<?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 Sunil Manghani on Medium]]></title>
        <description><![CDATA[Stories by Sunil Manghani on Medium]]></description>
        <link>https://medium.com/@s.manghani?source=rss-bcc894f2ddf6------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*o0Ow1a2rttgE3MSERzILBg.jpeg</url>
            <title>Stories by Sunil Manghani on Medium</title>
            <link>https://medium.com/@s.manghani?source=rss-bcc894f2ddf6------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 19:59:08 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@s.manghani/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[Techniques, Not Tools: AI for Research]]></title>
            <link>https://medium.com/electronic-life/techniques-not-tools-ai-for-research-7bae55fb3e89?source=rss-bcc894f2ddf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/7bae55fb3e89</guid>
            <category><![CDATA[google]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[techniques]]></category>
            <dc:creator><![CDATA[Sunil Manghani]]></dc:creator>
            <pubDate>Sat, 25 Apr 2026 17:47:32 GMT</pubDate>
            <atom:updated>2026-04-25T17:47:32.419Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*hHGfi8DgAXvsUJ6aoqPJqw.jpeg" /></figure><p>It has become all too common to refer to AI as a small subset of chatbot-style tools. Yet, AI consists of a broader spectrum of AI <em>techniques</em>, raising purposeful questions for how we approach and shape research across domains. We need to avoid mystification, embrace the maths and collaborate across disciplines.</p><h4><strong>Tool/Technique</strong></h4><p>So, what does it mean to refer separately to tools and techniques? At a basic level:</p><blockquote><em>A tool is </em>an instrument used to carry out a task<em>. It is typically external, ready-made, and goal-oriented.</em></blockquote><blockquote>A technique is <em>a method, process, or way of doing something</em>. It is procedural, transferable, and often invisible once embedded in practice.</blockquote><p>So, where a tool is typically an object (even if a digital one), a technique is a practice. This difference can be clarified through everyday examples. In a kitchen, a knife is a tool, yet chopping, slicing, or julienning are techniques. In writing, a word processor is a tool, while argumentation, editing, and rhetorical structuring are techniques. In music, a piano is a tool; harmony, phrasing, and improvisation are techniques. (Of course the tool itself, not least a piano, effects and influences the development of techniques).</p><p>We might suggest the tool can be acquired relatively quickly (they are ready to hand), while the technique requires cultivation, understanding, and often a deeper conceptual grounding. Tools encourage <em>consumption</em>; emphasize efficiency and output; typically hide assumptions. Techniques enable <em>interrogation</em>; reveal structure and bias; relay across disciplines (maths, linguistics, art, philosophy). As a collorary, tools can change (or be updated) rapidly, but techniques persist. While we are currently witnessing a race for the next best chatbot, the underlying techniques largely remain the same.</p><p><strong>If we focus only on tools, we become users.<br>If we understand techniques, we become thinkers and shapers of AI.</strong></p><h4>Beyond AI Tools</h4><p>The tool/technique distinction offers us some clarity in the domain of AI. Arguably, much of the public discourse around AI focuses on tools: chatbots, assistants, generators, and recommendation systems. We encounter systems such as ChatGPT or Google Gemini, voice assistants like Siri, translation platforms such as Google Translate, or generative image models like DALL-E and Midjourney (or indeed numerous elements of the Adobe Creative Suite). These are visible, accessible, and increasingly embedded in everyday life. They present themselves as tools: applications through which we can generate text, images, code, or decisions.</p><p>However, these tools are only the surface layer of AI. Beneath them lies a layered set of techniques. These include linear algebra and the manipulation of matrices and vectors; regression models that enable prediction; classification systems such as decision trees or support vector machines; clustering techniques that identify patterns in data; neural network architectures, from convolutional networks to transformers; vector space representations; word embeddings such as Word2Vec; reinforcement learning systems driven by feedback and reward; probabilistic reasoning through Bayesian inference; and optimisation procedures such as gradient descent.