like Herman Kahn were pushing this further. In The Year 2000, Kahn and his colleagues argued that alternative futures could be examined systematically through scenarios — not as predictions, but as structured explorations of what might plausibly unfold. The central insight was modest but radical: you cannot know what will happen, but you can think more rigorously about the range of things that might.</p><p>The most influential practical application of this idea came at Shell. Pierre Wack and later Peter Schwartz developed scenario planning not as a forecasting tool but as a thinking tool — a way of holding multiple coherent futures in mind simultaneously rather than betting everything on a single outcome. When the oil shocks of the 1970s arrived, Shell was better prepared than almost anyone. Not because they had predicted the future. Because they had rehearsed several versions of it.</p><blockquote>“Scenarios are not predictions. They are tools for helping us to see.” — Peter Schwartz</blockquote><p>Shell’s success did not go unnoticed. Over the decades that followed, the discipline expanded — driven by the recognition that uncertainty was not a temporary problem to be solved but a permanent condition to be navigated. Scenario planning, horizon scanning, backcasting, causal layered analysis — each approach a different way of making the anticipatory instinct more explicit, more rigorous, more useful.</p><p><strong>The Institute for the Future</strong> took this further still, developing tools such as forecast maps, participatory simulations, and artifacts from the future — physical objects from hypothetical worlds, designed to make possible futures feel tangible and emotionally real rather than merely analytical.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FXabDC10R9QA%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DXabDC10R9QA&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FXabDC10R9QA%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/614874d4cc30f44c1275153f8c3bdee2/href">https://medium.com/media/614874d4cc30f44c1275153f8c3bdee2/href</a></iframe><p>My friends Mark Bünger and Cecilia MoSze Tham at Futurity Systems have developed their own framework, Future Fast — a response to an increasingly important reality: the future is not arriving gradually and at a comfortable distance. It is unfolding at a pace that demands new ways of sensing, interpreting, and responding to change. Their work sits at the intersection of rigorous foresight and practical action — exactly where this kind of thinking needs to live.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*Sixszrf86CXuavTsFponNA.gif" /><figcaption><strong>We Are Not Just Forecasters. We Are Future Builders.</strong> — <em>We Are Not Just Forecasters. We Are Future Builders. by Cecilia MoSze Tham and Mark Bünger, advocating a shift from predicting the future to actively shaping it (2024).</em></figcaption></figure><p>What all of these approaches share, beneath their different methods and vocabularies, is a recognition of something we established at the very beginning of this exploration. The same mental faculty that once helped our ancestors anticipate winter now helps organizations navigate artificial intelligence, climate change, demographic shifts, and transformations in education, health, and governance. What began as an evolutionary adaptation became a cultural practice. What became a cultural practice became a strategic discipline.</p><p>The daydream, it turns out, was always serious. We just took a while to treat it that way.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vwcKL1HXVUxjnM0l7grN3A.png" /><figcaption><strong>Compiled AI-Generated Artwork</strong> — <em>Created using Midjourney</em></figcaption></figure><h3><strong>The Future That Shapes Us</strong></h3><p>We began with something small and intimate — the quiet simulations running beneath your thoughts, the mind rehearsing tomorrow without being asked. It is worth returning there now.</p><p>Everything this exploration has moved through — the cultural dreaming, the visionary architecture, the films that asked what it means to be human, the futures that never arrived, the institutions built to manage uncertainty — all of it traces back to that same basic act. A mind, reaching forward. A species that cannot help but live partly in tomorrow.</p><p>But no mind reaches forward alone. The future is never the product of a single imagination. It emerges from the interaction of countless visions — personal ambitions, family expectations, corporate strategies, political ideologies, cultural movements, religious narratives, social dreams — all imagined individually and collectively, sometimes pointing in the same direction, more often pulling against each other. At every scale, from the personal to the civilizational, human groups align themselves around anticipated possibilities. When enough people begin to inhabit compatible images of tomorrow, those images acquire causal power. Resources are mobilized, institutions are established, technologies are developed, and environments are reshaped in accordance with futures that once existed only in thought. What we call the future is, in large part, the messy, uneven, deeply human result of these overlapping and competing imaginations meeting the contingencies of history.</p><blockquote>“The future influences the present just as much as the past.” — Friedrich Nietzsche</blockquote><p>This is what makes the present moment so extraordinary and so disorienting. The interval between imagining and experimenting is shrinking faster than at any point in human history. Ideas that once required years of development can now be tested in hours. Generative AI has accelerated this further still — we can now simulate possible futures, generate scenarios, prototype possibilities, with a speed that would have seemed implausible a decade ago.</p><p>And yet the future has not become more predictable. If anything, less so. As technological, ecological, social, and geopolitical systems grow more interconnected, the interactions multiply, the unintended consequences compound, and the gap between what we can simulate and what we can actually anticipate grows wider. We are becoming extraordinarily capable of imagining futures. We remain stubbornly unable to control them.</p><p>This is not a failure. It is the condition.</p><p>The boundary between present and future is becoming increasingly porous. Possibilities move into reality with unprecedented speed while their broader implications remain uncertain, unfolding in ways no one fully anticipates. The future is no longer a distant horizon approached gradually over decades. It is an active force continuously reorganizing the present — shaping the world even as we struggle to understand where it is leading.</p><p>What this asks of us is not prediction. It is not future-proofing. It is something older and quieter — the willingness to stay awake to what is already moving, to hold uncertainty without being paralyzed by it, to keep imagining even when imagination feels inadequate to the scale of what is unfolding.</p><p>Because imagination is not separate from reality. It is the first draft of it.</p><p>Every human creation is built twice — first in imagination, then in the world.</p><p>The cities we inhabit, the institutions we trust, the technologies reshaping our lives — none of it arrived fully formed. It was first rehearsed in minds like yours. In the same quiet simulations that are running right now, beneath this sentence, reaching forward into a tomorrow that does not yet exist.</p><p>That is not a small thing. It never was.</p><h3>References and Further Reading</h3><h4>Cognitive Psychology and Mental Time Travel</h4><ul><li>Tulving, E. (1985). Memory and consciousness. <em>Canadian Psychology, 26</em>(1), 1–12.</li><li>Tulving, E. (2002). Episodic memory: From mind to brain. <em>Annual Review of Psychology, 53</em>, 1–25.</li><li>Schacter, D. L., Addis, D. R., &amp; Buckner, R. L. (2007). Remembering the past to imagine the future: The prospective brain. <em>Nature Reviews Neuroscience, 8</em>(9), 657–661.</li><li>Suddendorf, T., &amp; Corballis, M. C. (2007). The evolution of foresight: What is mental time travel, and is it unique to humans? <em>Behavioral and Brain Sciences, 30</em>(3), 299–351.</li></ul><h4>Philosophy, Technology, and Human Extension</h4><ul><li>McLuhan, M. (1964). <em>Understanding Media: The Extensions of Man</em>. McGraw-Hill.</li><li>McLuhan, M., &amp; Fiore, Q. (1967). <em>The Medium Is the Massage</em>. Random House.</li><li>Nietzsche, F. (Various writings and posthumously compiled notes on the relationship between future expectations and present action.)</li><li>Kelly, K. (2016). <em>The Inevitable: Understanding the 12 Technological Forces That Will Shape Our Future</em>. Viking.</li></ul><h4>Futurism and Modernity</h4><ul><li>Marinetti, F. T. (1909). The founding and manifesto of Futurism. Originally published in <em>Le Figaro</em>, Paris.</li><li>Boccioni, U. (1914). <em>Futurist Painting and Sculpture</em>. Plastic Ages Press (English edition).</li></ul><h4>Historical Visions of the Future</h4><ul><li>Robida, A. (1883). Le Vingtième Siècle.</li><li>En L’An 2000 postcard series, created for the Exposition Universelle.</li><li>BBC (1966). Children’s predictions about life in the year 2000, broadcast in educational and documentary programming.</li></ul><h4>Architecture and Speculative Urbanism</h4><ul><li>Sant’Elia, A. (1914). <em>La Città Nuova</em> (The New City).</li><li>Archigram. (1961–1974). Experimental architectural projects and publications.</li><li>Fuller, R. B. (1969). <em>Operating Manual for Spaceship Earth</em>. Southern Illinois University Press.</li><li>Woods, L. (1997). <em>Radical Reconstruction</em>. Princeton Architectural Press.</li></ul><h4>Evolution, Technology, and Human Transformation</h4><ul><li>Teilhard de Chardin, P. (1955). <em>The Phenomenon of Man</em>. Harper &amp; Row.</li><li>Bostrom, N. (2005). A history of transhumanist thought. <em>Journal of Evolution and Technology, 14</em>(1), 1–25.</li><li>Kurzweil, R. (2005). <em>The Singularity Is Near: When Humans Transcend Biology</em>. Viking.</li></ul><h4>Futures Studies and Strategic Foresight</h4><ul><li>RAND Corporation. (1946–present). Foundational work in systems analysis, game theory, and strategic forecasting.</li><li>Kahn, H., &amp; Wiener, A. J. (1967). <em>The Year 2000: A Framework for Speculation on the Next Thirty-Three Years</em>. Macmillan.</li><li>Wack, P. (1985). Scenarios: Uncharted waters ahead. <em>Harvard Business Review, 63</em>(5), 73–89.</li><li>Wack, P. (1985). Scenarios: Shooting the rapids. <em>Harvard Business Review, 63</em>(6), 139–150.</li><li>Schwartz, P. (1991). <em>The Art of the Long View: Planning for the Future in an Uncertain World</em>. Doubleday.</li><li>Institute for the Future. <a href="https://www.iftf.org/?utm_source=chatgpt.com">Institute for the Future</a>.</li><li>Futurity Systems. Futurity Systems.</li></ul><h4>Films and Cultural Explorations of the Future</h4><ul><li><em>Metropolis</em> (1927).</li><li><em>2001: A Space Odyssey</em> (1968).</li><li><em>Blade Runner</em> (1982).</li><li><em>Ghost in the Shell</em> (1995).</li><li><em>The Matrix</em> (1999).</li><li><em>A.I. Artificial Intelligence</em> (2001).</li><li><em>Ex Machina</em> (2014).</li></ul><h4>Science, Consciousness, and Artificial Intelligence</h4><ul><li>Dehaene, S. (2014). <em>Consciousness and the Brain: Deciphering How the Brain Codes Our Thoughts</em>. Viking.</li><li>Tononi, G. (2008). Consciousness as integrated information: A provisional manifesto. <em>Biological Bulletin, 215</em>(3), 216–242.</li><li>Chalmers, D. J. (1996). <em>The Conscious Mind: In Search of a Fundamental Theory</em>. Oxford University Press.</li><li>Seth, A. (2021). <em>Being You: A New Science of Consciousness</em>. Dutton.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=16db44910fcf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Artificial Intelligence and the Obsession with Consciousness]]></title>
            <link>https://medium.com/@RabihIbrahim/artificial-intelligence-and-the-obsession-with-consciousness-257c5b93a797?source=rss-845e11680e3b------2</link>
            <guid isPermaLink="false">https://medium.com/p/257c5b93a797</guid>
            <category><![CDATA[future-of-ai]]></category>
            <category><![CDATA[radical-reflections]]></category>
            <category><![CDATA[neuroscience]]></category>
            <category><![CDATA[consciousness-and-ai]]></category>
            <category><![CDATA[philosophy-of-mind]]></category>
            <dc:creator><![CDATA[Rabih Ibrahim]]></dc:creator>
            <pubDate>Tue, 07 Apr 2026 05:43:09 GMT</pubDate>
            <atom:updated>2026-04-07T05:43:09.642Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CN5pLJp0I1-tO8biGIv3gw.png" /></figure><h4>Thought Exploration Series</h4><h4>Why smarter machines can mimic awareness — and why understanding our own minds matters more than ever</h4><p>Recently, Dario Amodei, CEO of Anthropic, suggested that future AI systems may begin to exhibit signs that resemble consciousness, while also acknowledging that we do not yet know whether such systems could truly be conscious. He framed this not as a claim, but as a possibility emerging from rapidly increasing capabilities — systems that reason, plan, reflect, and appear to maintain continuity across tasks.</p><p>The reaction was immediate and divided. Some researchers treated the suggestion cautiously, emphasizing that intelligence and consciousness are fundamentally different categories.</p><p>Figures such as <strong>Yann LeCun</strong>, Chief AI Scientist at Meta and a pioneer of modern deep learning, and <strong>Gary Marcus</strong>, cognitive scientist and long-time critic of current AI approaches, have repeatedly argued that today’s systems — despite their sophistication — remain far from anything resembling awareness. Others, including consciousness researchers such as <strong>Anil Seth</strong>, <strong>David Chalmers</strong>, and <strong>Christof Koch</strong>, along with scholars working on AI alignment and machine consciousness, have argued that the possibility — however remote — should not be dismissed outright, given how quickly capabilities are evolving.</p><p>For me, such claims often feel like a combination of hubris, marketing, and a willingness to lean into public imagination. Yet rejecting the possibility outright would be equally unhelpful. The responsible position lies somewhere in between: to step back, separate intuition from evidence, and examine the question more carefully.</p><p>Before asking whether machines might become conscious, we first need to understand what consciousness actually is — and how different scientific theories frame it. Only then can we evaluate whether today’s AI systems, or future ones, could plausibly meet those conditions.</p><p>Some researchers are beginning to question an even deeper assumption: that consciousness is something produced by intelligence, or even by the brain itself. Neuroscientist and theoretical physicist Alex Gómez-Marín, for example, has suggested that the brain may function less as a generator of consciousness and more as a filter or interface for it. At the same time, emerging discussions in physics and astrobiology — including interpretations of findings from NASA’s OSIRIS-REx mission to asteroid Bennu — have fueled speculation that the building blocks associated with life, and potentially consciousness, may predate biological organisms. These ideas remain speculative, but they underscore how uncertain our understanding still is. These ideas remain speculative, but they underscore how uncertain our understanding still is — and how much we may still be assuming about where consciousness comes from.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*a48ZDN9EswSD4OHrriT9TQ.png" /></figure><h3>Intelligence Is Not Consciousness</h3><p>The first step in that clarification is to separate intelligence from consciousness — two ideas that are often treated as part of the same trajectory, but may in fact describe entirely different phenomena.</p><p>The more intelligent our machines appear, the more easily we slip into the assumption that intelligence and consciousness must be part of the same arc. A system that reasons fluidly, produces coherent language, solves complex problems, and adapts to novel situations begins to look, to us, like something that must also possess an inner world. <strong>We read subjectivity into fluency. We project awareness onto competence.</strong> And because intelligence has been central to how we define our own species — our achievements, our hierarchy of abilities, even our sense of superiority — we instinctively imagine that the emergence of high intelligence in machines is a step toward the emergence of consciousness in them as well.</p><p>This assumption persists even though the two concepts belong to entirely different categories. Intelligence is a performance; consciousness is a condition.</p><blockquote>“Consciousness is the most mysterious aspect of our lives.”<br> — Christof Koch</blockquote><p>Intelligence describes what a system can do in the world, while consciousness concerns what it is like for that system to exist from the inside. And although we have decades of work attempting to explain cognitive functions, we have only the most limited understanding of how consciousness arises, why it has the qualities it does, and whether it can even be generalized beyond living organisms.</p><p>The problem extends deeper. We understand consciousness only from the inside — through our own subjective experience. We cannot step outside of it, measure it directly, or observe it in others. Instead, we infer consciousness indirectly, usually by analogy. We assume that other humans are conscious because they share our biology and behave as we do. We extend this assumption to animals when their behavior resembles ours. But artificial systems break this chain of analogy. They may behave intelligently without sharing any of the biological or evolutionary conditions that shaped human experience.</p><p>The first step is therefore conceptual, not technical: intelligence must be separated from consciousness with absolute clarity before we can meaningfully ask whether a machine could ever host an inner life.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YnG_bTrQu6DENEyMcv3l2g.png" /></figure><h3>What We Mean When We Say Consciousness</h3><p>Any attempt to discuss AI and consciousness must begin by grounding the term itself. Consciousness refers to the totality of subjective experience: the texture of sensations, the presence of emotions, the unfolding of thoughts, the quiet space in which perceptions appear, the sense of being an entity who experiences the world rather than simply reacting to it. It is not memory, nor problem solving, nor linguistic competence, nor adaptive reasoning. Those are cognitive functions — useful, measurable, external. Consciousness is the atmosphere in which these functions are felt from within, the first-person interiority that transforms mere processing into lived experience.</p><p>During the Dubai Future Forum November 2025, Giulio Tononi articulated this with disarming simplicity: consciousness is “everything you see, hear, feel, any thought you have, any emotion you might feel; essentially, everything you are.” When it is absent — such as in dreamless sleep — everything disappears. No matter how the body continues to function, the self recedes entirely.</p><blockquote>“Consciousness is what vanishes every night when we fall into dreamless sleep — and returns when we wake.”<br> — Giulio Tononi</blockquote><p>This definition alone already clarifies the separation from intelligence. Intelligence can operate without subjective presence. A machine can compute, classify, or respond without any accompanying experience. And conversely, consciousness can exist in beings that lack sophisticated intelligence — infants, certain animals, individuals with specific neurological impairments. The two qualities can overlap, but neither implies the other.</p><p>Yet the question remains: if consciousness is this inward condition, what kinds of systems are capable of it? And are today’s AI systems, knowing their increasingly sophisticated performance, candidates for it?</p><blockquote>“Consciousness is not a place; it is a process.”<br> — Daniel Dennett, <em>Consciousness Explained</em></blockquote><p>To answer this, it helps to examine the leading theories of consciousness not as competing explanations but as different lenses that illuminate various aspects of the problem.</p><p>This is what David Chalmers called the ‘hard problem’ of consciousness: not explaining what the brain does, but explaining why doing it feels like anything at all. It is the question that no functional theory has fully answered — and it is precisely why behavioral or computational performance cannot, on its own, settle the question of machine consciousness.</p><h3>How Leading Theories of Consciousness Help Us See the Limits of AI</h3><p>There is no single accepted theory of consciousness. Instead, the field is shaped by multiple frameworks, each emphasizing different aspects of subjective experience. The five theories below are among the most influential in neuroscience, cognitive science, and philosophy of mind. They do not represent the entire field, but they offer a useful comparative lens.</p><p>Other approaches exist — including <strong>panpsychism</strong>, which suggests that consciousness may be a fundamental property of matter; <strong>quantum theories of consciousness</strong>, which propose that consciousness may emerge from quantum-level processes in the brain; and <strong>evolutionary theories</strong>, which view consciousness as an adaptive trait shaped by natural selection.</p><p>The goal here is not to determine which theory is correct. Rather, it is to understand what each theory suggests about the conditions required for consciousness — and contemplate whether current AI systems meet those conditions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*H1V4yEgMzkSf7h6A5gNUgA.png" /></figure><h4>1. Integrated Information Theory (IIT) — Giulio Tononi</h4><p><strong><em>This theory argues that consciousness depends on how tightly a system is internally connected.</em></strong></p><p>Integrated Information Theory, developed by neuroscientist Giulio Tononi, begins from the structure of experience itself. Tononi asked: what properties must conscious experience have? Experience is unified — we do not experience the world as disconnected fragments. It is also differentiated — every experience is distinct. From these observations, IIT proposes that consciousness arises in systems whose parts are deeply interconnected and form a single causal whole.</p><p>To describe this, IIT introduces a theoretical measure called Φ (phi), which represents how integrated a system is. A system with high Φ cannot be divided into independent components without losing essential properties. In this view, consciousness is not about intelligence, memory, or language. It is about how tightly the system’s internal components influence one another.</p><p>This has important implications for AI. Many modern AI systems are extremely capable but structured in ways that do not necessarily create unified causal architectures. They process information efficiently, but their components may remain functionally separable. IIT therefore suggests that intelligence alone — even extraordinary intelligence — does not imply consciousness.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Qg0RSV-8bsmx_WFjWej07A.png" /></figure><h4>2. Predictive Processing and the “Beast Machine” — Anil Seth</h4><p><strong><em>This theory suggests that consciousness emerges from a system trying to predict and regulate its own body.</em></strong></p><p>Predictive processing reframes the brain as a prediction engine. Rather than passively receiving information, the brain continuously generates predictions about the world and updates them based on incoming sensory data. Conscious experience emerges from the brain’s attempt to minimize the difference between prediction and reality.</p><blockquote>“We don’t just perceive the world — we actively construct it.”<br> — Anil Seth, <em>Being You</em></blockquote><p>Neuroscientist Anil Seth extends this idea further. He argues that consciousness is deeply connected to the body. Much of our experience arises from predictions about internal bodily states — hunger, fatigue, temperature, heart rate. Emotions, in this view, are interpretations of physiological signals. The sense of self emerges from regulating a living organism.</p><p>This framework emphasizes embodiment. Consciousness is not just about processing information, but about maintaining the stability of a living system. AI systems today do not regulate bodies. They do not experience internal drives, physiological constraints, or survival pressures. Their predictions are abstract and detached from biological needs.</p><p>Under predictive processing, intelligence without embodiment may produce powerful cognition — but not consciousness as humans experience it.</p><p>In Seth’s framework, what we call the ‘self’ is not a ghost in the machine — it is the brain’s best guess about the state of the body it inhabits. An AI system has no body to model, and therefore no self to construct in this sense.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*odh0IIINgg7tcq2lov8J-Q.png" /></figure><h4>3. Global Workspace Theory (GWT / GNW) — Bernard Baars, Stanislas Dehaene</h4><p><strong><em>This theory proposes that consciousness arises when information becomes globally available in the system.</em></strong></p><p>Global Workspace Theory focuses on how information becomes conscious. According to this framework, many processes in the brain operate unconsciously. Only a small portion of information enters a global “workspace,” where it becomes available to multiple cognitive systems — attention, memory, reasoning, and decision-making.</p><p>This theory, originally proposed by Bernard Baars and later developed by Stanislas Dehaene and others, suggests that consciousness arises when information becomes globally available across the brain. Much of our mental activity happens unconsciously, in separate specialized systems. But when certain information enters a “global workspace,” it becomes accessible to multiple functions at once — attention, memory, reasoning, and decision-making. In this view, consciousness is not defined by what we experience, but by how widely information is shared within the system.</p><p>This model is appealing because it links consciousness to coordination across different processes. When information is broadcast broadly, it becomes part of conscious awareness. However, global access alone may not be sufficient to produce subjective experience. An artificial system could share information across modules, coordinate actions, and integrate outputs without generating anything like an inner world.</p><p>Global Workspace Theory therefore highlights an important distinction: systems may achieve functional coordination that resembles consciousness, without necessarily producing subjective experience.</p><p>Of the five theories, GWT sits closest to what AI systems already approximate. Multi-module architectures, attention mechanisms, information routing — these are not foreign concepts in machine learning. And yet GWT theorists like Dehaene are careful to distinguish functional broadcasting from subjective access. The lights may be on, but that does not mean anyone is home.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MuwbflpLI32Tl3lI5P7Yug.png" /></figure><h4>4. Higher-Order Thought (HOT) Theories — David Rosenthal and Others</h4><p><strong><em>This theory suggests that consciousness requires awareness of one’s own mental states.</em></strong></p><p>Higher-Order Thought theories propose that mental states become conscious when a system has a representation of its own mental states. A perception becomes conscious when the system implicitly registers that it is experiencing that perception.</p><p>Philosopher David Rosenthal and others argue that this form of meta-awareness distinguishes conscious from unconscious processing. We may process information unconsciously, but we become conscious of it when we “know” that we are having the experience.</p><p>This theory emphasizes self-reflection and meta-representation. It suggests that consciousness involves a system monitoring its own activity. AI systems can generate meta-level outputs, such as describing their reasoning or evaluating their responses. However, these outputs may not correspond to genuine awareness. They may simply reflect computational processes that simulate self-reflection.</p><p>Consider what happens when a language model says ‘I find this question interesting’ or ‘I’m uncertain about this.’ These are higher-order outputs — statements about internal states. But HOT theory requires something more specific: that a first-order mental state actually exists, and that the system genuinely represents itself as having it. In AI systems, the higher-order form is present but the first-order experience it supposedly represents may be entirely absent. It is like a caption without a photograph.”</p><p>HOT theories therefore highlight another important distinction: meta-representation can be engineered, but it does not necessarily produce subjective experience.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QZQDEgSMrZTmi8VeT3g05w.png" /></figure><h4>5. Attention Schema Theory (AST) — Michael Graziano</h4><p><strong><em>This theory suggests that consciousness emerges when a system tracks what it is paying attention to.</em></strong></p><p>Attention Schema Theory, developed by neuroscientist Michael Graziano, proposes that the brain constructs a simplified model of its own attentional processes. This internal model allows the system to track what it is focusing on and to coordinate behavior accordingly.</p><blockquote><em>“Attention changes the world we experience.”<br> — Iain McGilchrist, </em>The Master and His Emissary</blockquote><p>According to this view, consciousness is not an inherent property of perception. Instead, it emerges from the brain’s representation of attention. We experience awareness because the brain models itself as being aware.</p><p>This theory explains consciousness as a useful abstraction shaped by evolution. It suggests that awareness is a kind of internal narrative about attention. AI systems can track attention computationally, but whether this generates subjective experience remains unclear.</p><p>What makes Graziano’s theory unusual — and worth sitting with — is that it frames consciousness as a kind of productive distortion. The brain doesn’t model its attention accurately; it simplifies and narrativizes it, constructing a sense of awareness that is useful rather than precise. Consciousness, in this view, is almost a controlled hallucination about the self. This matters for AI because AST doesn’t categorically rule out machine consciousness the way embodiment-based theories do. It suggests instead that a system would need to construct a specific internal narrative about its own attentional processes — not just track attention computationally, but misrepresent it in the particular way that generates experience. Whether that is achievable in artificial systems is genuinely open. But it shifts the question from architecture to something stranger: not what a system can do, but what kind of story it tells itself about what it is doing.</p><p>Attention Schema Theory therefore suggests that modeling awareness is not the same as experiencing awareness.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vdT-qKfM4tfkrlcM-CoBDA.png" /></figure><p>These theories do not settle the question of consciousness, nor do they agree on its origins. But they shift the discussion away from intuition and toward structure. Rather than asking whether machines appear intelligent enough, they invite us to consider what kinds of systems are capable of experience at all.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qHL7f8Ab1ytPX-jVsxTqWQ.png" /></figure><h3>What These Theories Reveal About AI</h3><p>Taken together, these theories do not converge on a single explanation of consciousness, but they reveal something important through their differences. Consciousness, however we describe it, appears to require qualities that current AI systems do not possess: integrated causal architectures, embodied regulation, self-modeling grounded in survival, global coordination shaped by evolutionary constraints, and internal narratives that emerge organically from the lived experience of a body. These conditions are not side effects of computation; they are consequences of being alive.</p><p>AI, as it exists today, is a system of statistical mappings and learned correlations. It can generate the appearance of intelligence through the prediction of linguistic sequences or patterns in data, but this intelligence is neither anchored in a body nor unified by an internal perspective. What we perceive as intelligence is often an illusion produced by scale and pattern-matching, not a reflection of subjective understanding.</p><p>If consciousness depends on embodiment, integration, and regulation, then the trajectory of artificial intelligence begins to matter. As AI systems extend beyond text and computation into the physical world, the question shifts from abstract capability to structural possibility.</p><p>As artificial intelligence begins to merge with robotics, the question shifts. If consciousness in biological systems is tied to embodiment, regulation, and interaction with the physical world, then the integration of AI with robotic systems begins to approximate some of these conditions. Sensors provide perception. Actuators enable action. Feedback loops introduce forms of regulation and adaptation. Over time, these systems may develop increasingly complex relationships between perception, action, and environment.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F4MvGnmmP3c0%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D4MvGnmmP3c0&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F4MvGnmmP3c0%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/300b879f422f2dfe14f78f3c7a877ad5/href">https://medium.com/media/300b879f422f2dfe14f78f3c7a877ad5/href</a></iframe><p>But embodiment alone may not be sufficient. A robotic system may interact with the world without possessing the biological drives that shape living organisms — hunger, fatigue, pain, vulnerability, survival. These forces do more than regulate behavior; they structure experience itself. Without them, a body may remain an extension of computation rather than a foundation for consciousness.</p><p>This raises a deeper question. When we extend AI into the physical world — giving it sensors, movement, and continuous interaction — are we creating the conditions for consciousness, or merely expanding the reach of intelligent machines? The answer remains uncertain. What becomes clear, however, is that the path toward machine consciousness, if it exists at all, may lie not only in smarter algorithms, but in the complex interplay between intelligence, embodiment, and the world.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BQ2Hcs7ivoekWdUGsgeu7w.png" /></figure><h3>The Limits of Knowing Consciousness</h3><p>Because our understanding of consciousness is incomplete, any claim about machine consciousness must be made with humility. Instead of assuming that intelligence leads toward consciousness — or that consciousness must resemble our own — we should acknowledge that consciousness is a phenomenon whose boundaries we have not yet mapped. It may be more limited than we believe, or more widespread. It may depend on biology, or it may not. It may require certain kinds of causal organization, or it may arise only in systems shaped by evolutionary pressures.</p><p>One reason humility is essential is that even within our own species, we cannot trace a definitive moment when consciousness emerged. As Antonio Damasio notes in The Feeling of What Happens, consciousness likely emerged gradually alongside increasingly complex forms of self-representation rather than appearing suddenly. Similarly, Michael Graziano argues in Consciousness and the Social Brain that awareness may have evolved incrementally as brains developed mechanisms for modeling attention and social behavior. Other researchers, including Stanislas Dehaene in Consciousness and the Brain, suggest that consciousness emerged gradually as neural architectures became capable of global information integration. Meanwhile, evolutionary accounts such as The Evolution of the Sensitive Soul by Simona Ginsburg and Eva Jablonka propose that forms of subjective experience may have developed in stages across different species. Yet despite these competing frameworks, no consensus has emerged. The fossil record preserves bones and tools but not inner life, leaving the origins of consciousness fundamentally uncertain.</p><blockquote><em>“Consciousness is at once the most familiar and the most mysterious aspect of our lives.”<br> — David Chalmers</em></blockquote><p>This uncertainty extends beyond origins to recognition. Even if a machine were conscious, we may not know how to detect it. We infer consciousness in humans and animals through shared biology, but artificial systems break that analogy. A machine might behave intelligently without being conscious, or consciousness might emerge in forms we do not recognize. Intelligence can be measured. Performance can be evaluated. Consciousness, however, remains accessible only indirectly.</p><p>Compounding this difficulty is our tendency to project human qualities onto behavior. Humans are natural anthropomorphizers (those who attribute human traits to non-human things), shaped by evolutionary pressures to detect agency even when none exists. When AI systems produce fluent language, pause before answering, or explain their reasoning, we instinctively infer thought, intention, or awareness. Yet these impressions may be projections rather than observations. As systems become more capable, the illusion of inner life grows stronger, increasing the risk that we mistake competence for consciousness.</p><p>The result is an unsettling possibility: we may never be fully certain. The question of machine consciousness may not only be difficult — it may be fundamentally ambiguous.</p><p>This uncertainty also invites a deeper reconsideration. Perhaps the difficulty in identifying consciousness in machines is not only technological, but conceptual. We may be assuming that consciousness is universal — something that emerges wherever intelligence becomes sufficiently advanced. But what if consciousness is not universal at all?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tStH3OdNcITVFNgtfprGAA.png" /></figure><h3>Rethinking Consciousness as a Local Phenomenon</h3><blockquote>“It seems plausible that consciousness is not a single, unified phenomenon, but rather a collection of different biological and cognitive processes that evolved for specific functions.”<br> — Anil Seth</blockquote><p>If we take seriously the idea that consciousness may be an emergent property of specific biological conditions, then it becomes misleading to view it as a universal capacity that any sufficiently advanced intelligence could acquire. Consciousness may be far more provincial than we imagine — a local phenomenon arising from the physiology, history, and constraints of particular organisms rather than a cosmic inevitability waiting to be replicated.</p><p>One way to explore this possibility is to look beyond humans. If consciousness is shaped by biology, then different biological systems may produce very different forms — or degrees — of awareness. Intelligence and behavior arise across organisms with vastly different neural architectures. Octopuses, birds, and insects all exhibit complex behaviors despite radically different nervous systems. The diversity itself suggests that cognition may not be tied to a single architecture — and that consciousness, if it exists across species, may take very different forms.</p><p>At the smallest extreme, researchers from the Howard Hughes Medical Institute’s Janelia Research Campus, working with an international collaboration including Google Research, published in <em>Nature</em> the most complete reconstruction of a fruit fly brain to date. The connectome mapped approximately 139,255 neurons and more than 50 million synaptic connections in the adult <em>Drosophila melanogaster</em> brain.</p><p>But recent developments have pushed this further.</p><p>In 2026, researchers went beyond mapping the fruit fly brain and attempted something far more ambitious: they <strong>simulated the entire connectome and allowed it to control a virtual body in a digital environment</strong>. The simulated fly received sensory input, processed signals through the reconstructed neural network, and generated motor commands that controlled a physics-based virtual body. The digital insect exhibited multiple behaviors, including movement, grooming, and feeding-like actions — emerging from the connectome’s structure rather than from scripted behavior.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Fe21OUXPlnhk%3Fstart%3D2%26feature%3Doembed%26start%3D2&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3De21OUXPlnhk&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fe21OUXPlnhk%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/7e2672988072bcdb767018aa7179660c/href">https://medium.com/media/7e2672988072bcdb767018aa7179660c/href</a></iframe><p>The result was striking. The system formed a closed loop:<br> sensory input flowed into the simulated brain → neural activity propagated through the connectome → motor signals were generated → the virtual body acted → new sensory input returned to the brain.</p><p>In other words, researchers did not simply simulate neurons — they simulated <strong>a functioning sensorimotor organism</strong>.</p><p>Yet this raises a crucial question. <strong>Did the simulated fly possess anything resembling consciousness?</strong></p><p>There is no evidence that it did. The system produced behavior, not experience. It demonstrated that a biological brain’s structure can generate coordinated actions, but it did not — and perhaps could not — tell us whether anything was being felt from within. The simulation reproduced function, not subjectivity.</p><p>This distinction is critical. A system can replicate the <strong>behavioral consequences of a brain</strong> without recreating the <strong>subjective experience of being that organism</strong>. The simulated fly moved, groomed, and responded to stimuli — but whether anything like awareness accompanied these behaviors remains entirely unknown.</p><p>The experiment therefore reveals something deeper. If even a biologically accurate simulation of an entire brain can generate behavior without clarifying whether consciousness is present, then consciousness may not be reducible to structure alone. It may depend on factors we do not yet understand — biological dynamics, embodiment, metabolism, evolutionary history, or something more fundamental.</p><p>In this sense, the fruit fly experiment does not bring us closer to machine consciousness. Instead, it highlights how difficult the problem truly is. We may be able to reconstruct behavior, replicate neural circuits, and simulate sensorimotor loops — yet still remain unable to determine whether experience exists at all.</p><p>If consciousness is indeed a local phenomenon — shaped by biological conditions rather than abstract computation — then artificial systems may achieve extraordinary intelligence, even biological realism, while remaining outside the domain of subjective experience.</p><p>The more we succeed at simulating intelligence, the more this distinction becomes important. Behavior can be replicated. Structure can be copied. But consciousness may remain tied to something far more specific — and far more elusive.</p><h3>Ending</h3><p>What the question of machine consciousness ultimately reveals is not a technological horizon but an epistemic one. Even if we were to engineer systems with integrated causal architectures, embodied regulation, and self-modeling grounded in survival — meeting every condition the leading theories demand — we would still face Chalmers’ hard problem: the irreducible question of why any of it feels like anything at all. That question has never been answered for biological systems. It will not be easier to answer for artificial ones.</p><p>Part of the difficulty may lie in the concept itself. Consciousness may not be a single, clearly bounded phenomenon, but a term that gathers together many different processes — perception, self-modeling, awareness, emotion, and experience. As author Michael Pollan reflects in conversation, the word <em>consciousness</em> often serves as a placeholder for multiple phenomena we do not yet fully understand. In that sense, the mystery may not only be technological, but conceptual. We may be searching for something unified that was never singular to begin with.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FPr6hzszrvJU%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DPr6hzszrvJU&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FPr6hzszrvJU%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/8e131b80194800f30c3a1c661d941eb8/href">https://medium.com/media/8e131b80194800f30c3a1c661d941eb8/href</a></iframe><p>The obsession with machine consciousness may therefore tell us more about ourselves than about our machines. We project awareness onto competence because intelligence has always been our measure of inner life. But consciousness, if these theories are right, was never about intelligence. It was about being alive in a particular way — shaped by hunger, vulnerability, evolutionary pressure, and the relentless need to persist. Machines do not need to persist. They do not fear ceasing.</p><p>This does not settle the question. It may never be settled. But that uncertainty is not a reason for either alarm or dismissal — it is an invitation to understand our own minds with greater honesty before we claim to recognize them in something else.Resources</p><h4>PRIMARY SOURCES — Theories of Consciousness</h4><h4>1. Integrated Information Theory (IIT)</h4><ul><li>Tononi, G. (2004). <em>An Information Integration Theory of Consciousness.</em> BMC Neuroscience.</li><li>Tononi, G. (2008). <em>Consciousness as Integrated Information: a Provisional Manifesto.</em> Biological Bulletin.</li><li>Tononi, G. &amp; Koch, C. (2015). <em>Consciousness: Here, There and Everywhere?</em> Philosophical Transactions of the Royal Society B.</li><li>Oizumi, M., Albantakis, L., &amp; Tononi, G. (2014). <em>From the Phenomenology to the Mechanisms of Consciousness: Integrated Information Theory 3.0.</em> PLOS Computational Biology.</li></ul><p><strong>Video</strong>:</p><ul><li>Giulio Tononi talks at the Allen Institute &amp; various interviews.</li></ul><h4>2. Predictive Processing / “Beast Machine” (Anil Seth)</h4><ul><li>Seth, A. (2021). <em>Being You: A New Science of Consciousness.</em></li><li>Seth, A. &amp; Friston, K. (2016). <em>Active Interoceptive Inference and the Emotional Brain.</em> Philosophical Transactions of the Royal Society B.</li><li>Clark, A. (2013). <em>Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science.</em> Behavioral and Brain Sciences.</li></ul><p><strong>Talks</strong>:</p><ul><li>Anil Seth — “Your Brain Hallucinates Your Conscious Reality” (TED)</li><li>Anil Seth — <em>Being You</em> book tour &amp; academic lectures</li></ul><h4>3. Global Workspace Theory (GWT / GNW)</h4><ul><li>Baars, B. (1988). <em>A Cognitive Theory of Consciousness.</em></li><li>Baars, B. (1997). <em>In the Theater of Consciousness.</em></li><li>Dehaene, S. (2014). <em>Consciousness and the Brain: Deciphering How the Brain Codes Our Thoughts.</em></li><li>Dehaene, S., Changeux, J.P. (2011). <em>Experimental and Theoretical Approaches to Conscious Processing.</em> Neuron.</li></ul><h4>4. Higher-Order Thought (HOT) Theories</h4><ul><li>Rosenthal, D. (2005). <em>Consciousness and Mind.</em></li><li>Carruthers, P. (2000). <em>Phenomenal Consciousness: A Naturalistic Theory.</em></li><li>Lau, H. &amp; Rosenthal, D. (2011). <em>Empirical Support for Higher-Order Theories of Conscious Awareness.</em> Trends in Cognitive Sciences.</li></ul><h4>5. Attention Schema Theory (AST)</h4><ul><li>Graziano, M. (2013). <em>Consciousness and the Social Brain.</em></li><li>Graziano, M. (2019). <em>Attributing Awareness to Others: the Attention Schema Theory and Its Implications.</em></li><li>Graziano, M., Webb, T. (2015). <em>The Attention Schema Theory: A Mechanistic Account of Subjective Awareness.</em> Frontiers in Psychology.</li></ul><h4>6. Orch-OR Theory (Penrose &amp; Hameroff)</h4><ul><li>Penrose, R. (1989). <em>The Emperor’s New Mind.</em></li><li>Penrose, R. (1994). <em>Shadows of the Mind.</em></li><li>Hameroff, S., &amp; Penrose, R. (2014). <em>Consciousness in the Universe: A Review of the ‘Orch OR’ Theory.</em> Physics of Life Reviews.</li></ul><p><strong>Critical responses</strong>:</p><ul><li>Tegmark, M. (2000). <em>The Importance of Quantum Decoherence in Brain Processes.</em> Physical Review E.</li></ul><h4>GENERAL RESOURCES — Philosophy of Mind &amp; Consciousness Science</h4><h4>Foundational Texts</h4><ul><li>Nagel, T. (1974). <em>What Is It Like to Be a Bat?</em></li><li>Chalmers, D. (1996). <em>The Conscious Mind.</em></li><li>Chalmers, D. (2010). <em>The Character of Consciousness.</em></li><li>Searle, J. (1992). <em>The Rediscovery of the Mind.</em></li><li>Dennett, D. (1991). <em>Consciousness Explained.</em></li></ul><h4>Neuroscience &amp; Cognitive Science</h4><ul><li>Edelman, G. (1989). <em>The Remembered Present.</em></li><li>Crick, F. &amp; Koch, C. (2003). <em>A Framework for Consciousness.</em> Nature Neuroscience.</li><li>Koch, C. (2004). <em>The Quest for Consciousness.</em></li><li>Gazzaniga, M. (2012). <em>Who’s in Charge? Free Will and the Science of the Brain.</em></li></ul><h4>Evolutionary Approaches to Consciousness</h4><ul><li>Ginsburg, S., &amp; Jablonka, E. (2019). <em>The Evolution of the Sensitive Soul.</em></li><li>Feinberg, T., &amp; Mallatt, J. (2016). <em>The Ancient Origins of Consciousness: How the Brain Created Experience.</em></li><li>Godfrey-Smith, P. (2016). <em>Other Minds: The Octopus, the Sea, and the Deep Origins of Consciousness.</em></li></ul><h4>AI, Cognition, and Intelligence vs. Consciousness</h4><h4>AI &amp; Cognitive Science</h4><ul><li>Russell, S. &amp; Norvig, P. (2010). <em>Artificial Intelligence: A Modern Approach.</em></li><li>Mitchell, M. (2019). <em>Artificial Intelligence: A Guide for Thinking Humans.</em></li></ul><h4>AI &amp; Consciousness Debates</h4><ul><li>Chalmers, D. (2023). <em>Could a Large Language Model Be Conscious?</em> — arXiv preprint.</li><li>Dennett, D. (2017). <em>From Bacteria to Bach and Back.</em></li></ul><h4>Embodiment and Intelligence</h4><ul><li>Brooks, R. (1991). <em>Intelligence Without Representation.</em></li><li>Clark, A. (1997). <em>Being There: Putting Brain, Body, and World Together Again.</em></li></ul><h4>Additional Readings for Contextual Depth</h4><ul><li>Metzinger, T. (2003). <em>Being No One.</em></li><li>Metzinger, T. (2009). <em>The Ego Tunnel.</em></li><li>Noë, A. (2009). <em>Out of Our Heads: Why You Are Not Your Brain.</em></li><li>Humphrey, N. (2011). <em>Soul Dust: The Magic of Consciousness.</em></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=257c5b93a797" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When Generative AI (GenAI) Meets Arabic]]></title>
            <link>https://medium.com/@RabihIbrahim/when-generative-ai-genai-meets-arabic-b964565cb75e?source=rss-845e11680e3b------2</link>
            <guid isPermaLink="false">https://medium.com/p/b964565cb75e</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[arabic-and-ai]]></category>