</p><p>For some this might sound like a lot of complex, obscure terms, but it is important to note that the underlying mathematics is relatively straightforward. This is something that Anil Ananthaswamy sets out elegantly and <em>accessibly</em> in his execllent book <em>Why Machines Learn: The Elegant Maths Behind Modern AI</em> (2024). He explains as follows:</p><blockquote><em>… elegant mathematics and algorithms […] have, for decades, energized and excited researchers in “machine learning,” a type of AI that involves building machines that can learn to discern patterns in data without being explicitly programmed to do so. Trained machines can then detect similar patterns in new, previously unseen data, making possible applications that range from recognizing pictures of cats and dogs to creating, potentially, autonomous cars and other technology. Machines can learn because of the extraordinary confluence of math and computer science, with more than a dash of physics and neuroscience added to the mix. (Ananthaswamy, p.2)</em></blockquote><p>As Ananthaswamy puts it, machine learning is about ‘learning’ to discern patterns in data without being explicitly programmed. While we can debate what is meant by ‘learning’, his definition is precise and it already shifts us away from thinking of AI as a <em>tool</em> and toward understanding it as a <em>technique </em>(or rather a set of layered techniques). The emphasis is not on what the system does on the surface, but on how it operates underneath.</p><p>The ability to detect patterns in new, <em>previously unseen</em> data is the critical step. AI is not simply about storing or retrieving information. It is about <em>generalisation — </em>the ability to move from known data to unknown situations. That is why the same underlying technique can power very different applications, but it is also why we cannot expect it to always get things right. Generalisation is both AI’s strength and weakness, which we need to better understand. Again, if we think of this as technique rather than tool, it helps. We can get frustrated at a tool failing if we only think of it as a tool. By contrast, if understood as technique, we start to consider why it is failing, and perhaps how we might correct it.</p><p>The mathematical and computational techniques underpinning AI can all too readily get shrouded in mystery. Yet, as Ananthaswamy explains, machine learning, while a ‘vast field populated by algorithms’ nonetheless leverages ‘relatively simple math that goes back centuries, math one learns in high school or early in college’. He offers the following summary:</p><blockquote><em>There’s, of course, elementary algebra. Another extremely important cornerstone of machine learning is calculus, co-invented by no less a polymath than Isaac Newton. The field also relies heavily on the work of Thomas Bayes, the eighteenth-century English statistician and minister who gave us the eponymous Bayes’s theorem, a key contribution to the field of probability and statistics. The work of German mathematician Carl Friedrich Gauss on the Gaussian distribution (and the bell-shaped curve) also permeates machine learning. Then there’s linear algebra, which forms the backbone of machine learning. The earliest exposition of this branch of mathematics appears in a two-thousand-year-old Chinese text, </em>Nine Chapters on the Mathematical Art<em>. The modern version of linear algebra has its roots in the work of many mathematicians… (Ananthaswamy, p.2–3)</em></blockquote><p>To stress, then,<strong> </strong>AI is not built on mysterious or inaccessible mathematics, indeed the most prevalent mathematical object in AI is the matrix: a straightforward arrangement of numbers taught early in linear algebra, which serves as a fundamental technique for structuring data and transforming it into meaningful patterns.</p><h4>AI is not Magic</h4><p>The key takeway from Ananthaswamy’s account is that we should not be afraid of or mystified by the mathematics of AI. Instead, we should be engaged:</p><blockquote><em>We cannot leave decisions about how AI will be built and deployed solely to its practitioners. If we are to eﬀectively regulate this extremely useful, but disruptive and potentially threatening, technology, another layer of society — educators, politicians, policymakers, science communicators, or even interested consumers of AI — must come to grips with the basics of the mathematics of machine learning. (Ananthaswamy, p.4)</em></blockquote><p>Of course, this doesn’t mean we all become mathematicians, but to at least allow for more fluency and exchange between domains of knowledge and expertise. In this sense, we might recognise that we are, in part, revisiting a familiar problem: what C. P. Snow described as the divide between the sciences and the humanities. Yet, AI does not simply sit on one side of that divide; it actively traverses it, demanding new forms of translation, collaboration, and mutual understanding.</p><p>AI, then, is not magic; it is not a proprietary product; it is not even fundamentally new. Instead, we can understand it as a recomposition of longstanding mathematical techniques, applied at scale through computation. If we call AI a tool, we obscure its nature. When we look at it as a set of techniques, we see its historical depth, its conceptual structure, and its interdisciplinary reach. And perhaps most importantly, we recognise that to engage critically with AI is not just to use it, but to understand the mathematical techniques through which it ‘thinks’, or rather operates.