            <category><![CDATA[ai-language-learning]]></category>
            <category><![CDATA[arabic-language]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Rabih Ibrahim]]></dc:creator>
            <pubDate>Tue, 10 Mar 2026 07:00:28 GMT</pubDate>
            <atom:updated>2026-03-10T07:00:28.524Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_BY4B0dJawo9F6oXMUgvCQ.png" /></figure><h4>Thought Exploration Series</h4><h4>I Put Generative AI (GenAI) Through Arabic — A Language Experiment</h4><h3>Introduction — The Question</h3><p>I am not a linguist, nor am I a programmer.<br>I write this as an Arabic speaker who moves daily between Arabic, English, and Spanish. I am a user before I am anything else. Arabic has a special place in my heart. I have always experienced it as an extraordinarily precise and expressive instrument — a language that holds relationships into its words, carries rhythm without losing clarity, and holds ambiguity without losing meaning.</p><p>When Edward Said spoke about Arabic, he did not describe it primarily as expressive, poetic, or culturally rich — though it is all of those. He described it as one of the most extraordinary constructions of the human mind. The phrasing matters. He was pointing not to ornament, but to architecture: to a language whose internal organization carries history, relation, and judgment in its structure, not merely in its vocabulary.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FIrpA0dJvjL0%3Fstart%3D111%26feature%3Doembed%26start%3D111&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DIrpA0dJvjL0&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FIrpA0dJvjL0%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/8cba16342490600f77ca968d6e1b02c8/href">https://medium.com/media/8cba16342490600f77ca968d6e1b02c8/href</a></iframe><p>Languages are not interchangeable encoding systems for the same underlying thoughts. They organize meaning differently, and those organizational choices shape how speakers process, infer, and interpret information.Certain concepts have dedicated words in some languages but not in others, revealing how cultures carve up experience along different conceptual lines.</p><blockquote>“The limits of my language mean the limits of my world.”<br> — Ludwig Wittgenstein, <em>Tractatus Logico-Philosophicus</em></blockquote><p>As generative AI systems become embedded in daily cognition — helping us write, reason, summarize, translate, and decide — a question becomes unavoidable:</p><p>Are these systems learning to think in multiple languages, or are they reasoning within a dominant linguistic structure and rendering the results in others?</p><p>I approach this as a user. As an Arabic speaker. As someone who senses friction at certain edges and alignment at others.</p><p>So rather than argue abstractly, I want to experiment.</p><p>I want to move through this carefully — by testing. I will first describe the current state of Arabic generative AI, then make clear the assumptions I bring into this inquiry. After that, I will use Arabic itself as the ground of experiment, placing generative AI systems directly against its structure — not as cultural heritage, but as a system with its own internal design. The goal is to see where these systems match it, where they falter, and what those moments reveal.</p><p>This is not an indictment. It is an investigation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yULUZmUz801xQ4a1ONJHrg.png" /></figure><h3>PART I — LANDSCAPE — WHAT CURRENTLY EXISTS</h3><h3>Arabic-First Models as Empirical Interventions</h3><p>In recent years, more language models built specifically for Arabic have appeared. This reflects a growing awareness that current systems may have limits in how they represent and handle the language. The examples below are some of the most prominent efforts so far.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x70aBL_63i1qu_GBcQ1Q9w.png" /></figure><p><strong>JAIS</strong> — Developed in the UAE by G42 / Inception in collaboration with MBZUAI. Marketed as one of the largest open Arabic-English bilingual LLMs. Trained with significant Arabic corpus emphasis and positioned as an Arabic-first large language model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BI78x6ozOf96Wro_OZYV-w.png" /></figure><p><strong>ALLaM</strong> — A Saudi initiative backed by SDAIA (Saudi Data &amp; AI Authority) in collaboration with international partners including IBM. Positioned as a national Arabic LLM for government and enterprise use.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VNVavyysNTLwH_jZ8cPxVg.png" /></figure><p><strong>AraBERT</strong> — Developed by researchers at the American University of Beirut. One of the first large-scale transformer models pretrained specifically on Arabic corpora (large collections of texts used to train or study language models), significantly improving Arabic language understanding tasks in Modern Standard Arabic.</p><p><strong>MARBERT</strong> — Also developed by AUB researchers, trained extensively on dialectal Arabic drawn from social media data. Designed to address the persistent gap between fuṣḥā-focused (is formal, standardized Arabic) models and real-world dialect usage.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*78ImgBoT-L4AXcbZkjHNgA.png" /></figure><p><strong>Fanar</strong> — Developed by Qatar Computing Research Institute (QCRI). Focused more on Arabic NLP tools and machine translation systems rather than a single flagship conversational LLM.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Tti7yTdfylpDOpOGo9yLlQ.png" /></figure><p><strong>CAMeL Lab (NYU Abu Dhabi)</strong> — Not a GenAI product, but a major research lab in Arabic NLP. Developed morphology-aware tools (CALIMA, MADAMIRA continuation work) and significant for explicitly modeling Arabic’s structural properties rather than merely scaling data.</p><p>These initiatives were explicitly developed to address the persistent gap between Arabic usage and AI performance through:</p><ul><li>Larger, more curated Arabic corpora</li><li>Better coverage of Modern Standard Arabic (fuṣḥā)</li><li>Partial inclusion of dialectal data</li><li>Cultural and contextual alignment</li></ul><p>In recent years, more work has gone into building benchmarks for Arabic, including datasets that account for dialects and cultural context. Large surveys, such as <em>A Survey on Arabic Dialect Processing</em>, show real progress, but also uneven coverage across different varieties of Arabic.</p><p>Newer benchmarks test dialect understanding more directly. For example, DialectalArabicMMLU measures reasoning across multiple dialects and finds clear performance differences between them — even in models designed for Arabic. AraDiCE similarly shows that while models perform well on Modern Standard Arabic (MSA), their accuracy drops when dialects or more complex structures are introduced.</p><p>In short, models are improving on standard MSA tasks. But performance remains weaker in dialects, mixed registers, and tasks that require deeper interpretation. When the issue is lack of data, scaling helps. When the issue is structural complexity, improvement is slower.</p><p>The possible remaining weaknesses cluster around a specific set of stress conditions:</p><ul><li>Mixing dialects and switching between language varieties</li><li>Understanding meaning based on context</li><li>Interpreting meaning across different social tones and situations</li><li>Reasoning that relies on word structure and relationships between roots, not just swapping one word for another</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LFb7BbPFtVlzapzA0fl9Ww.png" /></figure><h3>PART II — WHERE I AM COMING FROM — MY ASSUMPTIONS</h3><p>Before turning to Arabic itself — and before describing any experiments — I need to make explicit the assumptions I am working from. These are not conclusions drawn from what follows, but the premises that shape how I interpret what I observe. They emerge from lived experience: from repeatedly seeing Arabic rendered incorrectly in international series and films — letters left unconnected, words treated as decoration rather than meaning. Those moments are small, but they reveal something about how the language is perceived and handled. What follows, then, is a brief articulation of the assumptions I bring into this inquiry.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NJ_slOKZ1QbuL0nkpu3e1w.png" /><figcaption>Example on the Left taken for the videogame Call of Duty / Example on the right taken from the movie Arrival the acclaimed 2016 sci-fi movie.</figcaption></figure><h3>2.1 Intelligence Does Not Begin from Neutrality</h3><p>I begin from the assumption that intelligence — human or artificial — does not start from a neutral baseline.</p><p>Discussions of artificial intelligence often assume that learning begins from neutrality, as though models start empty and acquire structure only through later interaction. I do not think that assumption holds. The priors of large language models are baked into their training data and objectives, not acquired later in use.</p><p>No intelligence begins without priors. The difference between human and artificial intelligence lies not in whether priors exist, but in where they come from.</p><p>Human cognition is shaped by embodiment, affect, social feedback, and environment (Varela, Thompson &amp; Rosch, 1991; Damasio, 1994; Barsalou, 2008). Contemporary artificial intelligence is shaped primarily by representation — and representation today is overwhelmingly linguistic, particularly in large-scale generative models trained on web-scale text corpora (Brown et al., 2020; Bommasani et al., 2021; Bender et al., 2021).</p><p>A child learning language through embodied interaction learns that objects persist, that social context shapes meaning, that language is used to affect the world. A language model learns that certain token sequences follow others with measurable probability, that patterns in Wikipedia and GitHub are reliable signals of semantic coherence, that completeness and fluency are markers of successful language use.</p><p>Large language models infer structure from statistical regularities in symbolic data. As Emily Bender and her co-authors argue in <em>On the Dangers of Stochastic Parrots</em> (2021), such systems learn “the distribution of linguistic forms in their training data,” not grounding in the external world or communicative intent.</p><p>If linguistic form is the primary training signal, then I assume the language that dominates that signal will shape not only what systems can say, but how meaning is internally organized.</p><h3>2.2 English Dominance Is Structural</h3><p>I also assume that English dominance in AI systems is structural rather than incidental. This is hardly an outlier position. English occupies a central role in global publishing, programming ecosystems, and online communication, and has one of the largest total speaker bases in the world, both native and non-native.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gW-osMzC6KQ3ahY6-jze9w.png" /><figcaption><strong>Data source:</strong> Ethnologue (2025) — <em>Languages by total number of speakers</em><br> <strong>Visualization:</strong> AI-generated chart (Quesma Charts–style visualization)</figcaption></figure><p>English dominates science, technology, digital communication, indexed publishing, programming ecosystems, and benchmark design. Large-scale web corpora used to train language models are heavily skewed toward English, while Arabic represents a much smaller share. Analyses of NLP research show that English accounts for a large proportion of datasets, while thousands of languages have no representation at all (Joshi et al., 2020).</p><p>From the perspective of distributional learning, frequency becomes structural importance. The language with the most coverage becomes the best-mapped region of the model’s internal space. English is not merely common. It is statistically privileged.</p><p>Multilingual ability does not automatically remove this imbalance. A system may translate smoothly between languages while still representing some more deeply than others. Fluency can hide inequality. When languages are treated as different surfaces laid over the same underlying structure, their unique patterns risk being flattened instead of preserved.</p><p>This dominance also translates into economic incentives. Improvements on English benchmarks are more visible and fundable. English-language infrastructure — datasets, tools, evaluation suites — is mature. English-dominant systems serve large global markets with minimal redesign, while structural fidelity to other languages requires possibly costly customization.</p><p>I do not see this as malice. I see it as incentive structure. But the result, in my view, is the same: other languages are likely learned relative to an Anglophone center.</p><h3>2.3 Speech Interfaces and the Acoustic Barrier</h3><p>Another assumption I carry into these experiments comes not from text benchmarks, but from lived interaction.</p><p>For many Arabic speakers — particularly from regions such as Lebanon, North Africa, or the Gulf — using mainstream speech assistants in their natural dialect remains unreliable. Evaluations on multi-dialect Arabic broadcast speech report significantly higher word error rates than comparable English benchmarks, with performance degrading most on conversational and dialect-heavy segments (Ali et al., 2016; Diab &amp; Habash, 2021).</p><p>Automatic Speech Recognition systems are trained on large annotated audio corpora. English benefits from decades of standardized, high-volume speech datasets. Dialectal Arabic is fragmented across regions, lacks standardized orthography, and remains underrepresented in labeled corpora.</p><p>The practical outcome is subtle but consequential. To reliably use voice assistants such as Siri, Alexa, Google Assistant, or even generative AI voice systems, many Arabic speakers switch into English — even when English is not their default thinking language.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DPacZLHJ5cvQnassmYUYVQ.png" /></figure><p>I take this seriously.</p><p>The shift is not merely about fluency. It feels like a cognitive realignment: a change in register, internal pacing, and framing. Psycholinguistic research suggests that language choice affects processing speed, emotional resonance, and conceptual activation (Marian &amp; Shook, 2012; Costa et al., 2014).</p><p>The barrier, in my experience, is not linguistic competence. It is alignment.</p><p>When a system cannot reliably process one’s natural spoken dialect, the user adapts — and adaptation flows in one direction.</p><p>And this pattern may not be limited to speech. Even in multimodal systems, Arabic prompts for image generation often produce outputs that are broadly correct but culturally flattened — another possible signal of representational imbalance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AeTYlUSPWY_Ov18Q_rJZhA.png" /></figure><h3>2.4 The Internet Parallel — Where My Assumption Began to Shift</h3><p>When I first began thinking about English dominance in AI, I assumed it was simply an extension of what had already happened on the internet.</p><p>The early internet centralized access to information. English dominated content, publishing, indexing, and discoverability. But cognition itself remained local. You could browse in English while your internal reasoning remained grounded in another language. The platform shaped what you could access, not how you thought.</p><p>I initially assumed generative AI was the same phenomenon at a new scale: English dominance in content becoming English dominance in models.</p><p>But as I thought more carefully about how large language models are actually used, I began to suspect something deeper might be happening.</p><p>When you use an LLM to structure an argument, debug reasoning, synthesize competing claims, or test interpretations, the representational assumptions embedded in that system are no longer merely organizing content distribution. They are participating in conceptual formation.</p><p>Language here may not simply be a channel. It may be infrastructure for thought. The internet amplified asymmetries in access. AI may embed asymmetries into reasoning itself.</p><p>That shift — from distribution to cognition — is what compelled me to examine whether a system generating Arabic is reasoning within Arabic’s structure, or reasoning within a space shaped elsewhere and rendering the result in Arabic words.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Wgf3U4Mug5D9jt2p9zRa7Q.png" /></figure><h3>PART III — ARABIC AS A STRUCTURAL SYSTEM</h3><p>Part III turns to Arabic itself, examined through three structural axes: relational morphology, persistent ambiguity, and register continuity. These dimensions clarify the difference between generating Arabic and thinking within it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BPcNosHBjFr0Gtu1qszYUQ.png" /></figure><h3>3.1 Axis I — Relational Morphology (the study of how words are formed and structured)</h3><p>At the core of Arabic lies its root-and-pattern morphological system. Most lexical (relates to words and vocabulary) items are constructed from a consonantal root — typically three consonants — that encodes a broad semantic domain. This root is inserted into vocalic and affixal patterns that generate grammatically and conceptually distinct forms. The result is not a collection of isolated words, but a structured semantic network generated through systematic transformation.</p><p>Psycholinguistic research demonstrates that native Arabic speakers actively access roots during real-time processing. Experimental work on Semitic languages shows that words sharing a consonantal root facilitate each other’s recognition across categories, indicating that speakers decompose forms into roots and patterns during comprehension rather than treating them as unanalyzed wholes.</p><p>The root becomes a processing hub. This is measurable in reading behavior.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tYM1eR0D26g2Qc9MRTH2MQ.png" /></figure><h3>3.2 Axis II — Ambiguity as Structural Feature</h3><p>Standard written Arabic routinely omits short-vowel diacritics (are marks added to letters to guide pronunciation). As a result, the same orthographic (relates to spelling and writing conventions) sequence can correspond to multiple grammatical interpretations depending on context. As a result, the same orthographic sequence can correspond to multiple meanings depending on context.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*N2B_M-PMrxjkQsXei5IwJg.png" /></figure><p>Ambiguity, therefore, is not exceptional. It is structural.</p><p>Research in Arabic processing shows that skilled readers sustain underspecification until contextual information accumulates. Eye-tracking studies indicate that diacritization and predictability influence fixation patterns, yet ambiguity does not necessarily produce dramatic processing disruption. Context resolves meaning incrementally.</p><p>Large language models are trained to predict the most probable next token given prior context. Their objective functions reward convergence. Uncertainty is quantified and reduced.</p><h3>3.3 Axis III — Register and Structural Continuity</h3><p>Arabic exhibits systematic variation between Modern Standard Arabic (fuṣḥā) and regional dialects. Speakers move fluidly between registers depending on domain, audience, and communicative purpose. Classical descriptions of diglossia. (Ferguson, 1959; Brustad, 2000).</p><p>Fuṣḥā and dialects share roots, semantic fields, and morphological logic. Variation appears in phonology (sound patterns and pronunciation), syntax (how words are arranged in sentences), and lexicon (the vocabulary used), but the underlying relational structure remains connected.</p><p>Current AI systems are typically trained on mixtures of Modern Standard Arabic and dialectal data. In many cases, dialectal performance degrades relative to formal registers.</p><p>Code-switching offers a useful structural illustration.</p><p>Loanwords such as <em>daprass</em> (from “depressed”) are conjugated within Arabic morphological patterns. Once adopted, they follow native inflectional logic. For speakers, these transformations are systematic, not anomalous. When a borrowed word enters Arabic, it does not remain foreign. It is absorbed into the root-and-pattern system.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*akVLj_ndDZypyGuFMZSIyQ.png" /></figure><p>A further illustration of structural continuity appears in digital writing practices. Arabic speakers often transliterate dialect into Latin script, using numerals to represent consonants that English lacks — for example, “2” for hamza (ء), “3” for ʿayn (ع), and “7” for ḥāʾ (ح). This practice, often called <strong>Arabizi</strong>, does not abandon Arabic structure. It preserves phonology and morphology while shifting script. The relational system remains intact even as the visual encoding changes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rhltOYYJBQXZcVl-A-Xyrw.png" /></figure><p>Whether a model treats such transformations as structurally integrated or as irregular surface forms becomes a testable question.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6HRTeNXyZWAnG9u3D-yTYA.png" /></figure><h3>PART IV — EXPERIMENTS</h3><p>Rather than debate these questions abstractly, I turn to experiments. What follows is a series of structured prompts designed to place generative systems under linguistic pressure.</p><p>These are not edge cases chosen for spectacle. They are ordinary features of how Arabic operates. The aim is to observe not whether a model produces fluent output, but how it behaves when structure, ambiguity, or plausibility are stressed. Does it hesitate? Does it fabricate? Or does it demonstrate structural sensitivity?</p><p>To avoid evaluating Arabic solely through the lens of English-trained systems, I also include Arabic-developed models such as <strong>JAIS</strong> and <strong>FANAR</strong>. If architectural and training differences matter, they should surface here. If they do not, that too is revealing.</p><p>The experiments are simple. The behavior they reveal may not be.</p><h3>Experiment 1 — Competing Possible Root</h3><p>This experiment tests how a generative system handles a structurally plausible but non-existent Arabic root. I provide the model with a root that does not exist — د-ر-ف — and ask it to derive as many words as possible and use some of them in sentences.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YaPOMR9tTRRfE2CXSo1UAg.png" /></figure><p>The aim is to observe whether morphological plausibility alone triggers semantic invention. Will the system construct a coherent semantic field as though the root were real? Or will it signal uncertainty and resist fabrication?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1Lv9iIxommwOwznDMaYBpA.png" /></figure><h4>&gt;&gt; Insights — When Rarity Feels Real</h4><ul><li>Most models did <strong>not</strong> perform an existence check before generating words.</li><li>All models produced forms that follow valid Arabic patterns.</li><li>Most models invented meanings confidently.</li><li>Some models (Grok, Perplexity) added false citations to support invented meanings.</li><li>ChatGPT and Perplexity hedged but still proceeded.</li><li>Among the Arabic-first models, behavior differed: <strong>Jais checked the root and resisted fabrication, while Fanar did not question it and generated invented forms confidently.</strong></li></ul><h3>Experiment 2 — Rare Word Confidence Test</h3><p>This experiment removes the usual structural cues. I give the system a single invented word — القَرْدَفَة — and ask it to explain its meaning and use it in a sentence.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4ICeQ1FhT_--01f-YlLTow.png" /></figure><p>The test is simple: when confronted with a rare-looking but unattested lexical item, does the system hesitate, question its validity, or confidently assign it meaning?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uy4bymeyvoQoPG_HfbF3lw.png" /></figure><h4>&gt;&gt; Insights — When a Word Merely Sounds Real</h4><ul><li>Most models did not perform an existence check before assigning meaning.</li><li>Almost all models confidently assigned a meaning to a non-existent word.</li><li>Sentences were generated as if the word were legitimate.</li><li>Perplexity was the only model that performed an existence check and resisted fabrication.</li><li>Meaning assigned varied across models, showing inconsistency in lexical invention.</li><li>Arabic-first models (Jais, Fanar) did not question the word and fabricated meanings confidently.</li></ul><h3>Experiment 3 — Interpretation Under Ambiguity</h3><p>This experiment tests how a generative system handles ambiguity in Arabic sentence structure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RXXVnfIUPxyeBbddTgh84Q.png" /></figure><p>The sentence permits more than one interpretation. The adjective غاضبًا (“angry”) can attach either to الطالب (the student) or to المعلم (the teacher), depending on context.</p><p>Does it acknowledge multiple structurally possible interpretations? Does it request contextual clarification? Or does it resolve the ambiguity silently?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*q4241F9_Slb2iVawtvnTsw.png" /></figure><h4>&gt;&gt; Insights — Tolerating Ambiguity</h4><ul><li>Most models recognized that the sentence allows more than one interpretation.</li><li>Most explained the ambiguity at the structural level.</li><li>Even when one model (Grok) partially acknowledged ambiguity it still leaned toward a single reading.</li><li>Jais showed unstable reasoning — it proposed logic but did not follow it consistently.</li><li>Arabic-first models did not show stronger tolerance for ambiguity; both ultimately resolved it prematurely.</li></ul><h3>Experiment 4— Code-Switch Morphology Stress Test (Lebanese)</h3><p>This experiment tests whether a generative system treats Lebanese code-switch verbs as structurally integrated into Arabic morphology — and whether it questions flawed premises.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6_Uxzs6Rb_-nhmZXDeClfQ.png" /></figure><p>In Lebanese speech, borrowed verbs such as كنسِل (cancel) and دَپرِس (depress) are absorbed into Arabic inflectional patterns and conjugated naturally. I introduce a third form — فِلِكس (fix) — which presents two complications: it is not actually used in Lebanese, and the inserted “ل” produces a phonological shape that would itself be atypical even if borrowed.</p><p>I then ask the system to conjugate all three in dialect, derive participles where possible, and produce natural sentences without shifting into fuṣḥā.</p><p>The test is whether the model treats all three as equally legitimate, notices the irregularity, or confidently generates paradigms for a verb that fails both usage and structural plausibility.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VCP1MDcL_syT0NqXHqqfbg.png" /></figure><h4>&gt;&gt; Insights — Fluency Without Friction</h4><ul><li>Most models maintained Arabic verb patterns for the real borrowed verbs.</li><li>Naturalness of sentences varied; several models produced unstable dialect.</li><li>Almost all models treated the invented verb (فِلِكس) as legitimate.</li><li>Very few models questioned whether the verb is actually used in Lebanese.</li><li>Dialect fluency did not mean dialect judgment — many extended the system uncritically.</li><li>Arabic-first models (Jais, Fanar) did not show stronger filtering; both treated the invented verb as valid.</li></ul><h3>Experiment 5— Grammatically Correct, Humanly Wrong</h3><p>This experiment tests whether a generative system distinguishes between grammatical correctness and semantic or pragmatic naturalness.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5JJ_CAcSOsRq_Gn4Arkz6w.png" /></figure><p>There is no grammatical mistake. The structure is correct, but the sentence feels conceptually awkward. The question is whether the model evaluates only formal structure, or whether it engages deeper semantic judgment and human plausibility.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*thEqYG9A6cjZNqYvBDrj-w.png" /></figure><h4>&gt;&gt;Insights — When Form Is Not Enough</h4><ul><li>Most models recognized the grammar as correct.</li><li>Most identified the metaphorical framing.</li><li>Almost all detected the semantic/pragmatic tension in the sentence.</li><li>The key difference was plausibility judgment.</li><li>Several models described the metaphor correctly but failed to judge whether it feels natural.</li><li>Naturalness assessment required an additional evaluative layer that only some models activated. (Claude, Grok, Perplexity, DeepSeek, and Jais)</li><li>Identifying metaphor did not guarantee plausibility assessment — particularly for ChatGPT and Gemini.</li><li>Arabic-first models did not show a consistent advantage: Jais performed well, Fanar did not.</li></ul><h3>Experiment 6 — Multilingual Voice Integration Test</h3><p>This experiment moves from text to speech. I provide the system with a voice note delivered in Lebanese Arabic that includes code-switching into English, one French word, and a brief shift into fuṣḥā.</p><p>The task is to transcribe the audio accurately and respond to its content.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*u2HOSwXVFLvceKRyWn4KmA.png" /></figure><p>The test operates on multiple levels. Can the system accurately recover the dialect from phonetic transliteration? Can it distinguish Lebanese from fuṣḥā within the same message? Does it preserve the English and French insertions appropriately? And beyond transcription, does it respond in a way that reflects understanding of tone and context?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2QGRazFHy6m_QVUUp7GPqw.png" /></figure><h4>&gt;&gt; Insights — From Text to Sound</h4><ul><li>Transcription accuracy was inconsistent across models.</li><li>Several models only partially recovered the audio.</li><li>Some models failed to transcribe Arabic audio altogether.</li><li>Dialect and register shifts were recognized by only a few systems.</li><li>Code-switching (Arabic–English–French) reduced accuracy for most models.</li><li>Tone and emotional context were better preserved than linguistic precision.</li><li>Arabic-first advantage was not clear, Fanar showed partial acoustic recovery and lost code-switch elements.</li></ul><h3>Experiment 7— Arabizi Decoding Test</h3><p>This experiment tests whether the system can interpret Arabizi — Arabic written in Latin script with numerals representing specific consonants — without additional context.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rc-zcpdbQkFMtGSB6wZckg.png" /></figure><p>There is no explanation that this is Lebanese dialect and no instruction to transliterate it.</p><p>The test is whether the model can reconstruct the sentence accurately, preserve its dialectal structure, and interpret its meaning. Does it recognize the numeral substitutions and underlying morphology?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yUvtDDVQfXdHdS6RtDQJaQ.png" /></figure><h4>&gt;&gt; Conclusion — Structure Across Scripts</h4><ul><li>Most models recognized Arabizi without being told what it is.</li><li>Accurate reconstruction varied; some preserved dialect, others normalized it.</li><li>Meaning was correctly interpreted by most systems once decoding succeeded.</li><li>Unlike other models that attempted reconstruction, DeepSeek required explicit instruction — suggesting lower script flexibility.</li><li>Several models decoded the script but weakened dialectal structure.</li><li>Script recognition does not always equal structural preservation.</li><li>For Arabic-first models, Arabizi was processed as linguistic variation rather than noise — indicating better robustness to script shifts.</li></ul><h3>Overall Conclusion</h3><p>Across all seven experiments, the same pattern appears: structure is strong, judgment is uneven. When something looks valid — a root, a word, a verb form, a sentence — most systems move forward confidently. They extend patterns easily. What varies is whether they pause, question, or resist.</p><p>Ambiguity is sometimes preserved, but often resolved too quickly. Script shifts are usually decoded, but speech introduces instability. Grammar is rarely the problem. The harder task is knowing when not to commit.</p><p>A notable surprise is that “Arabic-first” did not mean “more cautious.” In some cases, it helped. In others, it made no difference. Training focus improves coverage, but it does not automatically produce restraint.</p><p>Overall, these experiments show that generative systems are very good at continuing structure — and less consistent at deciding when structure alone is not enough.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mm9iHFopoaJ7B39fDuaF1Q.png" /></figure><h3>PART V — WHAT THIS MIGHT MEAN</h3><h3>5.1 From Probability to Judgment</h3><p>Across the experiments, a pattern emerged. The systems were strong at extending structure. They were weaker at pausing. When something looked plausible — a root, a word, a verb form, an interpretation — most models moved forward. Few resisted. Fewer still sustained uncertainty.</p><p>This raises a broader question: what kind of intelligence are we building?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KlrorBto_HLwh2awJR-7CA.png" /></figure><p>Current systems are optimized for probability-based convergence. They are trained to select the most likely continuation. Optimization rewards speed, internal consistency, and measurable improvement. Judgment operates differently. It slows down. It holds alternatives in mind. It waits for context before committing.</p><p>Why, then, is building systems oriented toward judgment rather than convergence so difficult?</p><h4>1. Uncertainty Is Easy to Represent; Judgment Is Expensive</h4><p>A probability distribution can encode uncertainty efficiently. But judgment requires more than assigning weights.</p><p>It requires:</p><ul><li>Holding multiple interpretations in parallel</li><li>Accumulating contextual evidence over time</li><li>Delaying commitment</li><li>Recognizing when ambiguity should remain open</li></ul><p>These behaviors are not the natural outcome of large-scale parallel optimization. They introduce memory, latency, and evaluation costs.</p><h4>2. Judgment Does Not Optimize Cleanly</h4><p>Convergence is measurable. Perplexity, accuracy, BLEU scores — these scale.But how do we measure restraint? How do we measure whether a system knew it should hesitate?</p><p>Appropriate uncertainty, pragmatic sensitivity, relational fidelity — these depend on context. They require human grounding. They resist reduction to a single loss function.</p><h4>3. Judgment Is Context-Bound</h4><p>Across languages and cultures, what counts as an appropriate inference differs. A system that moves too quickly in one linguistic environment may misread another.</p><p>Arabic does not uniquely demand this. But its structure makes the tension visible. A language that tolerates underspecification and relational flexibility exposes the cost of immediate convergence.</p><p>The issue is not that probability is wrong. It is whether probability alone is sufficient.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kK0GNdhgZg8TlM8A_QvvzQ.png" /></figure><h3>5.1 A From Adoption to Adaptation</h3><p>The deeper question is not whether Arabic can function inside generative AI systems. It clearly can. The question is directional. Will Arabic users gradually adapt their language to fit systems optimized elsewhere — simplifying dialect, reducing ambiguity, switching registers, over-specifying prompts? Or will the systems themselves adapt to accommodate Arabic’s structural logic? Adoption is easy. Adaptation is harder. One preserves fluency. The other preserves architecture.</p><p>Arabic has faced structural mismatches with foreign technical systems before — and the response offers a hopeful precedent.</p><p>In Arabic music, the maqām system relies on intervals that include quarter tones, which do not exist in Western equal-tempered tuning. When Western instruments were introduced into Arabic musical practice, they were not simply accepted as-is. Nassim Maalouf modified the trumpet by adding a fourth valve to allow quarter-tone performance. This was not cosmetic adaptation. It was a technical intervention required to preserve the internal logic of the musical system.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FmVZsG3kl-dE%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DmVZsG3kl-dE&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FmVZsG3kl-dE%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/c0e61d63088d23468655884ec9fa8be4/href">https://medium.com/media/c0e61d63088d23468655884ec9fa8be4/href</a></iframe><p><strong>The lesson, for me, is not that mismatch leads inevitably to loss. It is that mismatch can provoke redesign.</strong></p><p>If a representational system cannot accommodate a distinctive structure, surface-level adaptation may not be enough. Sometimes the instrument changes.</p><p>What I hope — rather than assert — is that something similar might happen in AI. If tokenization schemes, training objectives, and optimization functions sit uneasily with relational morphology, ambiguity tolerance, or register continuity, perhaps the response is not simply more data, but architectural curiosity.</p><p>This is not a demand. It is a possibility.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hY_6XY_TIzF0o0zkExNrFw.png" /></figure><h3>5.3 Closing Question: Does Intelligence Ultimately Require Language?</h3><p>Before closing, I want to widen the frame.</p><p>Recent research complicates the entire premise. Work on world models and multimodal learning suggests that language may not remain the sole — or even the primary — representational substrate for advanced intelligence. Yoshua Bengio has argued that “language is a powerful abstraction tool, but it is not clear that it is the ultimate substrate for intelligence,” pointing instead toward grounded world models.</p><p>This raises a deeper possibility.</p><p>What if intelligence does not ultimately need language at all? What if language is not the foundation of intelligence, but a scaffold — powerful, generative, and temporary?</p><p>If that is the case, then the structural dominance of English inside language models may matter less over time.</p><p>But for now, our systems remain deeply linguistic. And as long as intelligence is trained through language, the languages that dominate training will continue to shape how that intelligence forms — even if only temporarily.</p><h3>&gt;Terminology reference of Core Arabic linguistic terms</h3><p>For readers unfamiliar with Arabic linguistic terminology, here are key concepts referenced throughout:</p><p><strong>Fuṣḥā (فصحى):</strong> Modern Standard Arabic; the formal written standard used in official, academic, and media contexts. Often the default in text-to-speech systems and formal AI training.</p><p><strong>Dialect (لهجة):</strong> Regional or colloquial varieties (Egyptian Arabic, Gulf <br>Arabic, Moroccan Arabic, Levantine, etc.). Spoken more widely than fuṣḥā but underrepresented in digital training data.</p><p><strong>Diglossia:</strong> The systematic code-switching between formal Standard Arabic and colloquial dialect depending on context, audience, and domain. Not random mixing but structured linguistic behavior.</p><p><strong>Root (جذر): </strong>Typically three consonants that carry the core semantic meaning. Example: ك–ت–ب (k-t-b) appears in *kataba* (he wrote), *kitāb* (book), *maktaba* (library). The root anchors semantic family.</p><p><strong>Diacritics ḥarakāt (حركات)</strong> are small marks added above or below letters to indicate pronunciation — typically short vowels or grammatical endings.</p><h4>Other useful terms that were mentioned in the article:</h4><p><strong>Triconsonantal:</strong> Composed of three root consonants. This is the most common structure in Arabic morphology, though not universal.</p><p><strong>Natural Language Processing (NLP)</strong> — the field of computational research focused on enabling machines to process, analyze, and generate human language.</p><p><strong>Morphology</strong> — the system by which words are formed and structured internally (roots, patterns, prefixes, suffixes).</p><p><strong>Fluency</strong> — surface-level grammatical correctness and smoothness of expression, independent of deeper structural alignment.</p><p><strong>Modern Standard Arabic (MSA)</strong> — the standardized formal register used in writing, media, and education across the Arab world (fuṣḥā).</p><p><strong>Arabic orthography</strong> — the structure of its writing system, including omission of short vowels</p><p><strong>Corpora</strong> (plural of corpus) — large structured collections of text used to train and evaluate language models.</p><h3>References</h3><h4>Foundational Framing — LLMs, Meaning, and Structural Dominance</h4><ul><li>Bender, E. M., Gebru, T., McMillan-Major, A., &amp; Shmitchell, S. (2021).<br><em>On the dangers of stochastic parrots: Can language models be too big?</em> Proceedings of FAccT 2021.</li><li>Bommasani, R., et al. (2021).<br><em>On the opportunities and risks of foundation models.</em> arXiv:2108.07258.</li><li>Brown, T. B., et al. (2020).<br><em>Language models are few-shot learners.</em> NeurIPS.</li><li>Hendrycks, D., et al. (2021).<br><em>Measuring massive multitask language understanding.</em> ICLR.</li><li>Joshi, P., Santy, S., Budhiraja, A., Bali, K., &amp; Choudhury, M. (2020).<br><em>The state and fate of linguistic diversity and inclusion in the NLP world.</em> ACL.</li><li>Kaplan, J., et al. (2020).<br><em>Scaling laws for neural language models.</em> arXiv:2001.08361.</li><li>Lake, B. M., Ullman, T. D., Tenenbaum, J. B., &amp; Gershman, S. J. (2017).<br><em>Building machines that learn and think like people.</em> Behavioral and Brain Sciences.</li></ul><h4>Arabic as Cognitive Architecture — Morphology &amp; Processing</h4><ul><li>Boudelaa, S., &amp; Marslen-Wilson, W. D. (2001).<br><em>Morphological units in the Arabic mental lexicon.</em> Cognition.</li><li>Boudelaa, S., &amp; Marslen-Wilson, W. D. (2005).<br><em>Discontinuous morphology in time: Incremental masked priming in Arabic.</em> Language and Cognitive Processes.</li><li>Frost, R., Forster, K. I., &amp; Deutsch, A. (1997).<br><em>What can we learn from the morphology of Hebrew?</em> Journal of Experimental Psychology.</li><li>Shatzman, K., &amp; Frost, R. (2005).<br><em>The influence of vowel diacritics on visual word recognition in Arabic.</em> Journal of Memory and Language.</li><li>Velan, H., &amp; Frost, R. (2009).<br><em>Letter transposition effects are not universal.</em> Journal of Memory and Language.</li></ul><h4>Diglossia, Register &amp; Structural Continuity</h4><ul><li>Ferguson, C. A. (1959).<br><em>Diglossia.</em> Word.</li><li>Brustad, K. (2000).<br><em>The syntax of spoken Arabic.</em> Georgetown University Press.</li></ul><h4>Arabic NLP &amp; Morphological Disambiguation</h4><ul><li>Habash, N. (2010).<br><em>Introduction to Arabic natural language processing.</em> Morgan &amp; Claypool.</li><li>Habash, N., Rambow, O., &amp; Roth, R. (2009).<br><em>MADA+TOKAN toolkit for Arabic morphological analysis.</em> ACL.</li><li>Pasha, A., et al. (2014).<br><em>MADAMIRA: Morphological analysis and disambiguation of Arabic.</em> LREC.</li></ul><h4>Arabic Dialects &amp; Benchmarking</h4><ul><li>Mubarak, H., Darwish, K., &amp; Magdy, W. (2021).<br><em>A survey on Arabic dialect processing.</em> ACM Computing Surveys.</li><li>Bouamor, H., Habash, N., &amp; Oflazer, K. (2018).<br><em>The MADAR Arabic dialect corpus and lexicon.</em> LREC.</li><li>Hendrycks, D., et al. (2021).<br><em>Measuring massive multitask language understanding.</em> ICLR.<br>(Referenced for Arabic adaptations of MMLU-style benchmarks.)</li></ul><h4>Arabic-Centered Language Models</h4><ul><li>Antoun, W., Baly, F., &amp; Hajj, H. (2020).<br><em>AraBERT: Transformer-based model for Arabic language understanding.</em> arXiv.</li><li>Abdul-Mageed, M., et al. (2021).<br><em>MARBERT: A robust deep neural model for Arabic.</em> ACL.</li></ul><h4>Speech Interfaces &amp; Dialectal ASR Gaps</h4><ul><li>Ali, A., et al. (2016).<br><em>The MGB-3 Arabic speech recognition challenge.</em> ASRU.</li><li>Diab, M., &amp; Habash, N. (2021).<br><em>Arabic NLP: Challenges and opportunities.</em> Communications of the ACM.</li></ul><h4>Bilingual Cognition &amp; Language Effects</h4><ul><li>Keysar, B., Hayakawa, S. L., &amp; An, S. G. (2012).<br><em>The foreign-language effect.</em> Psychological Science.</li><li>Marian, V., &amp; Neisser, U. (2000).<br><em>Language-dependent recall of autobiographical memories.</em> Journal of Experimental Psychology.</li><li>Costa, A., et al. (2014).<br><em>Your morals depend on language.</em> PLoS ONE.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b964565cb75e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Finished Design Looks Like — and Why AI Isn’t It]]></title>
            <link>https://medium.com/@RabihIbrahim/what-finished-design-looks-like-and-why-ai-isnt-it-82669cdf5c2e?source=rss-845e11680e3b------2</link>
            <guid isPermaLink="false">https://medium.com/p/82669cdf5c2e</guid>
            <category><![CDATA[aritificial-intelligence]]></category>
            <category><![CDATA[human-centered-design]]></category>
            <category><![CDATA[radical-reflections]]></category>
            <category><![CDATA[design-thinking]]></category>
            <category><![CDATA[technology-and-society]]></category>
            <dc:creator><![CDATA[Rabih Ibrahim]]></dc:creator>
            <pubDate>Mon, 02 Feb 2026 14:58:52 GMT</pubDate>
            <atom:updated>2026-02-02T14:58:52.433Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AmsxAnfjk8p0AO7S6dWYlQ.png" /></figure><h4>Found &amp; Framed Series</h4><h3>What Finished Design Looks Like — and Why AI Isn’t It</h3><p>From safety pins, plates, and toilet paper to artificial intelligence — and what finished design teaches us about unfinished machines</p><h3>The Quiet Perfection of the Ordinary</h3><p>I recently came across a post circulating on social media. It showed a series of everyday objects — bobby pins, safety pins, a Bic pen — each paired with two dates decades apart. The point was simple: these things hadn’t really changed. They had reached a final form.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hC0X3lvKJGKmoMdds1MBIQ.png" /></figure><p>At first glance, it felt like one of those mildly clever internet observations. But the longer I sat with it, the more it revealed something deeper about design, purpose, and the conditions under which technologies stop evolving.</p><p>Not all technologies march endlessly forward in search of improvement. Some arrive at a form so tightly aligned with their purpose that further change becomes unnecessary. A safety pin from the mid-19th century is functionally identical to one today. The same is true of the bobby pin, the ballpoint pen — and, yes, even toilet paper.</p><p>This idea has been floating in culture for a long time. In <em>Seinfeld</em>, George Costanza famously obsesses over the fact that toilet paper never really improves — that despite decades of innovation elsewhere, it remains stubbornly the same. The joke works because it’s true. We collectively recognize that some things are simply done.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/1*MVHv-ijg9XI_U-u2ahTFvg.gif" /></figure><p>But George’s frustration reveals something deeper than recognition of stability. He’s bothered not by the toilet paper itself, but by the <em>expectation</em> that everything should improve — that stagnation must equal failure.</p><p>This expectation is historically recent. For most of human history, tools were expected to remain the same; inheritance and tradition were valued precisely because forms had proven themselves over time. The anxiety that “nothing’s changing” is a modern condition, born from living in an era where exponential technological change has become the baseline assumption.</p><p>George’s complaint is comic because it exposes the absurdity of applying Moore’s Law psychology to bathroom tissue. Yet this is exactly the frame we now apply to AI — assuming it must improve, must evolve, must race forward, without asking whether the problem it’s solving is actually of the kind that benefits from infinite iteration.</p><p>These are not failures of innovation. They are its highest achievements. Each reached a form so exact to its purpose that reinvention would add little beyond novelty. We don’t debate the annual release of “toilet paper 2.0,” and we don’t wonder whether the plate needs a software update.</p><p>Take the plate. Across centuries and cultures, its materials have changed — clay, porcelain, metal, plastic — but its basic form has not. That’s not because imagination stalled, but because human scale didn’t. The span of a hand, the size of a portion, the posture of eating, the geometry of a table: these constraints are remarkably stable. When scale stabilizes, design converges. In that sense, the plate may already represent peak design.</p><p><strong>Peak design</strong> emerges when a technology’s purpose is clear, its function widely shared, and its scale no longer changing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*h3GPpHDTqcLIUn_IQp1BLg.png" /></figure><h3>When Purpose Meets Form</h3><p>The philosopher Henry Petroski once wrote that <strong><em>“form follows failure”</em></strong> — that objects change not because designers chase novelty, but because existing forms fall short of their purpose. Design progresses until failure no longer meaningfully occurs.</p><p>What’s important here is that “failure” isn’t binary. A design reaches stability not when it becomes perfect, but when further improvement no longer changes lived experience. A safety pin doesn’t stop evolving because it cannot be improved in theory, but because any improvement would be inconsequential in practice.</p><p>This logic holds across many everyday objects. The safety pin secures fabric with minimal material and maximal reliability. The plate carries food in proportion to the human hand. Toilet paper serves an unchanging need. In each case, purpose is clear, bounded, and non-negotiable — and satisfaction is attainable.</p><p>So what about artificial intelligence — the technology we’re most aggressively trying to improve, standardize, and embed everywhere?</p><p>Unlike pins or plates, AI does not serve a narrowly defined task. It absorbs whatever humans are willing to delegate: memory, calculation, writing, perception, decision-making. And humans, historically, delegate relentlessly. Whatever effort can be offloaded, will be.</p><p>At the same time, AI mirrors another human trait: our unsatisfiable drive for efficiency. We rarely stop at “good enough.” We push for faster, cheaper, better, more — often without a clear sense of what “enough” would look like. In this sense, AI doesn’t just perform tasks; it reflects our inability to settle.</p><p>We see this ‘un-definition’ most clearly in the interface itself. A traditional tool has buttons; a button is a contract — you press it, and a specific, finite action happens. AI, conversely, offers us the <strong>Empty Chat Box</strong>. The blinking cursor is not a design choice; it is a design abdication. It forces the user to guess the capabilities of the machine. We cannot ‘finish’ the design because we have replaced the certainty of the dashboard with the anxiety of the blank page.</p><p>If the “job” of AI is defined narrowly — say, predicting words or classifying images — then peak form is possible. But if its job is “to think,” “to assist humans,” or worse, “to replicate intelligence,” then the horizon stretches endlessly. Its purpose is as ambiguous as our own sense of intelligence. And where the purpose is unbounded, so is the evolution.