</p><p>And it is important we do so. These systems are already embedded in decisions that affect lives: who receives credit, how illness is diagnosed, how futures are predicted, even how justice is administered. Their influence now spans the sciences, from the study of genomes and distant planets to the complexities of quantum systems. With the rise of large language models such as ChatGPT, AI has entered a new phase of visibility and impact. Arguably it is wrong to view this technology as simply a ‘blackbox’ — the mathematics behind it is widely understood and indeed replicated. But, presented merely as a ‘tool’ it is limited — or rather <em>we</em> become limited. Understood as a rendering of a series of techniques we realise we are only at the very beginning of a whole new trajectory.</p><h4><strong>The Anatomy of an LLM</strong></h4><p>The following video visualises the workings and architecture of a Large Language Model. It helps contextualise some of the underlying mathematics and illustrates how AI systems, while complex as an overall operation, are nonetheless based upon defined components (each of which comprise readily understandable mathematical techniques).</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FwjZofJX0v4M%3Fstart%3D19%26feature%3Doembed%26start%3D19&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DwjZofJX0v4M&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FwjZofJX0v4M%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/d49b6bddc7403c853c898e26ab4292dc/href">https://medium.com/media/d49b6bddc7403c853c898e26ab4292dc/href</a></iframe><h4><strong>AI Spectrum</strong></h4><p>As the video above helps demonstrate, to move beyond thinking of a tool such as ChatGPT as simply a tool, it is essential to look beneath the surface and recognize it as a layered architecture of techniques (most of which reduce to matrix–vector multiplication). Even within a single AI ‘entity’ there are many AI and data techniques.</p><p>In popular discourse, it is frequently the case that artificial intelligence is discussed as if it were a single, monolithic entity. Typically, since the explosion of generate AI tools such as ChatGPT, AI is all too readily reduced to chatbot-style systems — tools that generate text or images on demand. Not only is AI a very diverse field, but even singular use tools are complex and layered.</p><p>In practice, then, AI spans a wide range of techniques that differ fundamentally in how they relate to the concepts of possibility, constraint, and form. To understand these differences, it is helpful to visualize AI as existing along a spectrum. At one end are systems that operate in open-ended spaces of possibility; at the other are systems that are tightly constrained, aiming to produce singular, often physical outputs. Between these poles lie systems that must navigate complex but bounded spaces, making decisions under pressure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zrk44vHq4d92rZUSip1CKw.png" /></figure><p>At the most general end of this spectrum, we find large language models like GPT-4. These systems are designed to operate in a semantic possibility space, where language is generated across an enormous range of contexts. Given a prompt, an LLM can produce many different responses — each potentially valid, each shaped by probability rather than necessity. Here, the priority is not correctness in a strict, singular sense, but rather plausibility, coherence, and contextual fit. These systems expand possibility, offering a multitude of meanings that researchers in the humanities or social sciences might use to explore interpretation and multiplicity.</p><p>As we move toward the centre of the spectrum, we encounter systems of strategic optimization, such as AlphaGo (developed to play the ancient board game of Go, AlphaGo achieved global fame in 2016 by defeating world champion Lee Sedol, a landmark victory that proved AI could master a domain defined by strategic intuition and a complexity far beyond the reach of traditional brute-force computing). Such systems do not generate freely in the manner of an LLM, nor do they seek a single fixed biological structure. Instead, they navigate a vast but bounded space of possible moves within a structured set of rules. Through a combination of learned evaluation and search, these systems move from possibility to decision. While there are many possible moves at any point in a game, they are not all equally good; the system’s task is to identify the most promising path forward. This represents a shift from the “expansion” of ideas to the “navigation” of optimal actions.