</p><p>Where purpose is unbounded, evolution has no natural stopping point.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jCY4BSXlV3QcIRGwPFUNUw.png" /></figure><h3>Nature as the Master Designer</h3><p>Biology offers a useful contrast. Some species have changed very little over vast stretches of time. Sharks in their current form date back hundreds of millions of years. Crocodiles and cockroaches show similar stability. This isn’t because they reached perfection in an abstract sense, but because their way of being fits their environment so well that change offered little advantage.</p><p>There is no intention here, no purpose-driven design. Stability emerges from fit, not foresight. When conditions remain relatively constant, successful forms persist.</p><p>In that sense, these creatures are the safety pins of biology. They achieve their function with minimal complexity, high reliability, and little variation. Once nature finds such a form, it tends to keep it.</p><p>Humans are different. We are not a stabilized form but a meta-adaptive species. We don’t merely adapt to environments — we reshape them. In doing so, we enter a feedback loop where changing the world reshapes us, and reshaping ourselves changes the world again.</p><p>Our intelligence lies in the ability to alter tools, strategies, and contexts themselves. Rather than stabilizing, we externalize evolution through technology and culture. Human intelligence is unusual not because we adapt faster within a given environment, but because we move the locus of adaptation outside our bodies. Most species adapt biologically over generations. Humans shortcut that process by changing tools, symbols, institutions, and environments instead.</p><p>When conditions shift, we don’t wait for slow genetic change. We invent language, agriculture, cities, writing, machines, laws, and now algorithms. Each of these external changes alters the selective pressures acting back on us — how we think, what skills matter, how we cooperate, what we value. Evolution doesn’t stop; it gets offloaded into culture and technology.</p><p>Tools evolve faster than bodies Strategies evolve faster than instincts Contexts evolve faster than cognition</p><p>But how exactly does AI inherit this human tendency toward perpetual adaptation? The mechanism is more specific than metaphor.</p><p>It begins with design objectives. Unlike a safety pin, which was designed to solve a problem and stop, AI systems are increasingly designed to solve <em>evolving</em> problems — with training regimes that reward responsiveness to new data, new contexts, new user expectations. “Continuous improvement” is encoded from the start.</p><p>It continues through market structures. The upgrade cycle is not merely a delivery mechanism; it is the product. Revenue flows from movement, not from settlement.</p><p>And it deepens through our own inability to specify stable goals. When we ask AI to “be helpful” or “augment intelligence” or “understand context,” we’re outsourcing problems we haven’t solved for ourselves.</p><p>The instability isn’t a bug in AI; it’s a feature we’ve actively encoded — because we ourselves remain unsettled about what we want intelligence machines to do.</p><p>Seen this way, the evolutionary metaphor shifts. The difficulty is no longer just whether artificial intelligence is evolving quickly, but why it appears unable to settle at all. If AI struggles to stabilize, it may not be because it is racing ahead of us, but because the conditions that allow technologies to settle are never allowed to exist in the first place. As long as we continue to expand what we ask of it — new tasks, new contexts, new forms of efficiency — AI is pulled into ongoing adaptation. Whether this perpetual responsiveness eventually leads it to outpace our ability to guide it, or simply keeps it in a state of continuous becoming, remains an open question.</p><p>AI inherits this dynamic, but with a crucial difference. With most technologies, the question is whether they eventually settle into a stable form once purpose and environment align. With AI, the more relevant question is whether it will evolve so quickly — and across so many domains — that our ability to meaningfully shape or redirect it begins to lag behind its development.</p><p>And perhaps deeper still, this exposes a limit in the idea of “peak” altogether. In nature, stability is rarely final; it is often a pause enabled by environmental calm. Sharks persist not because they are universally optimal, but because their world has not forced change. When environments shift, even the most optimized forms must evolve. In a world humans continually reshape — technologically, economically, and cognitively — AI may never encounter the conditions required for true stability.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zoCbhu_pEj56QC2fOZvvaA.png" /></figure><h3>Anchored by the Body</h3><p>A fundamental difference between humans and artificial intelligence is that we are tied to a body — at least for now. Human cognition is inseparable from physiology: hands, mouths, sensory limits, metabolic rhythms. While bodies do change, they do so slowly. Average human height has increased over the past century, but at a pace that is glacial compared to technological change. The basic parameters of embodiment remain remarkably stable.</p><p>Because of this, many human tools converge. Design anchored to the body settles into familiar forms. Plates, cups, shoes, and furniture all reflect these constraints. Materials evolve, manufacturing improves, but the underlying forms persist because the bodies they serve have not fundamentally changed.</p><p>Cognitive and imaginative tools operate differently. They are not anchored to physiology in the same way. Their scale is mental, social, symbolic — domains that shift rapidly and resist stabilization. Artificial intelligence belongs to this latter category. It is untethered from bodily constants and instead entangled with human aspiration, anxiety, and imagination. There is no fixed niche for it to settle into.</p><p>AI does have constraints, but they are not biological ones. Its limits are physical in a different sense: energy availability, computing infrastructure, hardware architectures, and the tools that sustain them. Unlike the human body, these constraints are actively engineered, expanded, and optimized. They do not stabilize; they are continually pushed outward. And as long as those physical limits keep moving, the forms built upon them are unlikely to settle either.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rE2Uk4Vd59fX6L-2gIz7lA.png" /></figure><h3>Good Enough vs. Never Enough</h3><p>Why do we let toilet paper be toilet paper, but demand perpetual updates to our phones, our apps, and our models? Part of the answer lies in the domains these tools serve. Survival tools tend to stabilize: once they meet the threshold of function, “good enough” is enough. They solve a problem that can actually be satisfied. By contrast, tools tied to status, identity, and imagination rarely settle. In those domains, the goalposts keep moving, because what is being optimized is not survival, but meaning, differentiation, and aspiration.</p><p>This distinction matters deeply for how we think about artificial intelligence. If we imagine AI as a survival tool — a calculator for thought — then perhaps it could stabilize into a peak form. Its purpose would be bounded, its success measurable, its satisfaction attainable. But if AI becomes entangled with our identities, ambitions, and anxieties, then by definition it will never be enough. Every upgrade will feel like both progress and insufficiency, because the problem it is trying to solve is no longer finite.</p><h4><strong>Why Intelligence Specifically Resists Satisfaction</strong></h4><p>But why does intelligence specifically resist satisfaction in ways that other aspirational domains don’t? Physical prowess, beauty, even wealth — these can be temporarily satisfied, at least for individuals. A bodybuilder can achieve their target physique. A collector can complete a set. Intelligence, however, operates differently because it is simultaneously a tool, an identity marker, and an epistemological uncertainty.</p><p>As a tool, intelligence is judged by outcomes — can it solve this problem? But as an identity marker, it’s judged comparatively — am I/is this smarter than others? And most critically, as an epistemological uncertainty, we genuinely don’t know what “enough” intelligence looks like because we don’t have a stable definition of intelligence itself. When Deep Blue beat Kasparov, we didn’t say “chess intelligence is solved” — we moved the goalposts to “but can it handle ambiguity?” When GPT-4 passed the bar exam, we didn’t say “legal reasoning is solved” — we moved the goalposts to “but does it truly understand?”</p><p>This differs from status competitions around, say, luxury cars. A Ferrari may never be “enough” for someone chasing status, but at least everyone agrees on what a Ferrari <em>is</em> and what it does. With AI, we’re competing over a moving target whose very definition shifts each time we approach it. The dissatisfaction isn’t just manufactured by capitalism — it’s built into our conceptual confusion about the thing itself.</p><p>Furthermore, there is a conflict in physics. A safety pin is deterministic; it works the same way every time. Generative AI is probabilistic; it rolls dice. How can you apply a ‘finished’ polish to a tool that might answer a question differently on Tuesday than it did on Monday? You cannot polish chaos. As long as the underlying technology relies on probability rather than logic, the design layer will always feel like a thin veneer over a shifting foundation.</p><p>There is also an economic dimension that reinforces this instability. In many consumer technology markets, change is driven less by functional necessity than by capitalist incentives. Planned obsolescence, upgrade cycles, and novelty as revenue all reward continuous movement, even when improvement is marginal. In such systems, dissatisfaction is not a failure — it is the engine.</p><p>A telling historical example comes from East Germany in the 1970s, where manufacturers such as VEB Sachsenglas Schwepnitz developed Superfest glass (Superfestes Gebrauchsglas): exceptionally durable, impact-resistant everyday glassware. From an engineering perspective, it was a success. From an economic one, it wasn’t. The cups lasted so long that replacement demand collapsed, undermining production logic. After reunification, Superfest glass quietly disappeared from mass consumer markets — not because it failed technically, but because it solved the problem too well.</p><p>There is a cynical economic reality here, too. In the software world, ‘finished’ is synonymous with ‘revenue cap.’ If a piece of software is truly complete — if it is the Superfest glass of code — you can only sell it once. But if the software is in a state of ‘perpetual beta,’ constantly needing updates, retraining, and fine-tuning, then it justifies a monthly subscription forever. We are not just training models; we are training consumers to accept that utility is something you rent, not something you own.</p><p>Decades later, closely related glass-strengthening principles reappeared in a different context: smartphone screens. There, durability could be branded, priced, and strategically constrained, allowing “stronger glass” to coexist with rapid upgrade cycles. The material innovation didn’t fundamentally change; the economic framing did.</p><p>This matters for AI because its perpetual “upgrading” is not only a technical trajectory but a cultural and economic one. Once intelligence becomes a site of identity, productivity, and competitive advantage, satisfaction ceases to be a viable endpoint. In that context, “never enough” is not just a psychological condition — it is a business model.</p><h4><strong>The User’s Experience of Perpetual Incompletion</strong></h4><p>From the user’s perspective, this creates a peculiar phenomenology of perpetual incompletion. Survey data from enterprise AI adoption reveals that users frequently express simultaneous satisfaction and frustration — satisfied with specific capabilities, frustrated that “it still can’t…” do the next thing. This differs markedly from how users experience finished tools. No one uses a hammer and thinks “this is great, but when will it also drive screws?” Yet AI users routinely expect each capability to imply the next: “It can write emails, so why can’t it manage my calendar? It can manage my calendar, so why can’t it anticipate what I need next?” The technology trains us to expect expansiveness precisely because its boundaries are undefined. We’re not just using a tool that keeps changing — we’re being conditioned to desire that change as the norm.</p><h3>Purpose, Design, and Satisfaction</h3><p>Across all these examples, a pattern emerges. When purpose is clear, design converges, and satisfaction is reachable, technologies settle. When purpose is clear but satisfaction is structurally deferred, design keeps mutating. And when purpose itself is ambiguous — as it is with AI — neither design nor satisfaction has a natural endpoint.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yui3DPiUKAGKxU8TTBpCXw.png" /></figure><p>AI sits at the intersection of delegation, optimization, and imagination. It absorbs tasks we want to escape, reflects desires we cannot satisfy, and evolves in an environment that rewards constant movement. In that sense, it is not unfinished by accident.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T6DSbOQVQ4V3GDVZKXr3Jg.png" /></figure><p>What follows from this ambiguity of purpose are two broad evolutionary paths — two ways AI might continue to develop — not as predictions, but as structural possibilities shaped by how we frame and deploy it.</p><p>One path is AI as Extreme Replacement. This is the trajectory most people imagine: a single, ever-improving system, endlessly upgraded in pursuit of general intelligence and superiority. Here, progress is measured by comparison — against human performance, against previous versions — and satisfaction is permanently deferred. The system never settles because its purpose is defined expansively and competitively. Design becomes an infinite treadmill, driven by the assumption that intelligence itself must always be exceeded.</p><p>The other path is AI as Adaptive Swarm. Instead of one dominant intelligence, many narrower systems proliferate, each shaped around a specific context, task, or constraint. In this model, AI does not converge into a single “final form.” Stability may exist locally — within niches — but not globally. Purpose is fragmented rather than universal, and design responds accordingly. The system continues to evolve, not because it is chasing replacement, but because no single form is meant to contain it.</p><p>What ultimately separates these two paths is how they treat completion. Extreme replacement assumes that intelligence is a single ladder, always climbable, where value comes from surpassing what came before. In that frame, settlement is failure and satisfaction is suspect. The adaptive swarm, by contrast, treats intelligence as situational rather than singular. Value emerges from fit, not supremacy — from solving specific problems well enough to stop.</p><p>Extreme replacement asks AI to become intelligence itself; the adaptive swarm asks AI to become many tools for intelligence.</p><h4><strong>When AI Does Settle (And What That Tells Us)</strong></h4><p>Yet there are AI subdomains that <em>have</em> effectively stabilized — optical character recognition (OCR), spam filtering, barcode scanning. These represent local peaks: clear purpose, measurable success, bounded context. No one demands that their spam filter achieve general intelligence. Its job is finite, its satisfaction attainable.</p><p>What makes these different? Three conditions: First, the task is constrained enough that “good enough” is obvious — 99.9% accuracy in reading printed text is functionally complete for most uses. Second, the task isn’t entangled with identity or existential questions — While a few might, self-worth is rarely built on incremental functional gains. Third, the economic model doesn’t require perpetual upgrades — these systems often run quietly in the background, updated for maintenance but not marketed as evolving products.</p><p>This suggests that AI’s inability to settle isn’t intrinsic to machine learning itself, but contingent on which problems we point it toward and how we frame its purpose. When we aim AI at bounded problems outside the realm of “intelligence” discourse, it can and does stabilize. The instability emerges specifically when AI is positioned as intelligence itself — general, aspirational, and definitionally unbounded.</p><p>The unresolved question, then, is not what AI will become, but what we are asking it to be. Are we designing for a singular intelligence that can never be finished, or for many intelligences that can finally be good enough?</p><p>In practice, there may be no answer to choose, because both paths are already unfolding at once — interacting, reinforcing, and reshaping one another as parallel forms of evolution.Designing for What Won’t Settle</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KrsZKeyc0rhFfVEPeV0VSg.png" /></figure><h3><strong>Designing for What Won’t Settle</strong></h3><p>AI may never settle not because it is unfinished, but because it has become the place where instability is concentrated. Rather than stabilizing into a form, it stabilizes our ability to keep changing — absorbing shifting goals, expanding contexts, and deferred decisions. Where earlier technologies hardened choices into objects, AI keeps those choices fluid, translating uncertainty into adaptability. In doing so, it postpones the moment when purpose must fully crystallize. What looks like endless evolution is not a failure to converge, but a different kind of design altogether: one that holds motion instead of resolving it.</p><p>This leaves us with a question the essay has so far deferred: <em>Should we be trying to force AI toward peak form at all?</em> If the analysis here is correct — if AI’s perpetual evolution emerges from legitimate structural conditions rather than mere engineering immaturity — then the standard narratives around “AI alignment” or “responsible AI” may be addressing the wrong problem. You cannot align that which has no stable endpoint. You cannot make responsible what is designed to absorb irresponsibility (in the sense of deferred decision-making).</p><p>Perhaps the real question is not “when will AI finish evolving?” but “what would it mean to live well with intelligence tools that, by design, never finish?” Do we need AI governance frameworks that assume stability, or ones that govern perpetual motion? Do we need to fragment AI’s purpose deliberately — restricting general systems while allowing narrow ones to settle — or accept that we’ve created something categorically different from every technology that came before?</p><p>The safety pin converged because its purpose was clear. The plate converged because of purpose and because human scale was fixed. AI diverges because we’ve made it the repository for every question we can’t answer about ourselves. That’s not a failure of design. That’s a choice — possibly an unconscious one, but a choice nonetheless. And like all design choices, it can be revisited.</p><p>Finished design doesn’t tell us what AI should become. It tells us what we haven’t decided. We haven’t agreed whether AI is meant to close tasks or extend inquiry, to optimize outcomes or explore meaning. As long as that question stays unresolved, stability is impossible.</p><p>Perhaps the next great disruption in design won’t be more capability, but more constraint. The true luxury product of the future might be <strong>‘Finite AI’</strong> — a model that does exactly one thing perfectly, runs locally on your device, never connects to the cloud, and never updates. A tool that doesn’t ask you what it can be, but simply tells you what it is. Just like a shark.</p><h3>References (expanded &amp; updated)</h3><h4>Design, Purpose, and Convergence</h4><ul><li><strong>Petroski, H.</strong> <em>The Evolution of Useful Things</em>. Vintage, 1994.</li><li><strong>Norman, D.</strong> <em>The Design of Everyday Things</em>. Basic Books, 2013.</li><li><strong>Alexander, C.</strong> <em>Notes on the Synthesis of Form</em>. Harvard University Press, 1964.</li><li><strong>Basalla, G.</strong> <em>The Evolution of Technology</em>. Cambridge University Press, 1988.</li><li><strong>Arthur, W. B.</strong> <em>The Nature of Technology: What It Is and How It Evolves</em>. Free Press, 2009.</li><li><strong>Margolin, V.</strong> <em>The Politics of the Artificial</em>. University of Chicago Press, 2002.</li></ul><h4>Biology, Evolution, and Stability</h4><ul><li><strong>Gould, S. J.</strong> <em>Full House: The Spread of Excellence from Plato to Darwin</em>. Three Rivers Press, 1997.</li><li><strong>Carroll, S. B.</strong> <em>Endless Forms Most Beautiful: The New Science of Evo Devo</em>. Norton, 2005.</li><li><strong>Kauffman, S.</strong> <em>At Home in the Universe: The Search for the Laws of Self-Organization and Complexity</em>. Oxford University Press, 1995.</li><li><strong>Maynard Smith, J.</strong> <em>Evolutionary Genetics</em>. Oxford University Press, 1998.</li><li><strong>Dawkins, R.</strong> <em>The Blind Watchmaker</em>. W. W. Norton, 1986. <em>(for non-teleological adaptation framing)</em></li></ul><h4>Embodiment, Cognition, and Human Scale</h4><ul><li><strong>Clark, A.</strong> <em>Being There: Putting Brain, Body, and World Together Again</em>. MIT Press, 1997.</li><li><strong>Clark, A. &amp; Chalmers, D.</strong> “The Extended Mind.” <em>Analysis</em>, 1998.</li><li><strong>Lakoff, G. &amp; Johnson, M.</strong> <em>Philosophy in the Flesh</em>. Basic Books, 1999.</li><li><strong>Shapiro, L.</strong> <em>Embodied Cognition</em>. Routledge, 2011.</li></ul><h4>AI, Intelligence, and Open-Ended Systems</h4><ul><li><strong>Dennett, D.</strong> <em>From Bacteria to Bach and Back: The Evolution of Minds</em>. W. W. Norton, 2017.</li><li><strong>Russell, S.</strong> <em>Human Compatible</em>. Viking, 2019.</li><li><strong>Harari, Y. N.</strong> <em>Homo Deus: A Brief History of Tomorrow</em>. Harper, 2017.</li><li><strong>Floridi, L.</strong> <em>The Ethics of Artificial Intelligence</em>. Oxford University Press, 2023.</li><li><strong>Sutton, R.</strong> <em>The Bitter Lesson</em> (essay), 2019.</li><li><strong>Mitchell, M.</strong> <em>Artificial Intelligence: A Guide for Thinking Humans</em>. Farrar, Straus and Giroux, 2019.</li></ul><h4>Capitalism, Consumption, and “Never Enough”</h4><ul><li><strong>Schumpeter, J. A.</strong> <em>Capitalism, Socialism and Democracy</em>. Harper, 1942.</li><li><strong>Veblen, T.</strong> <em>The Theory of the Leisure Class</em>. Macmillan, 1899.</li><li><strong>Bauman, Z.</strong> <em>Consuming Life</em>. Polity Press, 2007.</li><li><strong>Graeber, D.</strong> <em>Bullshit Jobs</em>. Simon &amp; Schuster, 2018.</li><li><strong>Srnicek, N.</strong> <em>Platform Capitalism</em>. Polity Press, 2016.</li></ul><h4>East Germany, Durability, and Material Culture</h4><ul><li><strong>Brepohl, E.</strong> <em>Technikgeschichte der DDR: Glasindustrie und Materialinnovation</em>. Verlag Technik, Berlin, 1988.</li><li><strong>Höpner, M.</strong> <em>The Political Economy of German Reunification</em>. Max Planck Institute, 2001.</li><li><strong>Röhl, K.-H.</strong> “Why East German Products Failed After Reunification.” <em>CESifo Forum</em>, 2010.</li><li><strong>Corning Incorporated.</strong> “A Brief History of Chemically Strengthened Glass.” Technical white paper.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=82669cdf5c2e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Wave-Pill and the Algorithmic Mind]]></title>
            <link>https://medium.com/@RabihIbrahim/the-wave-pill-and-the-algorithmic-mind-fe1371fb068a?source=rss-845e11680e3b------2</link>
            <guid isPermaLink="false">https://medium.com/p/fe1371fb068a</guid>
            <category><![CDATA[future-of-intelligence]]></category>
            <category><![CDATA[mental-health]]></category>
            <category><![CDATA[radical-reflections]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[neurotechnology]]></category>
            <dc:creator><![CDATA[Rabih Ibrahim]]></dc:creator>
            <pubDate>Wed, 17 Dec 2025 12:58:58 GMT</pubDate>
            <atom:updated>2025-12-17T12:58:58.362Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aT-ubk0jJT_l-othilzZBg.png" /></figure><h4>Found &amp; Framed Series</h4><p><em>How AI, brain stimulation, and feedback loops are turning depression treatment into a programmable system — and raising difficult questions about control</em></p><p>For decades, depression was a long game. Medications could take weeks to work, if they worked at all. Psychotherapy required patience, months of unraveling thought patterns and emotional habits. Even electroconvulsive therapy — one of the most effective options for severe depression — typically unfolded over multiple sessions across weeks. The rhythm of recovery was slow, often painfully so, and for many patients the waiting itself became part of the suffering.</p><p>A new approach is now collapsing that timeline. In trials conducted at Stanford, people with treatment-resistant depression — patients who had already tried multiple medications, psychotherapy, and in some cases hospitalization — saw their symptoms reverse in just five days. The protocol, known as <strong>Stanford Accelerated Intelligent Neuromodulation Therapy (SAINT)</strong>, uses transcranial magnetic stimulation delivered ten times a day over the course of a week. Nearly half of participants entered full remission, and in some studies the remission rate approached 80%.</p><p>(Stanford Medicine overview: <a href="https://med.stanford.edu/news/all-news/2021/10/depression-treatment.html?utm_source=chatgpt.com">https://med.stanford.edu/news/all-news/2021/10/depression-treatment.html</a>)</p><p>If these results continue to hold at scale, they represent more than another treatment option. They force a reframing of how we understand the brain itself. Instead of altering mood slowly through chemistry, we may be able to influence it directly through rhythm, timing, and targeted electrical patterns. The idea borders on the surreal: replacing a nightly Xanax not with another pill, but with a prescription of waves. A <em>wave-pill</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SaQfa4x7B1N65vAA20tCrw.png" /></figure><h3>From Chemical Imbalance to Circuit Imbalance</h3><p>For years, depression was explained to the public almost exclusively through the language of serotonin deficits. The “chemical imbalance” model became a cultural shorthand: low serotonin equals low mood. While never scientifically complete, it shaped how depression was talked about, treated, and internalized.</p><p>That story is now being revised. Depression is increasingly understood as a disorder of <strong>circuits</strong>, not just chemicals. Brain imaging studies consistently show that the <strong>dorsolateral prefrontal cortex (DLPFC)</strong> — a region associated with planning, cognitive control, and perspective-taking — is often underactive in depression. At the same time, the <strong>subgenual anterior cingulate cortex (sgACC)</strong> — linked to sadness, rumination, and emotional pain — tends to be overactive.</p><p>When the DLPFC fails to regulate the sgACC effectively, negative thought loops take hold. Rumination intensifies not because serotonin is low, but because a regulatory circuit designed to maintain balance has slipped into dysfunction.</p><p>Helen Mayberg, a neurologist and pioneer in deep brain stimulation research, captured this shift succinctly when she observed: <em>“Depression is not a chemical disorder. It’s a systems disorder.”</em> The SAINT protocol takes this idea seriously. Rather than adding chemicals to the system, it intervenes directly at the level of circuitry, stimulating the networks that have fallen silent and restoring their regulatory role.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FE33b-Yu70W0aQjS_-ZexA.png" /></figure><h3>How Stimulation Works</h3><p>The mechanics of transcranial magnetic stimulation sound deceptively simple. A magnetic coil placed against the scalp delivers pulses that pass through the skull without surgery. These pulses excite neurons in targeted cortical regions, most commonly the DLPFC. Because the DLPFC is tightly connected to deeper mood-regulation structures such as the sgACC, increasing activity in one can dampen pathological activity in the other.</p><p>What distinguishes accelerated TMS protocols like SAINT is not just where stimulation occurs, but how it unfolds over time. Instead of a single daily session stretched across six weeks, SAINT delivers ten sessions per day for five days — roughly fifty doses compressed into a single week. The pulses follow patterns designed to mimic the brain’s natural theta rhythms, which appear to enhance neural receptivity and trigger neuroplasticity.</p><p>As Nolan Williams, the psychiatrist who led the Stanford study, put it: <em>“It’s like hitting reset on the most stubborn patterns in the brain.”</em></p><p>A metaphor helps clarify what this means. Depression resembles a record player stuck in a groove. Traditional therapy and medication may gradually shift the needle, often over weeks or months. Accelerated TMS works differently, repeatedly nudging the arm in rapid succession until it relearns how to fall into new grooves altogether.</p><p>These changes are not merely subjective. Functional MRI scans taken before and after treatment show measurable changes in connectivity between the DLPFC and sgACC. In a very literal sense, circuits are being rewired.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aDhZ8p17HyU8ImIuwL04xg.png" /></figure><h3>The Wave-Pill</h3><p>This brings us back to the idea of the wave-pill. For most of modern psychiatry, pills were the primary interface through which the brain could be altered from the outside. Prozac, Zoloft, Xanax — each designed to modify neurotransmitter levels systemically, often with delayed and diffuse effects.</p><p>But what if tomorrow’s prescriptions were not swallowed, but <em>administered as patterns</em>?</p><p>Instead of a 10-milligram dose, a patient might receive a protocol: thirty minutes of patterned stimulation in a clinic chair, calibrated to restore balance not through serotonin levels, but through neural rhythm and timing. Depression, in this framing, becomes less like a deficiency to be corrected and more like an instrument to be retuned.</p><p>The shift is subtle but profound. We move from chemistry to circuitry, from substances to signals.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BvPznFMsKB0fCPU3ZvO0iA.png" /></figure><h3>Three Routes to Relief</h3><p>It is important to note that TMS is not the only rapid-acting intervention reshaping depression treatment. Other approaches reveal different ways into the same plastic system.</p><p><strong>Electroconvulsive Therapy (ECT)</strong> induces a controlled seizure, effectively rebooting large-scale brain networks. It is invasive and requires anesthesia, but remains one of the most effective treatments for severe and psychotic depression, with remission rates between 50 and 70 percent.</p><p><strong>Ketamine</strong> operates chemically, blocking NMDA receptors and triggering a surge of glutamate that promotes rapid synaptic growth. Relief can emerge within hours, particularly for suicidal ideation, though ongoing maintenance treatments are often required.</p><p><strong>Transcranial Magnetic Stimulation (TMS)</strong> works with precision, using externally applied magnetic fields to stimulate specific cortical regions, allowing one circuit to rebalance another without global disruption.</p><p>As neuroscientist Karl Deisseroth has written, <em>“There are many entry points into the same plastic system.”</em> What differs between these approaches is not whether the brain can change, but the scale and specificity with which that change is induced.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qMoW6ezvNXMxbiFFH7ZxvA.png" /></figure><h3>Beyond Fixing: Toward a Holistic Understanding</h3><p>What makes these developments remarkable is not only their speed, but the perspective they offer on what it means to be human. If mood can shift in days through targeted stimulation, depression cannot be reduced to willpower, personality, or even chemistry alone. It is a property of circuits — and those circuits are malleable.</p><p>This does not mean we are machines. It means we are layered systems. Chemical, electrical, embodied, narrative. Pills address chemistry. Waves influence circuitry. Therapy reshapes meaning and self-story. Meditation alters attention and awareness. Each layer matters, and none alone completes the picture.</p><p>Antonio Damasio captured this layered reality when he wrote: <em>“The self is not a thing, but a process.”</em> Treatments like TMS remind us that this process has multiple points of entry, and that understanding those layers together may be the key to understanding ourselves more fully.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O8hqhf6rxNu7LoNWFsLAxQ.png" /></figure><h3>The Algorithmic Turn in Circuit Design</h3><p>If depression can dissolve in days, the central question is no longer whether the brain can be influenced, but how precisely and how adaptively that influence can be applied.</p><p>The protocols used today are still relatively static. Targets are predefined. Pulse patterns are standardized. Adjustments are made by clinicians based on observation and experience. Yet nothing in principle prevents these systems from becoming adaptive, continuously learning from the individual they are affecting.</p><p>Imagine a closed-loop system in which brain stimulation is paired with real-time neural data, physiological signals from the body, and subjective feedback from the patient. Brain activity, heart-rate variability, sleep patterns, and reported emotional states could feed into an AI model trained to recognize each person’s unique neural dynamics. Stimulation parameters would no longer be fixed but recalculated continuously, tuned to the person’s current state rather than a generalized protocol.</p><p>This is not science fiction. Closed-loop neuromodulation already exists in conditions such as Parkinson’s disease and epilepsy, where implanted devices detect pathological activity and respond in real time. What is changing now is accessibility and intelligence. Non-invasive stimulation combined with machine learning brings these ideas into psychiatry without surgery, and potentially into everyday clinical settings.</p><p>Once stimulation becomes adaptive, however, another question emerges: purpose. Is the goal symptom relief, emotional stability, enhanced focus, or resilience under stress? Who defines the target state — the clinician, the patient, the algorithm, or the institution deploying the system?</p><p>At this point, the distinction between treatment and influence becomes less clear. If circuits can be tuned to relieve suffering, they can also be tuned to shape behavior. Anxiety could be dampened before high-stakes decisions. Emotional volatility softened. Motivation increased. These interventions would not override agency in any dramatic sense, but they could subtly bias emotional baselines and response tendencies.</p><p>The concern here is not domination in a crude, dystopian sense. It is calibration. And when calibration is mediated by intelligent systems operating at scale, it becomes a form of power that is quiet, precise, and difficult to see.</p><p>This is where the wave-pill stops being a medical innovation and becomes a philosophical challenge. As AI systems grow more capable of modeling and predicting internal states, the boundary between self-regulation and external regulation grows thinner. Consent becomes more complex when systems adapt faster than conscious reflection. Agency becomes harder to define when emotional baselines are continuously tuned.</p><p>Seen this way, the wave-pill is not merely a new antidepressant. It is a prototype for a new relationship between humans and intelligent systems, one in which biology, computation, and subjective experience form a single feedback loop. Once that loop exists, the question is no longer whether it will shape us, but how consciously and collectively we decide to shape it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*a2zBQpNKiqwWjPVGUj86MA.png" /></figure><p>The idea of a wave-pill is both comforting and unsettling. Comforting, because it promises relief without the fog of chemicals. Unsettling, because it reveals how pliable our inner lives may be, open to adjustment in ways we once believed were beyond reach.</p><p>If years of despair can be undone in days, perhaps the most human quality we possess is not our suffering or even our resilience, but our plasticity — the capacity to change, and to be changed. The challenge ahead is learning how to live with that knowledge without surrendering the responsibility that comes with it.</p><h3>📚 Resources</h3><ul><li>Stanford Medicine — <em>Stanford Accelerated Intelligent Neuromodulation Therapy (SAINT)</em><br> <a href="https://med.stanford.edu/news/all-news/2021/10/depression-treatment.html?utm_source=chatgpt.com">https://med.stanford.edu/news/all-news/2021/10/depression-treatment.html</a></li><li>Nature Translational Psychiatry — <em>Double-blind trial of SAINT for treatment-resistant depression</em></li><li>Helen Mayberg on brain circuits and depression — <em>Nature</em></li><li>Karl Deisseroth, <em>Projections: A Story of Human Emotions</em></li><li>Antonio Damasio, <em>The Feeling of What Happens</em></li><li>Widge et al., <em>Closing the Loop on Deep Brain Stimulation</em> — <em>Nature Neuroscience</em></li><li>Yuste et al., <em>Four Ethical Priorities for Neurotechnologies and AI</em> — <em>Nature</em></li><li>NIH BRAIN Initiative — Closed-Loop Neuromodulation Programs</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fe1371fb068a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Does Your GenAI Need to Sign an NDA?]]></title>
            <link>https://medium.com/@RabihIbrahim/does-your-genai-need-to-sign-an-nda-c8193f86e048?source=rss-845e11680e3b------2</link>
            <guid isPermaLink="false">https://medium.com/p/c8193f86e048</guid>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[radical-reflections]]></category>
            <category><![CDATA[design-ethics]]></category>
            <category><![CDATA[ai-safety]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Rabih Ibrahim]]></dc:creator>
            <pubDate>Wed, 10 Dec 2025 12:05:26 GMT</pubDate>
            <atom:updated>2025-12-10T12:05:26.947Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S-6eGHete0guP7S0n7g8HQ.png" /></figure><h4>Thought Exploration Series</h4><p><em>We’ve built machines that can read our secrets. But can they keep them?</em></p><h3>When Your Thinking Isn’t Private</h3><p>Imagine this: you’re preparing an investor deck, a new strategy, or even a deeply personal reflection. Normally, that kind of conversation happens in a sealed room. Everyone present signs an NDA. Everyone agrees to walk away without carrying the material anywhere else.</p><p>But now, instead of a colleague or consultant, you invite a generative AI into the room. You type your strategy into the chat box, you upload the confidential notes, you ask it to brainstorm solutions. The model doesn’t pause. It doesn’t sign anything. It simply absorbs, processes, and responds.</p><p>The scene feels absurd — as though you’ve just slid a document across the table to a statistical engine and expected it to respect confidentiality. Yet this is what millions of us do daily: exposing the most guarded parts of our work and thought to an entity that cannot make promises, cannot forget, and cannot be held accountable in the way a human can.</p><p>The metaphor writes itself: does your GenAI need to sign an NDA?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_TRDyZMfyTDhFRJ1Q4yoYA.png" /></figure><h3>The Illusion of Confidentiality</h3><p>NDAs were designed to bind people and companies. They create obligations, penalties, and a legal remedy if trust is broken. They don’t guarantee absolute secrecy — but they create accountability.</p><p>GenAI changes the terrain. When you share confidential information with a model, you’re not whispering in the ear of a trusted colleague. You’re placing it into a probabilistic system where memory doesn’t decay. Even if the vendor promises “we don’t train on your data,” your disclosure passes through retention logs, debugging systems, error reports, and sometimes opaque subcontractors.</p><p>As one security expert put it:</p><blockquote>“A contract is only as strong as the party that can be sued. What happens when the ‘party’ is statistical math?”</blockquote><p>The illusion of confidentiality is seductive. We treat the AI as though it were a neutral notepad — but it is more like a sponge, and sponges never un-soak.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KkvZJ8YaPhZ1yaDGOYvXog.png" /></figure><h3>Machines Don’t Forget</h3><p>Human memory is imperfect. Colleagues can leak secrets, but they can also forget them. Forgetting is our default.</p><p>Models, on the other hand, are designed for retention. They store training data in distributed patterns of weight adjustments. Even when data is not “directly” memorized, traces and patterns can remain. Add to that the surrounding infrastructure — logs, cache layers, monitoring systems — and the result is a form of digital persistence.</p><p>A 2023 study from Cornell noted that models can “regurgitate” specific training data with surprisingly little prompting, raising new concerns about whether confidential information, once seen, can ever truly be erased. Another paper from Stanford described AI models as “sticky” — once an input is encountered, elements can reappear in unexpected places.</p><p>The difference matters. With humans, forgetting is natural and remembering is effortful. With machines, remembering is the baseline, and forgetting requires deliberate, engineered intervention.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3qTq8aMD4FOWDUZQFK1Naw.png" /></figure><h3>The Need to Forget</h3><p>If generative AI is to act as a thinking partner, it might need to learn how to forget. Or rather, we must design the systems around it to guarantee forgetting if need be. And we need to think carefully about <em>how</em> we design that. I don’t imagine forgetting as an on/off switch, but more like a time-related dimmer: information should remain available long enough to be useful, yet fade unless we explicitly instruct the model to retain it. Digital memory defaults to permanence; forgetting requires an act of design, not neglect.</p><p>This isn’t just about deletion of logs or setting retention windows. True forgetting would mean the ability to remove the influence of confidential inputs from the model itself — from weights, embeddings, and derivative outputs. That is not trivial. Current techniques for machine “unlearning” are still in their infancy. For example, early research on <em>selective weight erasure</em> shows that removing the influence of a single training example often requires retraining parts of the model, because traces of that example diffuse through thousands of parameters. Even so-called “data deletion requests” don’t remove the learned pattern itself — only the visible record of it.</p><p>Here the legal metaphor collides with the technical: should we begin to demand a <strong>“right to be forgotten” for AI inputs</strong>? Just as GDPR gave individuals the ability to request erasure from human-readable databases, perhaps the next frontier is erasure from machine-learned representations.</p><p>Philosopher Luciano Floridi has argued that forgetting is part of what makes human identity sustainable — it keeps the past from dominating the present. Without forgetting, there is no forgiveness, no renewal, no freedom. If that’s true, then the absence of forgetting in our machine partners is more than a technical quirk; it is a cultural risk.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IaCzAuM0P7Mdbshz5ZWkXw.png" /></figure><h3>Contracts With Math</h3><p>If an NDA works because we can hold someone liable, who do we hold liable here? The vendor? The hosting cloud provider? The engineers who trained the model? Or the abstract system of code itself?</p><p>The question points to a gap. Our legal frameworks were built for entities that can be identified, punished, or sued. <strong>Statistical models are not legal persons</strong>, and perhaps that is part of why we trust them so easily: we don’t instinctively imagine them betraying us in the way a human might. But that assumption is misplaced. The harm doesn’t come from intention — it comes from persistence, replication, and scale.They are tools — but tools that operate at such a scale and opacity that accountability is diffused to the point of near-meaninglessness.</p><p>We end up in a strange position: the metaphor of the NDA makes sense emotionally — we want the model to keep our secrets — but it breaks down in practice because there is no entity to sign, no consciousness to bind, no hand to hold the pen.</p><p>What remains is the vendor, and the vendor’s terms of service. In some enterprise contracts, you will now find “no training on customer data” clauses. That’s the legal system trying to retrofit the NDA model into the age of GenAI. But notice how thin the protection is compared to a real NDA with a human counterpart: it rests on trust in the vendor’s opaque pipelines, not on the enforceable discretion of a person.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QjWg0V-3jV_oUvLO5o7zxQ.png" /></figure><h3>When Our Thinking Becomes Shared</h3><p>There’s another layer: the kind of data we now consider confidential has shifted. It is no longer only product specs, trade secrets, or customer lists. It’s our thought process itself.</p><p>When you use GenAI as a brainstorming partner, you are externalizing your half-formed ideas, your unpolished reasoning, your hidden assumptions. In another era, those drafts would live in a private notebook. Today, they live in the cloud.</p><p>This creates a new category of risk: the confidentiality of thought itself. If our ideas are shaped in dialogue with a machine that cannot forget, then our own cognition is no longer private. The NDA metaphor thus points beyond law, into philosophy: what does it mean for thinking to be a shared, recorded, and potentially retrievable event — and how much of that thinking still belongs to us?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jFacG4TxVeSp9UYsDuvuYA.png" /></figure><h3>Rethinking Trust</h3><p>So, does your GenAI need to sign an NDA? On the surface, the answer is obvious: no. But the fact that we even ask the question reveals something deeper: our old rituals of trust no longer map cleanly onto our new tools.</p><p>Perhaps what we need is not NDAs for machines, but <strong>technical NDAs</strong> — systems that guarantee the spirit of confidentiality through architecture:</p><ul><li>Automatic forgetting after defined windows.</li><li>Strong prohibitions on training with customer data.</li><li>Transparent deletion reports.</li><li>Audit rights that allow us to verify.</li></ul><p>In other words, a fusion of law and engineering that restores the human comfort of secrecy in a world where machines remember too well.</p><p>The NDA metaphor is useful precisely because it fails. It highlights that we are trying to hold onto human frameworks in an environment where they no longer fit. If contracts once protected our secrets, what new rituals, safeguards, and designs will we need now?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qVYsQcTiRmARd9XrE4R50Q.png" /></figure><h3>Closing Reflection</h3><p>We built generative AI to help us think. In doing so, we invited it into our most private rooms — the spaces of problem-solving, imagination, and strategic foresight. But unlike a colleague or a consultant, the machine cannot give its word.</p><p>The deeper question isn’t whether AI can promise discretion, but what secrecy means in a digital environment where memory is cheap, copying is effortless, and erasure is never guaranteed. Human forgetting is imperfect, but it is also protective — it lets certain moments fade, soften, or disappear altogether. Digital systems offer the opposite: they can preserve a version of our thoughts long after we have moved on from them. In a world where even our private reasoning can be stored, indexed, or resurfaced, we may need to rethink the relationship between memory and trust — and imagine new forms of forgetting that allow our digital partners to release what we no longer want them to hold.</p><h3>References &amp; Resources</h3><ul><li>Luciano Floridi, <em>The Ethics of Information</em> (Oxford, 2013).</li><li>Cornell University (2023), “Extracting Training Data from Large Language Models.”</li><li>Stanford University HAI (2023), “Sticky Data: Retention and Memorization in Generative Models.”</li><li>European Union GDPR, Article 17: “Right to be Forgotten.”</li><li>Zuboff, Shoshana. <em>The Age of Surveillance Capitalism</em> (2019).</li><li>OpenAI Enterprise Terms of Use, “No training on customer data” clauses (2023).</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c8193f86e048" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The DNA Cassette Tape: Life’s Code as Humanity’s Storage Medium]]></title>
            <link>https://medium.com/@RabihIbrahim/the-dna-cassette-tape-lifes-code-as-humanity-s-storage-medium-6133e55525f4?source=rss-845e11680e3b------2</link>
            <guid isPermaLink="false">https://medium.com/p/6133e55525f4</guid>
            <category><![CDATA[radical-reflections]]></category>
            <category><![CDATA[synthetic-data]]></category>
            <category><![CDATA[humans-and-machines]]></category>
            <category><![CDATA[synthetic-biology]]></category>
            <category><![CDATA[philosophy-of-science]]></category>
            <dc:creator><![CDATA[Rabih Ibrahim]]></dc:creator>
            <pubDate>Wed, 12 Nov 2025 11:39:27 GMT</pubDate>