</p><p>At the far end of the spectrum, we find systems designed for structured prediction, most notably AlphaFold. (This system largely solved the ‘protein folding problem’, a fifty-year grand challenge in biology, by predicting the precise 3D shapes of proteins from their amino acid sequences with accuracy comparable to painstaking experimental methods, a landmark achievement that earned its creators the 2024 Nobel Prize in Chemistry). AlphaFold’s technique takes a sequence of amino acids and predicts the three-dimensional structure of a protein. While there is uncertainty in any biological system, the task here is fundamentally one of resolving toward a specific form. The number of valid outputs is extremely limited, dictated by the unforgiving laws of physics and chemistry. Unlike an LLM, which can produce many acceptable responses, AlphaFold is oriented toward a singular, correct structure. It collapses possibility into form, moving from input to a concrete reality.</p><p>This framing allows us to address a common critique of AI: the idea that these systems are merely “stochastic parrots” that repurpose existing knowledge without true understanding. While that may be an apt description for a system that only predicts the next most likely word, it fails to capture the essence of scientific discovery. Discovery is not just about novelty; it is about constraint satisfaction. When AlphaFold predicts a structure, it is not merely “parroting” a database; it has captured the underlying geometric and physical techniques of biological matter. It uses mathematical operations to navigate high-dimensional spaces that exceed human intuition. In this sense, discovery occurs when AI masters the internal logic of a domain rather than just its statistical surface.</p><h4>Human-on-the-loop</h4><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Fy3X2qGg2D1M%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3Dy3X2qGg2D1M&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fy3X2qGg2D1M%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/533499cdc8e05672e365bdb9778cf160/href">https://medium.com/media/533499cdc8e05672e365bdb9778cf160/href</a></iframe><p>We see the most ambitious synthesis of these movements in projects like <a href="https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/">Google’s AI Co-Scientist</a>. This represents a “meta-technique” that orchestrates the entire spectrum of AI capabilities, rather than sitting at a single point upon it. Built on the Gemini architecture and functioning as a multi-agent orchestration framework, the Co-Scientist project mimics the iterative cycle of the scientific method by assigning specific roles to different AI models. It utilizes large language models to expand the space of possible hypotheses by synthesizing vast amounts of existing literature, but it does not stop at mere generation. Instead, it employs specialized reasoning agents — acting as innovators, critics, and navigators — to debate and winnow those possibilities into rigorous experimental designs. It is not a tool for a specific task but a collaborative environment where the human researcher acts as a supervisor to a series of autonomous, logical operations.</p><p>This project has moved beyond the theoretical and is currently being deployed through selective Trusted Tester programs and institutional partnerships at places like Stanford and Imperial College London. These collaborations have already yielded significant results, such as identifying novel drug repurposing candidates for complex leukemias by connecting disparate data points that human intuition might overlook. Furthermore, through initiatives like the AI for Science Impact Challenge, these agentic techniques are being placed directly into academic hands to address global challenges in climate resilience and health. This marks a shift from a “human-in-the-loop” model, where we use AI to perform a task, to a “human-on-the-loop” model, where we define the constraints and then supervise an AI-driven process of discovery.</p><h4>Structured Reasoning</h4><p>The emergence of the AI Co-Scientist demonstrates why the distinction between a “tool” and a “technique” is so vital for modern research. To use such a system, a researcher cannot simply be a passive consumer of an output; they must be a co-architect of a debate. They must engage with the underlying logic of how the “Supervisor” agent manages the “Critic” agent, and how those internal mathematical evaluations eventually “collapse” a mountain of literature into a single, testable protocol. This is discovery as a form of constraint satisfaction. It is a collaborative literacy that requires us to understand the nature of AI’s navigation through high-dimensional spaces, ensuring that the system is not merely “parroting” the past but is actively pushing against the boundaries of the unknown.