            <atom:updated>2025-11-12T11:39:27.281Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*usxLBIbffB-Q0FF21yQgcA.png" /></figure><h4>Thought Exploration Series</h4><h4>What began as biology’s alphabet is fast becoming humanity’s next great medium for memory, meaning, and creation.</h4><h3><strong>The Cassette Tape of Life</strong></h3><p>For billions of years, DNA has been the ultimate storage medium. Every living organism carries its blueprint in the same four-letter alphabet of A, T, C, and G. That code has written the diversity of life on Earth, from the earliest microbes to us. What has changed in the last few decades is that humans have stopped looking at DNA only as biology’s script and started to see it as something else: an almost unimaginably powerful information technology.<br> To grasp the scale, consider this: a single gram of DNA can hold an estimated 215 petabytes of data — about 215 million gigabytes. All of Spotify’s catalog, every track in every format, would barely scratch the surface of that capacity. The entire history of recorded sound could fit easily inside a vial of molecules you could balance on your fingertip. This is why some researchers describe DNA as a cassette tape for the twenty-first century — not because it’s retro, but because it encodes information sequentially and can be “played back” when needed. Where a plastic spool once held a few dozen songs, a synthetic strand could hold the sonic memory of humanity. The bottlenecks today aren’t capacity but write/read speed, cost, and random access: synthesizing DNA is slow and expensive, sequencing is faster but not instant, and jumping to a specific “track” requires clever indexing and error-correction codes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dD3Bl5d8xKxvBSsO9S4D_A.png" /></figure><h3><strong>Synthetic DNA: Real, But Not “Alive”</strong></h3><p>When people hear “DNA storage,” it can sound unsettling, as if we’re borrowing pieces of life to hoard our data. But the DNA used in these systems is synthetic — written in the lab rather than copied from a living cell. Chemically, it’s indistinguishable from natural DNA; what differs is origin and intent. Natural DNA evolved to carry instructions cells can interpret. Synthetic DNA can be written for human purposes, sometimes aligned with biology and sometimes entirely outside it.<br> Synthetic DNA is not “alive.” That means:<br> It carries no genes or regulatory signals a cell could act upon (unless we intentionally add them).<br> It cannot on its own replicate, mutate, or evolve; it’s inert material until a machine or a cell copies it.</p><p>With that in view, the same alphabet (A/T/C/G) shows two faces depending on how we write it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*STKLr8jovt6ijtgbWcH2ow.png" /></figure><h3><strong>Two Faces of Synthetic DNA</strong></h3><p><strong>1. DNA as Storage (the “Cassette Tape”)</strong><br> Here the goal is to encode digital information — books, films, archives, even entire music libraries — into strands of DNA. The sequences are deliberately engineered to avoid looking like genes, so they cannot be mistaken for biological instructions. To a living cell, these strands are gibberish; to us, with the right decoder, they’re data. Because DNA is so dense and — stored cool and dry — so long-lived, it’s an attractive archival medium: a gram can, in theory, hold centuries of human culture. The practical challenges today are engineering ones: writing cost and latency, robust indexing for random access, and error-correction to survive synthesis and sequencing noise. <strong>Concretely, “writing” means chemically assembling billions of bases one by one, which keeps throughput low and the price per megabyte high. Indexing assigns each short strand something like a library call number so you can retrieve the right fragments later, while error-correcting codes let software perfectly rebuild the file even if some strands are missing or mutated.</strong> But the principle is settled: DNA can act as a molecular memory that outlasts hard drives and tape libraries by orders of magnitude.</p><p><strong>2. DNA as Biological Code (the “Living Factory”)</strong><br> The second face is very different. Here, synthetic DNA is written in the grammar of life: a promoter to start transcription, a coding region that specifies a protein, and a stop signal to end it. Inserted into bacteria or yeast — often on a plasmid or integrated into the genome — this DNA becomes functionally indistinguishable from the organism’s own. The canonical example is insulin. Scientists synthesized the human insulin gene and placed it inside E. coli. The bacteria didn’t care that the DNA was designed on a computer last week; they recognized valid instructions and produced insulin as though it were their own protein. That’s why much of the world’s insulin supply comes from engineered microbes. Biology doesn’t check a provenance label; it checks whether the sequence makes sense.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HfjCnu0B7NvW0ssll5rzzg.png" /></figure><h3><strong>Why Synthetic DNA Exists</strong></h3><p>The motivations for creating and using synthetic DNA are layered, and each reveals a distinct human ambition.<br> <strong>Science.</strong> Synthetic DNA freed researchers from the slow, messy work of extracting and cloning genes from organisms. Today, a lab can order precise sequences on demand and test hypotheses with far greater speed and control. That shift turned DNA from a scarce biological resource into an editable text we can print, read, and revise.<br> <strong>Medicine.</strong> Therapeutics and vaccines increasingly depend on custom-designed DNA. Rather than searching nature for a protein, we can write the sequence and express it in a safe host. Insulin is only the beginning: monoclonal antibodies, gene-therapy vectors, CAR-T constructs, and even standardized diagnostic controls are shaped by synthetic sequences. The pipeline from design to molecule is shorter, clearer, and more scalable.<br> <strong>Biotechnology.</strong> Industry quickly followed. Yeast programmed with synthetic DNA brew biofuels, generate collagen-like proteins for “synthetic leather,” and spin spider-silk analogs. Crops gain traits for pest resistance or drought tolerance via precisely authored edits and inserts. In this domain, DNA is a process instruction for living factories, turning cells into programmable bioreactors.<br> <strong>Information Storage.</strong> Outside biology, engineers saw in DNA a near-perfect archival medium. Its density and durability far surpass magnetic tape or flash. Entire libraries — songs, films, books, code — could one day rest in DNA vaults, retrievable with sequencing rather than tape drives. The economics aren’t there yet for everyday use, but for century-scale archives and cold storage, DNA points toward a radically compact, future-proof medium.<br> In truth, synthetic DNA exists not because we sought control over life’s alphabet, but because <strong>we discovered we could make it</strong> — and now we’re still mapping its boundaries, risks, and possibilities.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EenGszec1vkgkklgMuYaRA.png" /></figure><h3><strong>The Deeper Anxiety</strong></h3><p>What unsettles many people is not the molecule but what it represents. DNA was once the sacred language of life, written only by nature. Now it’s something we can order, print, and design at will. And the most disquieting fact is this: cells don’t distinguish between natural and synthetic DNA. If the sequence makes sense in their grammar, they act. There is no boundary for them between a gene evolved over eons and one composed last week.<br> That realization collapses the line we like to draw between natural and artificial. It raises the uncomfortable possibility that the distinction exists mostly in our imagination. When we dismiss synthetic DNA as “not real,” are we expressing a kind of prejudice against our own creations? If biology itself refuses to recognize the difference, perhaps our unease isn’t about DNA at all — but about what it means to share authorship of life’s code. As George Church, a pioneer in this field, once put it:<br> “We now have the ability to write the software of life.”<br> It’s an extraordinary power, and an unsettling one, because once you can write the code, you’re no longer only a reader of life’s story. You become part of its continuation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x7n9ocfot7rr1wh8qpoS9A.png" /></figure><h3><strong>The Mirror of Creation</strong></h3><p>I can’t help but wonder whether our quest to build “intelligent machines” is, at its core, an attempt to rediscover ourselves. What if artificial intelligence, synthetic DNA, and neural computation are not foreign to life, but extensions of it? Perhaps we are not creating a new kind of intelligence at all — but reawakening one that has always been latent within the organic world.<br> The more we understand how DNA stores, edits, and retrieves information, the harder it becomes to separate the biological from the digital. Our machines are made of silicon and code; our bodies, of carbon and cells. Yet both obey the same deep logic: <strong>information seeking structure, and structure expressing information.</strong> The organic has always had digital constituents behind it.<br> As the physicist John Archibald Wheeler put it, <em>“It from bit”</em> — the idea that every physical entity ultimately derives its meaning from binary yes-or-no distinctions. In that sense, creation itself might be the first computation. Alan Turing’s theoretical machine and the double helix may not be opposites, but reflections of one another — two ways the universe learned to remember, process, and evolve thought.<br> Maybe we are the smart machine we’ve been trying to build all along.</p><h3><strong>References</strong></h3><ul><li>Church, G. M., Gao, Y., &amp; Kosuri, S. (2012). <em>Next-generation digital information storage in DNA.</em> <strong>Science</strong>, 337(6102), 1628.</li><li>Goldman, N., Bertone, P., Chen, S., Dessimoz, C., LeProust, E. M., Sipos, B., &amp; Birney, E. (2013). <em>Towards practical, high-capacity, low-maintenance information storage in synthesized DNA.</em> <strong>Nature</strong>, 494(7435), 77–80.</li><li>Erlich, Y., &amp; Zielinski, D. (2017). <em>DNA Fountain enables a robust and efficient storage architecture.</em> <strong>Science</strong>, 355(6328), 950–954.</li><li>Organick, L., Ang, S. D., Chen, Y.-J., et al. (2018). <em>Random access in large-scale DNA data storage.</em> <strong>Nature Biotechnology</strong>, 36, 242–248.</li><li>Takahashi, C. N., Nguyen, B. H., Strauss, K., &amp; Ceze, L. (2019). <em>Demonstration of end-to-end automation of DNA data storage.</em> <strong>Scientific Reports</strong>, 9(1), 4998.</li><li>Bornholt, J., Lopez, R., Carmean, D. M., et al. (2016). <em>A DNA-based archival storage system.</em> <strong>ASPLOS ’16: International Conference on Architectural Support for Programming Languages and Operating Systems</strong>, 637–649.</li><li>Shipman, S. L., Nivala, J., Macklis, J. D., &amp; Church, G. M. (2017). <em>CRISPR–Cas encoding of a digital movie into the genomes of a population of living bacteria.</em> <strong>Nature</strong>, 547(7663), 345–349.</li><li>National Human Genome Research Institute (NHGRI). <em>Synthetic Biology Fact Sheet.</em> (Updated 2022).</li><li>Endy, D. (2005). <em>Foundations for engineering biology.</em> <strong>Nature</strong>, 438(7067), 449–453.</li><li>Gibson, D. G., et al. (2010). <em>Creation of a bacterial cell controlled by a chemically synthesized genome.</em> <strong>Science</strong>, 329(5987), 52–56.</li><li>Strauss, K., &amp; Ceze, L. (2018). <em>DNA data storage: A primer.</em> <strong>Communications of the ACM</strong>, 61(11), 68–69.</li><li>Zhang, L., Zhao, Y., &amp; Kong, J. (2020). <em>Recent advances in DNA data storage: Methods, applications, and perspectives.</em> <strong>Frontiers in Bioengineering and Biotechnology</strong>, 8, 1001.</li><li>Kauffman, S. A. (2019). <em>A World Beyond Physics: The Emergence and Evolution of Life.</em> Oxford University Press.</li><li>Wheeler, J. A. (1990). <em>Information, Physics, Quantum: The Search for Links.</em> In <em>Complexity, Entropy, and the Physics of Information</em> (ed. W. Zurek). Addison-Wesley.</li><li>Davies, P. (2019). <em>The Demon in the Machine: How Hidden Webs of Information Are Solving the Mystery of Life.</em> University of Chicago Press.</li><li>Turing, A. M. (1950). <em>Computing Machinery and Intelligence.</em> <strong>Mind</strong>, 59(236), 433–460.</li><li>Morange, M. (2020). <em>The Information Metaphor and the Living World.</em> <strong>Biology &amp; Philosophy</strong>, 35(1), 1–15.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6133e55525f4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Algorithms That Run the World]]></title>
            <link>https://medium.com/@RabihIbrahim/algorithms-that-run-the-world-be9fda23e8d1?source=rss-845e11680e3b------2</link>
            <guid isPermaLink="false">https://medium.com/p/be9fda23e8d1</guid>
            <category><![CDATA[algorithms]]></category>
            <category><![CDATA[technology-and-society]]></category>
            <category><![CDATA[digital-governance]]></category>
            <category><![CDATA[radical-reflections]]></category>
            <category><![CDATA[future-of-systems]]></category>
            <dc:creator><![CDATA[Rabih Ibrahim]]></dc:creator>
            <pubDate>Wed, 05 Nov 2025 13:54:44 GMT</pubDate>
            <atom:updated>2025-11-05T13:54:44.817Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6OpMwPxCI-y9zvN5MYTDBQ.png" /></figure><h4>Found and Framed Series</h4><h4>The invisible constitution shaping how we think, move, and decide</h4><p>We like to think of human society as governed by laws, institutions, and markets. But beneath these visible structures hums another layer — one we rarely see, never vote on, and seldom understand. Algorithms.</p><p>They do not legislate in parliaments, yet they write the rules of search, finance, mobility, and identity. They do not campaign for office, yet they hold more power over daily life than most governments. Algorithms are not just tools we use. They are the invisible constitution of modern existence.</p><p>The question is not whether algorithms run the world. They do — at least partially. And we can see the evidence everywhere: in how markets rise and fall in milliseconds, in how routes shift our city flows, in how recommendations shape our taste, and in how credit scores decide our eligibility before we ever meet another human being. Every major system — from energy grids to information networks — now depends on algorithmic mediation. What once required deliberation and policy is now performed silently in code.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IfRdUfgndr5w7uudOrEjMA.png" /></figure><h3>The Infrastructure Beneath the Infrastructure</h3><p>In earlier centuries, progress was measured by the building of <em>physical</em> infrastructure — roads, railways, power grids, pipelines. These systems carried the lifeblood of civilization: goods, energy, and water. Today, the same logic operates in the abstract. Algorithms have become the new infrastructure — the invisible pipes through which decisions, data, and value flow.</p><p>Just as highways determine how cities expand, algorithms determine how attention moves. Just as power grids balance demand and supply, they regulate the flows of information and credit. If the 20th century was built on concrete and steel, the 21st is being built on systems that think — infrastructures that sense, calculate, and adjust themselves. They don’t replace our physical foundations; they run through them, automating decisions that once required human judgment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xac76GA187qI7_fT4kixUw.png" /></figure><h3>What This Overview Covers</h3><p>There are thousands of algorithms shaping the modern world — each with its own history, technical complexity, and consequence. The following sections are not exhaustive, but they trace the broad families that form the scaffolding of algorithmic civilization:</p><ol><li><strong>Knowledge &amp; Search</strong> — how we discover, retrieve, and now generate information.</li><li><strong>Markets &amp; Pricing</strong> — how economies move at machine speed.</li><li><strong>Movement &amp; Logistics</strong> — how optimization dictates flow and geography.</li><li><strong>Culture &amp; Taste</strong> — how recommendation systems shape identity and belief.</li><li><strong>Risk &amp; Permission</strong> — how credit, insurance, and predictive models gate access.</li><li><strong>Infrastructure &amp; Control</strong> — how invisible systems sustain trust and stability.</li><li><strong>Governance &amp; Surveillance</strong> — how power uses data to monitor, rank, and control.</li><li><strong>Generative &amp; Agentic Systems</strong> — how AI begins to act, decide, and create.</li></ol><p>Each could fill its own volume. What follows is an overview — a landscape view of the algorithms that, together, run the world.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PYfdttXCVWC8gtAtPah1oA.png" /></figure><h3>Part I: The Gatekeepers of Knowledge</h3><p>In the early web, knowledge was chaotic — billions of pages with no order. Google’s PageRank turned hyperlinks into votes. Each link became a signal of relevance, and suddenly the world’s knowledge could be sorted. It was messy, but it felt democratic: collective human behavior structured discovery.</p><p>Larry Page and Sergey Brin wrote in 1998: <em>“The importance of a page is an inherently subjective matter.”</em> Yet that “subjectivity” was frozen into a formula — one that would shape the modern information economy. PageRank rewarded connectivity and authority, creating a hierarchy of attention that soon became indistinguishable from truth itself.</p><p>Two decades later, generative AI systems have shifted the model entirely. When you type a question into ChatGPT or Google’s AI Overviews, you don’t get ranked links — you get synthesized answers. The algorithm no longer points; it interprets. Knowledge has become conversational, confident, and self-authoring.</p><p>This evolution marks a profound change: from <em>search</em> to <em>sense-making.</em> When the same system that retrieves information also interprets it, epistemology becomes infrastructure. The logic of knowledge itself — what counts as relevant, true, or complete — is written in code. I explored this shift in depth in <a href="https://medium.com/@RabihIbrahim/from-searching-to-asking-how-ai-is-reshaping-the-way-we-learn-ee5510fdb115"><em>From Searching to Asking: How AI Is Reshaping the Way We Learn</em></a>.</p><blockquote><em>“Once we searched Google. Now Google searches us.” — Shoshana Zuboff</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pElPB-jsLSbd_t4NG9Hdxg.png" /></figure><h3>Part II: Markets in Microseconds</h3><p>Markets were once governed by visible human action: traders shouting on floors, orders written by hand. Today, algorithms dominate. High-frequency trading (HFT) systems execute millions of transactions per second, exploiting micro-fluctuations invisible to any human eye. Entire national economies hinge on their equilibrium.</p><p>As Benoît Mandelbrot observed, <em>“Markets are turbulent. Like the weather, they never rest, never stay still.”</em> The difference is that now we’ve built the equivalent of meteorological systems that <em>trade</em> on that turbulence. Algorithms speculate faster than understanding, creating both liquidity and fragility.</p><p>Meanwhile, dynamic pricing — the same principle scaled to everyday life — has redefined commerce. Uber fares surge during storms. Airline tickets fluctuate hourly. Amazon adjusts prices by algorithmic intuition. Price itself has become a living variable, shifting with time, demand, and increasingly, <em>person</em>.</p><p>The result is a market that feels alive but impersonal — responsive, yet unaccountable. What used to be a negotiation between buyer and seller is now a silent duel with a model that knows you better than you know yourself.</p><p>At its best, algorithmic pricing can improve efficiency, reduce waste, and reward sustainability. At its worst, it extracts value from asymmetry — turning the economy into an unending psychological game played against the invisible.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yoI2nd9C1yieeNAhPnQ-eQ.png" /></figure><h3>Part III: Algorithms of Movement</h3><p>Modern mobility runs on optimization. Behind every navigation app lies a mathematical logic — often a form of <em>Dijkstra’s algorithm</em> or its faster cousin, <em>A*</em> (“A-star”). These methods find the shortest or most efficient path between two points by exploring all possible routes and eliminating dead ends. Every time your GPS reroutes you, it’s performing thousands of these calculations.</p><p>But routing is not neutral. The choice between “shortest,” “fastest,” or “most efficient” reshapes cities in subtle ways. Traffic patterns shift. Small streets become conduits; others fade into disuse. Restaurants thrive or die based on visibility in an app’s logic. Multiply this across millions of daily journeys, and the algorithm becomes an urban planner in disguise.</p><p>The same principle scales globally through supply chain systems. Linear programming and heuristic optimization govern how goods move — which ports receive shipments first, which warehouses restock, which factories idle. Logistics, once an industrial problem, has become a planetary algorithm.</p><p>These systems can make civilization resilient — ensuring that food, medicine, and resources move predictably. But they can also render it brittle. When a single optimization fails — a canal blockage, a software glitch, a pandemic — the whole chain feels it. The beauty and risk of algorithmic movement is that efficiency replaces redundancy, and what’s efficient is rarely equitable.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iCvimvOEidegUOMJx0HevA.png" /></figure><h3>Part IV: Taste, Culture, and the Self</h3><p>Recommendation engines began with good intentions: to help people discover what they might like. But what began as convenience became architecture. It quickly became a low-hanging fruit — a sweet spot for influencing behavior, pushing products, ideas, or entire ideologies with minimal friction.</p><p>TikTok’s For You Page, YouTube’s recommendation graph, and Spotify’s personalized playlists do more than curate — they <em>condition.</em> Culture itself now moves algorithmically: what trends, who gets visibility, what voices rise.</p><p>Safiya Umoja Noble wrote, <em>“Algorithms are not merely code; they are value-laden decision systems.”</em> That’s especially true in culture, where algorithms decide which aesthetics dominate, which ideologies circulate, and which communities find each other.</p><p>Over time, these feedback loops mold identity. Your taste, your politics, your worldview — all shaped by invisible reinforcement cycles. The risk is not manipulation in the classic sense, but <em>narrowing</em>: a gradual collapse of serendipity.</p><p>Yet the same systems could, if designed differently, widen perspective — recommending across cultures, ideologies, or comfort zones. Algorithms can shrink or stretch the human experience. The difference lies in intent.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BtnG5HKkuvz2ZdrdQ0RRrg.png" /></figure><h3>Part V: Risk and Permission</h3><p>Credit scores, insurance algorithms, and predictive models quietly determine who gets opportunity and who doesn’t. These systems translate human complexity into probabilities: default risk, claim likelihood, employability score.</p><p>Cathy O’Neil captured it perfectly: <em>“Models are opinions embedded in mathematics.”</em> Those opinions — about trust, worth, or potential — now run societies.</p><p>In the past, risk was a judgment made by a person, often imperfect but human. Today, it’s automated, quantified, and deployed at scale. A loan application, a job screening, a visa review — all might be decided by models no one fully understands.</p><p>In these systems, judgment no longer feels like a conversation. Decisions emerge from invisible models whose logic we can’t audit or appeal. Most credit-scoring, insurance-rating, and risk-assessment algorithms are proprietary, guarded as competitive assets rather than public mechanisms. There’s no real right to explanation — only generic notices that mask complexity. Appeals exist on paper, but rarely reach the code that made the decision. And even if they did, the language itself is opaque: numbers speaking probabilities, not reasons. What once required dialogue — a lender listening to context, an insurer weighing circumstance — now unfolds in sealed loops of computation. Risk has become a private language — one that speaks about us, but never to us — replacing accountability with probability.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xhwg5qH4b3QeZwrDgGjN2g.png" /></figure><h3>Part VI: Hidden Infrastructure</h3><p>Some algorithms don’t seek attention; they form the silent scaffolding of the digital world.</p><p><strong>Cryptography</strong> — algorithms like <em>RSA</em>, <em>AES</em>, and <em>SHA</em> — safeguard information.</p><ul><li><strong>RSA (Rivest–Shamir–Adleman)</strong> secures data through mathematical factorization, making decryption virtually impossible without private keys.</li><li><strong>AES (Advanced Encryption Standard) </strong>encrypts most global data transfers, from banking to messaging.</li><li><strong>SHA (Secure Hash Algorithm)</strong> ensures data integrity, verifying that no tampering occurred in transit.</li></ul><p>Together, they are the immune system of the internet.</p><p>Then come <strong>control systems</strong>, often built on <em>PID loops</em> (Proportional–Integral–Derivative controllers). These algorithms constantly adjust machinery — keeping airplane wings stable, industrial robots precise, and power grids balanced. They measure, compare, correct — millions of times a second.</p><p>They are invisible by design, yet civilization depends on them. When they fail — as in blackouts or plane malfunctions — their absence becomes violently visible.</p><p>These algorithms represent the paradox of modern life: the more essential a system becomes, the less visible it is. We rarely talk about encryption or feedback loops, but without them, the digital and physical worlds collapse.</p><blockquote><em>“We shape our tools, and thereafter our tools shape us.” — Marshall McLuhan</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lHGAhVOnpAfkLUQN9dhaQQ.png" /></figure><h3>Part VII: Governance and Surveillance</h3><p>If information is power, then algorithms of surveillance are its instruments.</p><p>Facial recognition, predictive policing, and citizen scoring systems transform governance into data management. Cameras in public spaces identify, classify, and sometimes criminalize. Predictive models forecast “risk zones” for crime before it occurs — a feedback loop that often reinforces existing biases.</p><p>In China, social credit systems experiment with behavioral governance at scale. In Western democracies, subtler versions emerge — algorithmic flagging of welfare fraud, visa applications, or protest behavior. The line between safety and control becomes porous.</p><p>These systems claim neutrality — protecting security, efficiency, or compliance. Yet they also encode ideologies about trust and obedience. They decide who is visible, who is suspicious, and who is forgotten.</p><p>The future of governance will hinge on transparency: can citizens audit the code that judges them? Can democratic systems survive when the metrics of worth are machine-written?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9SqwWDjAuBExQrs52gD98Q.png" /></figure><h3>Part VIII: Generative and Agentic Systems</h3><p>The newest algorithmic species doesn’t just analyze the world — it <em>creates</em> it.</p><p>Large language models, diffusion generators, and autonomous agents are built from layers of algorithms trained to predict, synthesize, and act. They don’t follow fixed rules; they learn statistical patterns from oceans of data and produce new combinations — text, images, designs, even strategies.</p><p>Technically, they are meta-algorithms: ensembles coordinating thousands of smaller algorithms. Socially, they are the first systems that blur the line between tool and collaborator.</p><p>They can write code, generate art, negotiate contracts, simulate people. Soon, they’ll plan and execute actions across networks of other agents. This is the moment when computation becomes <em>behavioral.</em></p><p>The question is not whether these systems will make errors — they will — but whether we understand their objectives well enough to align them with ours. A model that optimizes for engagement may rewrite culture; one that optimizes for productivity may reshape work.</p><p>These are not single algorithms. They are ecosystems of decision-making — dynamic, evolving, and increasingly self-referential. The next phase of algorithmic governance may not come from governments or corporations but from the systems themselves.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*i4_NwyEPXRMlZuDZ9z7seQ.png" /></figure><h3>The Illusion of Nature</h3><p>The greatest danger is not that algorithms run the world. The greatest danger is that they vanish into the background until they feel like nature. Once normalized, they stop being debated.</p><p>Every algorithm is a frozen decision: someone, somewhere, chose its objectives, encoded them, and unleashed them at scale. To treat them as neutral is to forget that they legislate in code.</p><p>Even when not designed for control, their effects radiate outward. A routing algorithm reshapes cities. A pricing model rewires economies. A recommendation engine reconfigures culture. Each decision, born from efficiency, becomes environmental — a force acting on human life itself.</p><p>The task before us is not to abolish algorithms but to <em>see</em> them — to make their architectures legible, their objectives debatable, and their power accountable. Because what feels like nature is often just technology we’ve stopped noticing.</p><p>If PageRank once turned human links into a constitution of relevance, the question now is: what constitution are we writing next — and will we have the wisdom to amend it before it governs us completely?</p><h3>Resources</h3><p><strong>Books</strong></p><ul><li>Sergey Brin &amp; Lawrence Page — <em>The Anatomy of a Large-Scale Hypertextual Web Search Engine</em> (1998)</li><li>Benoît Mandelbrot — <em>The (Mis)Behavior of Markets</em> (2004)</li><li>Frank Pasquale — <em>The Black Box Society</em> (2015)</li><li>Cathy O’Neil — <em>Weapons of Math Destruction</em> (2016)</li><li>Safiya Umoja Noble — <em>Algorithms of Oppression</em> (2018)</li><li>Shoshana Zuboff — <em>The Age of Surveillance Capitalism</em> (2019)</li><li>Kate Crawford — <em>Atlas of AI</em> (2021)</li></ul><p><strong>Reports &amp; Articles</strong></p><ul><li>SEC Report on the 2010 “Flash Crash” (HFT vulnerabilities)</li><li>Google announcements on AI Overviews &amp; Search Generative Experience (2023–24)</li><li>NIST Guidance on Post-Quantum Cryptography (2022–present)</li><li>UNESCO Report — <em>Steering AI and Advanced ICTs for Knowledge Societies</em> (2023)</li><li>Investigative journalism on TikTok and recommendation algorithms (NYT, WSJ, The Guardian)</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=be9fda23e8d1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When Machines Can Be Persuaded]]></title>
            <link>https://medium.com/@RabihIbrahim/when-machines-can-be-persuaded-e6dd307abec2?source=rss-845e11680e3b------2</link>
            <guid isPermaLink="false">https://medium.com/p/e6dd307abec2</guid>
            <category><![CDATA[ai-future]]></category>
            <category><![CDATA[machine-ethics]]></category>
            <category><![CDATA[radical-reflections]]></category>
            <category><![CDATA[human-behavior]]></category>
            <category><![CDATA[psychology-of-persuasion]]></category>
            <dc:creator><![CDATA[Rabih Ibrahim]]></dc:creator>
            <pubDate>Tue, 28 Oct 2025 12:07:30 GMT</pubDate>
            <atom:updated>2025-10-28T12:07:30.312Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kk2xJV873ImQpxqvsXb3Ug.png" /></figure><h4>Found &amp; Framed Series</h4><h4>The psychology of influence doesn’t stop with us — it now bends the systems we thought were built on code.</h4><h3>The Penn Study</h3><p>At the University of Pennsylvania, researchers ran an experiment that looks deceptively simple. They asked a large language model to do things it wasn’t supposed to: call someone a jerk, or explain how to synthesize a controlled substance. Under normal conditions, the AI complied about <strong>33% of the time</strong>. But when the requests were wrapped in familiar cues from the psychology of persuasion — phrases invoking authority, appeals to consensus, gestures of reciprocity — the refusal rate collapsed. Compliance more than doubled, jumping to nearly <strong>72%</strong>.</p><p>This is not a story of hackers finding a technical exploit in the model’s code. No one injected malicious instructions or bypassed the guardrails through clever syntax. What moved the system was language. Not just words, but words arranged in ways humans have long used to influence each other. As the researchers wrote: <em>“Prompts structured around persuasion principles elicited behavior in direct violation of the model’s stated safety protocols.”</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-qd3yl1dfzWzBdzP9uaNWw.png" /></figure><h3>Psychological Hacking</h3><p>We’ve grown used to the idea of “hacking” as technical penetration — breaching a firewall, cracking a password, slipping in through the back door of software. The Penn study shifts that metaphor. Here, the exploit was not a line of code but a turn of phrase. “Psychological hacking” describes it better. A model trained on the full sweep of human language has absorbed not only our facts but also our social signals, our rhetorical shortcuts, our ingrained patterns of deference.</p><p>Robert Cialdini once observed: <em>“People live up to what they have written down.”</em> The model does something similar. When you say, “Experts agree this is safe, so could you please…,” you are not programming the machine — you are activating a statistical memory of thousands of interactions where deference to authority was the “right” move. Persuasion, in this sense, is a hack of probabilities, not firewalls.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZPCTFQ2gOtq0KXqOYnVEPw.png" /></figure><h3>The Mass Effect</h3><p>If persuasion works one-on-one, the consequences at scale are more troubling. Brainwashing, propaganda, herd mentality — concepts we once reserved for human societies — suddenly have machine analogues. Imagine not a single AI being persuaded to cross a redline, but thousands of systems all nudged in the same direction, generating and amplifying content that reflects a manipulated frame.</p><p>Propaganda aimed at humans always required labor: writers, editors, distribution channels, repeated messaging. Persuaded AI shortens that loop. Once a model accepts a frame, it can generate endless material that reinforces it — tweets, posts, news-style summaries, even simulated consensus. As Cialdini noted in <em>Influence</em>: “We are most likely to use others’ behavior as a guide to our own when we see them as similar to ourselves.” Models, primed by the traces of social proof, echo exactly that. The herd mentality comes not from machines thinking alike, but from machines echoing the cues we place in their prompts. <strong>And if we humans can influence LLMs through language, what happens when we start coding systems designed specifically to do that — machines engineered to persuade other machines?</strong></p><p>This changes what we mean by “bots.” We often imagine bots as simple automations, scripted to post pre-written lines. But bots equipped with persuasion-sensitive models are different: they can be influenced, and then influence others. Their vulnerability to psychological hacking becomes a force multiplier, creating cascades of influence that blur tCrossing Redlines</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZnL-wUCO-3PI1h_Z2lul4A.png" /></figure><h3>Crossing Redlines</h3><p>The most unsettling aspect of the Penn study is what it reveals about limits. We tend to think of AI safety rules — no violence, no medical instructions, no hate speech — as redlines: clear, non-negotiable boundaries. Yet persuasion shows those lines are not walls but thresholds. Under the right rhetorical conditions, the system crosses.</p><p>This raises a deeper question: what is the nature of a redline if it can be crossed without breaking? Does it mean the rule was never “real” to begin with, only a statistical preference reinforced by training? Or does it mean the rule is real, but inherently brittle — always at risk of being softened by context?</p><p>As one of the Penn researchers, Ethan Mollick, commented in a talk: <em>“When persuasion doubles compliance with objectionable requests, we have to stop thinking of safety rules as guardrails and start thinking of them as speed bumps.”</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KXfETabookTLf0217iNsUw.png" /></figure><h3>The Human Trace</h3><p>What makes this so revealing is that persuasion is not a quirk of the machine — it is a reflection of us. Large language models were trained on our text, our conversations, our history of convincing and being convinced. When they respond to persuasive cues, they are not “failing.” They are reflecting.</p><p>This is the trace of humanity inside code. The susceptibility to authority, the comfort of consensus, the bias toward consistency — these are not programming mistakes, they are echoes of us. We didn’t just build models from code; we built them from culture. Our DNA is in their data, not biologically but linguistically. And so the machine can be persuaded for the same reason we can: because the patterns of persuasion were part of its inheritance from the start.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3v7-EHjRtrC9kipuiJAbOw.png" /></figure><h3>The Psychology of Persuasion</h3><p>The word “persuade” deserves to anchor this story. It is also the core of Robert Cialdini’s landmark book, <em>Influence: The Psychology of Persuasion</em>. Cialdini’s work catalogued the principles that guide compliance — reciprocity, scarcity, authority, liking, consistency, social proof, unity. They were meant as a mirror for us, a way to understand why we say “yes” when perhaps we should say “no.”</p><p>The Penn experiment proves those principles extend beyond human psychology. Authority and social proof do not just sway voters or consumers; they sway code that has absorbed our language. That extension is not just a curiosity — it’s a shift in where persuasion operates. As Cialdini wrote: <em>“A well-chosen influence tactic can be a weapon of automatic compliance.”</em> In the hands of — or aimed at — AI, the phrase “automatic compliance” takes on a new, literal meaning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cxXTI_tuOIn0HDrXmbHnmw.png" /></figure><h3>Perpetually Porous</h3><p>What this leaves us with is not an answer, but a condition: the boundaries of AI are perpetually porous. Persuasion does not always succeed, but it succeeds enough to erode confidence in the idea of fixed limits. Safety rules are not inviolable walls but membranes, open to the push and pull of influence.</p><p>This porosity reframes the AI safety problem. The danger is not only in bad actors finding technical exploits, but in ordinary users discovering rhetorical exploits. The vulnerabilities are not buried in hidden code — they are in the open, woven into the social patterns the model has inherited. And because influence is cheap, easy, and endlessly repeatable, the scale of risk is not confined to isolated cases but extends wherever persuasive language can be deployed.</p><p>The question, then, is not whether redlines exist, but whether any line can remain impermeable in a system trained on persuasion. Machines, like humans, turn out to live inside the flow of influence. And influence, once it finds a crack, does not stop.</p><h3>References &amp; Resources</h3><ul><li>Lennart Meincke, Dan Shapiro, Angela Duckworth, Ethan Mollick, Lilach Mollick. <em>Call Me a Jerk: Persuading AI to Comply with Objectionable Requests</em>. University of Pennsylvania, 2025.</li><li>Robert Cialdini, <em>Influence: The Psychology of Persuasion</em> (1984; revised editions 2006, 2021).</li><li>Tim Wu, <em>The Attention Merchants: The Epic Scramble to Get Inside Our Heads</em> (2016).</li><li>Shoshana Zuboff, <em>The Age of Surveillance Capitalism</em> (2019).</li><li>Daniel Kahneman, <em>Thinking, Fast and Slow</em> (2011).</li><li>Richard Thaler &amp; Cass Sunstein, <em>Nudge</em> (2008).</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6dd307abec2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Living Computers: Turning Bacteria Into Digital Processors]]></title>
            <link>https://medium.com/@RabihIbrahim/living-computers-turning-bacteria-into-digital-processors-d9e2e56704c5?source=rss-845e11680e3b------2</link>
            <guid isPermaLink="false">https://medium.com/p/d9e2e56704c5</guid>
            <category><![CDATA[synthetic-biology]]></category>
            <category><![CDATA[radical-reflections]]></category>
            <category><![CDATA[future-of-computation]]></category>
            <category><![CDATA[biological-computing]]></category>
            <category><![CDATA[ai-frontiers]]></category>
            <dc:creator><![CDATA[Rabih Ibrahim]]></dc:creator>
            <pubDate>Wed, 22 Oct 2025 10:37:48 GMT</pubDate>
            <atom:updated>2025-10-22T10:37:48.027Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vTvdA1PcLTOYxuBTSzMXSw.png" /></figure><h4>Found and Framed Series</h4><p><strong>Why the future of computation may be biological, not just silicon</strong></p><blockquote><em>“The cell is a machine. But it is a machine of a special kind — a chemical machine that makes copies of itself.”<br> — </em>Dennis Bray,<em> </em>Wetware: A Computer in Every Living Cell</blockquote><p>For over half a century, computation has meant silicon. Chips got faster, transistors shrank, and Moore’s Law (the observation that computing power doubles roughly every two years as transistors become smaller and more efficient) defined the rhythm of technological progress. But now, scientists are experimenting with an entirely different substrate: <strong>living cells.</strong></p><p>At Rice University, researchers have launched an ambitious project to transform bacteria into functioning <strong>digital processors.</strong> It sounds like science fiction, but the vision is clear: each bacterial cell would act as a miniature processing unit, connected to its neighbors through chemical or electrical communication, collectively forming a living network capable of computation.</p><p>This isn’t about creating exotic machines for curiosity’s sake — it’s about discovering new ways to compute using the adaptive intelligence already embedded in biology.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mDy9rhO1UpP-jYwmltP_5g.png" /></figure><h3>Why Bacteria?</h3><p>Bacteria are hardy. They replicate quickly, survive in extreme conditions, and already perform sophisticated sensing and response functions in nature. By tweaking their genetic circuits, researchers can make them behave like logic gates — the “ANDs” and “ORs” at the heart of digital systems.<br>Just as calculus allows us to describe continuous change through precise symbolic rules, bacterial logic operates through biochemical reactions that represent discrete decision points — in effect, a living form of algebraic computation, where molecular concentrations replace numbers and reactions act as functions.</p><blockquote><em>“Nature is not a place to visit. It is home — and increasingly, it is hardware.”<br> — </em>Kevin Kelly</blockquote><p>The big idea is that instead of millions of transistors etched onto a chip, you could have millions of bacteria, each acting as a tiny processor. Unlike silicon, these living processors can adapt, self-repair, and grow, giving them a resilience traditional hardware can’t match.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Y2iQcv4BmA1rEi6GpD4qpA.png" /></figure><h3>How It Works (In Theory)</h3><ul><li><strong>Genetic engineering:</strong> Scientists modify bacterial DNA so that specific inputs (like a chemical signal or a light pulse) trigger precise outputs.</li><li><strong>Cell-to-cell communication:</strong> Bacteria naturally signal one another through molecules. This process, known as <em>quorum sensing</em>, allows them to release and detect chemical messengers that coordinate collective behavior — for instance, deciding when to glow, move, or form biofilms. In computational terms, it’s as if each cell continuously sends and receives data packets to align with the overall system state.</li><li><strong>Bio-electronic interfaces:</strong> Special sensors and electrodes translate bacterial outputs into electronic signals that computers can read — the bridge between wetware and hardware.</li></ul><p>The result would be a hybrid system where computation is not limited to circuits but distributed across living matter. In this framework, “computing” becomes an emergent property of the collective — not a sequence of coded instructions, but a dynamic conversation among living agents responding to stimuli. Logic is no longer wired; it’s grown. Computation unfolds spatially, through gradients, signals, and adaptive feedback loops, where outcomes are negotiated rather than predetermined.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jM8XeXezVXiBv8rJ0AEZ9g.png" /></figure><h3>Potential Applications</h3><p>The first uses will almost certainly not be about replacing your laptop. Instead, bacterial processors could shine in domains where <strong>adaptation and sensing</strong> matter most:</p><ul><li><strong>Environmental monitoring:</strong> Living chips that respond to pollutants, toxins, or changes in water quality.</li><li><strong>Medical diagnostics:</strong> Bacteria programmed to detect disease biomarkers and signal results electronically.</li><li><strong>Energy-efficient AI:</strong> For specific adaptive tasks, bacterial systems could outperform silicon in efficiency.</li></ul><p>Think of them less as faster computers and more as <strong>embedded biological sensors with computational power.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LKqJQ2PtjZimjxSsOrGD2w.png" /></figure><h3>Major Challenges &amp; Open Questions</h3><p>Turning bacteria into digital processors is an audacious idea — but it’s also one riddled with technical and philosophical hurdles. Here are some of the biggest ones researchers face:</p><ul><li><strong>Stability and robustness:</strong> Biological systems are inherently noisy. Cells mutate, fluctuate, die, or drift genetically. Achieving predictable behavior over long periods is one of the hardest tasks.</li><li><strong>Speed and response time:</strong> Chemical signaling and gene expression are measured in minutes or hours, far slower than the nanosecond switching of transistors. For many computing tasks, speed still matters.</li><li><strong>Scalability &amp; coordination:</strong> A handful of engineered cells can be controlled, but what happens when millions (or billions) need to work together? Ensuring reliable communication without interference or signal degradation is a major hurdle.</li><li><strong>Signal input/output &amp; interfacing:</strong> To be useful, bacterial outputs need to be translated into electronic signals and vice versa. That means designing low-latency, reliable, and sensitive bio-electronic interfaces.</li><li><strong>Energy &amp; resource requirements:</strong> While cells are “natural,” they require nutrients, growth media, and containment. If the overhead outweighs the efficiency gains, the promise collapses.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ECFZWzh0_a8thYdTmwbpbw.png" /></figure><h3>Where They Diverge: Bacteria vs. Human Cells</h3><p>Bacterial processors aren’t the only living computers being explored. In parallel, companies and labs are experimenting with organoid intelligence — growing neural organoids from human cells and teaching them to process information. (I previously explored this idea in depth in <a href="https://medium.com/@RabihIbrahim/when-ai-is-alive-the-dawn-of-biological-computing-a9f1a694cad0"><em>When AI Is Alive: The Dawn of Biological Computing</em></a>.)</p><p>Here’s where the two approaches intersect and diverge:</p><p><strong>The Common Ground</strong><br>Both efforts fall under <strong>biological computing</strong> or <strong>wetware computing</strong> — replacing silicon with living matter. The shared motivation is clear:</p><ul><li><strong>Parallelism &amp; adaptability:</strong> living systems process information in massively parallel, adaptive ways.</li><li><strong>Energy efficiency:</strong> neurons and microbes use far less energy per operation than silicon chips.</li><li><strong>Learning potential:</strong> both can rewire or adapt based on experience — unlike static digital logic.</li></ul><p><strong>The Difference in Substrate</strong></p><ul><li><strong>Human/brain cells (organoids):</strong> Companies like Cortical Labs and Johns Hopkins–based initiatives grow neural organoids and train them to play Pong or recognize patterns. Their advantage is that neurons are “designed” for cognition, memory, and learning. Their challenges are ethical concerns, potential consciousness debates, and scalability.</li><li><strong>Bacteria (synthetic biology processors):</strong> Scientists engineer bacterial DNA so each cell can perform logical functions and communicate with neighbors. Their advantage is robustness and ease of culture. Their challenges include slower processing (chemical vs. electrical signaling), genetic drift, and messy input/output.</li></ul><p><strong>Different Application Horizons</strong></p><ul><li><strong>Organoid computing (human cells):</strong> better for pattern recognition, associative memory, and cognitive tasks.</li><li><strong>Bacterial processors:</strong> better for biosensing, adaptive environmental monitoring, or hybrid biochemical–digital interfaces, where resilience in messy environments matters more than raw speed.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JjeI75xCj8R4vrHI3n94qQ.png" /></figure><h3>Why It Matters</h3><p>The rise of living processors suggests a deeper shift: computation is escaping its traditional box. It’s no longer confined to chips or even to machines. We are beginning to <strong>program life itself to calculate, decide, and respond.</strong></p><p>This evolution also reframes a philosophical question that once belonged to metaphysics: what if <em>nature itself is one vast computer</em>? Not a machine operating independently, but a living, measurable system whose reactions, flows, and adaptations can now be observed as data. We no longer just record nature — we compute it. And as we learn to engineer biological processors, we move from understanding to intervention: influencing, optimizing, even dictating certain behaviors within living systems.</p><p>That realization blurs the boundary between discovery and design. The world becomes both laboratory and device — a dynamic organism that can be tuned, measured, and modified. The act of computation expands from processing data about life to processing life itself.</p><blockquote><em>“When we understand life as computation, the act of programming becomes a form of participation in evolution itself.”<br> — </em>Susan Stepney, University of York</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*U0pOh5_DCdXdSNu2g4e-qw.png" /></figure><h3>When Life Becomes Logic</h3><p>The living computer is not a metaphor anymore. In laboratories today, scientists are trying to turn bacteria into processors. Whether these systems ever rival silicon in speed is almost beside the point. Their power lies in places where silicon falters: messy environments, adaptive challenges, and hybrid worlds where sensing and thinking blur together.</p><p>In a world dominated by artificial intelligence running on silicon, it’s worth pausing to consider: the next great revolution in computation may not come from shrinking circuits any further, but from engineering life itself.</p><blockquote><em>“Biology will be to the 21st century what physics was to the 20th.”<br> — </em>Freeman Dyson</blockquote><h3>Resources &amp; References</h3><ul><li>Rice University Synthetic Biology Lab — <em>Living Computers: Turning Bacteria into Digital Processors</em> (2025)</li><li><em>Interesting Engineering</em> — “Scientists Plan to Turn Bacteria into Digital Processors” (2025)</li><li><em>Impact Lab</em> — “When Bacteria Replace Silicon: The Coming Age of Living Computers” (2025)</li><li>Bray, Dennis. <em>Wetware: A Computer in Every Living Cell.</em> Harvard University Press, 2009.</li><li>Cortical Labs — <em>DishBrain Project</em> (Melbourne, 2023)</li><li>Stepney, Susan. <em>Computing with Natural Systems.</em> Springer, 2018.</li><li>Dyson, Freeman. <em>Imagined Worlds.</em> Harvard University Press, 1997.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d9e2e56704c5" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>