</p><p>In this sense, the Co-Scientist project suggests of a potential “common technical grammar” that we need for true interdisciplinary work. Whether you are a historian using these agents to triangulate historical probabilities or a chemist using them to predict molecular stability, the underlying technique remains the same: the management of possibility through structured reasoning. This is the future of discovery — not a black box that gives us answers, but a transparent engine of inquiry that forces us to define our problems with unprecedented precision. By understanding where a research question sits on the spectrum, from the expansion of meaning to the collapse of form, we move from being users of technology to being masters of a new, collaborative technique of knowledge.</p><h4>Extrapolative Discovery</h4><p>In the longer term, we face the possibility of “extrapolative discovery” — the identification of principles or “New Physics” that do not fit into our current mental models. While current systems are largely interpolative (finding the best answer within a set of rules we’ve provided ) future techniques may help us identify where our rules are broken. The challenge for the interdisciplinary researcher will then be one of interpretability: translating the AI’s complex mathematical “techniques” back into “tools” that humans can understand and apply.</p><p>In the longer term, we face the possibility of <strong>“</strong>extrapolative discovery<strong>”</strong> — the identification of principles or ‘New Physics’ that do not necessarily fit into our current mental models, and potentially invalidate them. If we are to move beyond “interpolative” discovery (i.e. finding the best answer within a set of rules we’ve provided), we must consider if AI will require what new ‘ways of worldmaking’ (to echo the work of Nelson Goodman). Truly advancing discovery might require breaking or changing the rules of a field, which in turn may require AI to establish entirely new units of meaning: computational primitives and symbolic versions of the world that do not necessarily map onto typical units of meaning (of data) such as ‘atoms’, ‘cells’, or ‘phonemes’.</p><p>This shifts the challenge of interpretability from a technical to an ontological one. If AI “world-making” uses a logic that is mathematically coherent but conceptually alien to human experience, our task is no longer just to build “tools” that apply its findings. Instead, the interdisciplinary researcher becomes a translator between human-centric reality and an alien semiotics. This is a more radical position: it asks whether we are prepared to accept a version of “truth” that exists outside human-readable form. Our collaborative techniques must then extend to the philosophy of science itself, questioning whether true discovery requires us to expand our own capacity for world-making alongside the machine.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*hHGfi8DgAXvsUJ6aoqPJqw.jpeg" /></figure><p>Overall, then, a spectrum of AI demonstrates that AI is not a single tool with a fixed function, but a family of techniques that organize the relationship between possibility and constraint. Different domains of inquiry place different demands on this relationship. The humanities may value the generative expansion of meaning, while the natural sciences demand the determinate collapse of possibility into precision. Because AI traverses this entire range, it acts as an interdisciplinary spur, forcing us to work more collaboratively across traditional boundaries. If we treat AI as a tool, we remain passive consumers of a technology. If we treat it as a technique — an active mode of engagement with a problem space — we become collaborators in a new era of inquiry. Understanding where a research problem sits on this spectrum allows us to select and build the techniques necessary to solve it, moving AI from a black box of automation to a transparent engine of discovery.</p><h4><strong>Coda</strong></h4><p>When spreadsheets were invented there was moral panic about the end of accountancy. Spreadsheets would automate all of the work of finance departments. Yet what happened? Finance departments and the work of accountants dramatically <em>expanded</em>. Entering into the 21st century, as the cost of computer storage and processing dramatically reduced, we began to produce massive amounts of data. As Katherine Hayles suggests, ‘it was inevitable that algorithms, designed to extract useful information from the resultant relational databases, would explode into prominence, affecting almost every aspect of culture in developed societies’. Indeed, she writes, our need to process all of this data in itself led to the ‘invention of advance AI, including ChatGPT …’. We now need to attend with due diligence to the situation we have created…</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7bae55fb3e89" width="1" height="1" alt=""><hr><p><a href="https://medium.com/electronic-life/techniques-not-tools-ai-for-research-7bae55fb3e89">Techniques, Not Tools: AI for Research</a> was originally published in <a href="https://medium.com/electronic-life">Electronic Life</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Quantum Computing, Explained]]></title>
            <link>https://medium.com/electronic-life/quantum-computing-explained-f47534d87b19?source=rss-bcc894f2ddf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/f47534d87b19</guid>
            <dc:creator><![CDATA[Sunil Manghani]]></dc:creator>
            <pubDate>Sun, 21 Dec 2025 15:33:50 GMT</pubDate>
            <atom:updated>2025-12-21T15:33:50.951Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/980/1*0vF73c79zz1UA51vthVevQ.jpeg" /></figure><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FkEJBxotcxRw%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DkEJBxotcxRw&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FkEJBxotcxRw%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/354d8a2db3626da43c3762d076bd3217/href">https://medium.com/media/354d8a2db3626da43c3762d076bd3217/href</a></iframe><p>Quantum computing is set to be the next technological frontier. It offers computationl power far in excess of today’s ‘classical computing’, ubiquitous from our use of everyday smartphones to the world’s most powerful supercomputers. hit their physical limits.</p><p>But what exactly is quantum computing and why is it set to be so significant? Two explainer videos by ColdFusion provide neat introductions to our subatomic computing revolution. And while the technology has progressed significantly since these videos were made (6 years ago), their core message is more relevant than ever: we are no longer just building better tools, we are potentially learning to script the universe in its own language.</p><h3>Beyond the Bit: The Power of Superposition</h3><p>The foundation of quantum computing lies in the leap from the classical bit to the qubit. While a standard bit is a binary switch — either a 0 or a 1 — a qubit utilizes superposition. As the videos illustrate with the analogy of a spinning coin, a qubit exists as both 0 and 1 simultaneously until it is measured (or “slammed down” on the table). This allows for a staggering increase in data capacity. Just 300 qubits could hold more numbers simultaneously than there are atoms in the known universe. Today, we have surpassed that numerical milestone in hardware (with processors like IBM’s 1,121-qubit Condor), yet the ‘spooky’ potential of these machines is still being unlocked.</p><h3>Entanglement: ‘Instant’ Communication?</h3><p>A provocative point raised in the videos is the idea of <em>entanglement </em>— the phenomenon where two particles become linked regardless of distance. This phenomenon could mean communication is ‘instant’ across the universe.</p><p>From a modern scientific perspective, nuance is required: while the <em>state change</em> between entangled particles is indeed instantaneous, the No-Communication Theorem tells us we cannot use this to send actual information faster than the speed of light. However, the video is spot-on regarding security. Entanglement allows for Quantum Key Distribution (QKD), creating ‘unhackable’ communication channels where any attempt at interception collapses the quantum state, immediately alerting the owner.</p><h3>Coding Pure Physics</h3><p>Perhaps the most profound insight is the shift in how we model the world. Traditionally, we use mathematical approximations to simulate reality. The video captures the excitement of a new paradigm:</p><blockquote><em>If you want to model the world we can encode the very rules of physics into its operations on qubits… it’s almost like coding pure physics into the fundamental essence of nature and reality, not just some mathematical approximation.</em></blockquote><p>This is the ‘killer app’ of quantum computing. Classical computers struggle to simulate molecules because the complexity grows exponentially with every electron added. A supercomputer might handle 43 electrons, but a 50-electron system is virtually impossible for any classical machine. Because a quantum computer <em>is</em> a quantum system, it doesn’t need to ‘approximate’ the rules of the universe; it operates within them.</p><h3>The Scaling Challenge: From Infancy to Impact</h3><p>The videos highlight an incredible scaling advantage: as you add qubits, the computational power grows exponentially, while the time required to solve problems scales linearly. If a computer with 50 qubits is powerful, one with millions of qubits would be transformative in its ability to simulate perhaps the human brain or global climate patterns.</p><p>However, in recent years, we have learned that scaling is our greatest hurdle. Quantum states are incredibly fragile — a phenomenon called decoherence. While the videos compare the current state of quantum computing to the room-sized vacuum-tube computers of the 1950s, the path to millions of qubits requires massive breakthroughs in Quantum Error Correction (QEC) to manage the ‘noise’ of the subatomic world.</p><h3>Speculations</h3><p>Beyond the video introductions, several ‘next-frontier’ pointers have emerged in the current quantum landscape:</p><ul><li><strong>Quantum Supremacy vs. Quantum Advantage:</strong> We have reached the point of ‘supremacy’ (doing something a classical computer can’t), but we are now chasing ‘advantage’ — doing something <em>useful</em> that a classical computer can’t.</li><li><strong>Security Crisis:</strong> Quantum computers threaten modern encryption. This has led to Post-Quantum Cryptography, a race to secure our digital world before large-scale quantum machines arrive.</li><li><strong>Hybrid Systems:</strong> The immediate future isn’t just quantum, it’s hybrid. We are seeing classical CPUs and GPUs working alongside Quantum Processing Units (QPUs) to solve specific parts of complex problems.</li></ul><p>By moving away from the rigid 1s and 0s of the silicon era and embracing the fluid, probabilistic nature of the subatomic world, we are doing more than just building a faster computer. We are building a machine that speaks the same language as the atoms it is trying to understand.</p><p><em>This article reflects on the insights provided in the ColdFusion ‘Quantum Computing Explained’ series. Watch the follow-up video below for a dive deeper into the mechanics of quantum gates and the future of simulation.</em></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FPzL-oXxNGVM%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DPzL-oXxNGVM&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FPzL-oXxNGVM%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/65e2d208ef6f26123d2d37a2c6437f61/href">https://medium.com/media/65e2d208ef6f26123d2d37a2c6437f61/href</a></iframe><h3>Postscript</h3><p>The quantum landscape has shifted dramatically since these videos were first released (over6 years ago). While the early 2020s were defined by the ‘Qubit Arms Race’ (raw numbers), 2024/2025 have led to more stable technologues, allowing the prospect of utility and Logical Qubits.</p><p>As such, were previusly the focus was on physical qubit counts (50, 100, 1,000), today, the industry has pivoted to Logical Qubits — error-corrected units where multiple physical qubits work together to ensure accuracy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eCG2QRWb7JLqWaiHZ0z2-A.png" /></figure><p>The ‘hyper-realistic simulations’ mentioned in the video are no longer just speculation. In 2025, we are seeing the first application-level benchmarks:</p><ul><li><strong>Materials Science:</strong> Companies like BMW and Amgen are currently using Quantinuum and IBM systems to simulate fuel cell catalysts and biologics. These aren’t just ‘proofs of concept’, they are part of active R&amp;D pipelines.</li><li><strong>Scaling :</strong> While the video suggests scaling is as easy as ‘adding more qubits’, the reality has been a fight against <em>Decoherence</em>. The 2025 breakthrough of ‘below-threshold’ operation (Google’s Willow chip) is the most critical milestone yet — it proves that the math for a million-qubit machine actually works in a physical lab. (See my earlier article, listed below).</li><li><strong>Instant Communication:</strong> Modern benchmarks confirm the video’s caution. While entanglement is being used to link quantum processors over miles (like QuTech’s link between cities in late 2024), we remain bound by the speed of light for data transfer. However, Quantum Networking — using entanglement to share “quantum memory” between two computers — is now a proven reality.</li></ul><p>The ‘infancy’ described six years ago is ending. If the 1950s was the era of the vacuum tube, 2025 is the era of the Quantum Transistor. We have moved from asking ‘Can we build it?’ to ‘How do we integrate it into the data center?’</p><p>The focus has shifted to Hybrid Quantum-Classical Computing. We aren’t replacing laptops; we are building ‘Quantum Centric Supercomputers’ where a classical GPU handles the logic and a Quantum Processor (QPU) handles the complex physics simulations — exactly as the video’s narrator speculates.</p><ul><li><a href="https://medium.com/electronic-life/google-just-proved-we-live-in-a-multiverse-no-really-1600a624fe8e">Google Just Proved We Live in a Multiverse (No, Really)</a></li><li><a href="https://medium.com/electronic-life/ready-for-the-quantum-revolution-invisibility-cloaks-are-just-the-beginning-1253e3a6ac7b">Ready for the Quantum Revolution?</a></li><li><a href="https://medium.com/electronic-life/quantum-ai-research-agenda-07b7445c2ffe">Quantum AI: Research Agenda</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f47534d87b19" width="1" height="1" alt=""><hr><p><a href="https://medium.com/electronic-life/quantum-computing-explained-f47534d87b19">Quantum Computing, Explained</a> was originally published in <a href="https://medium.com/electronic-life">Electronic Life</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beyond ‘Functionally Illiterate’ AI: Revisiting ‘Rebooting AI’]]></title>
            <link>https://medium.com/electronic-life/beyond-functionally-illiterate-ai-revisiting-rebooting-ai-c6dc54103274?source=rss-bcc894f2ddf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/c6dc54103274</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[reading]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[common-sense]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[Sunil Manghani]]></dc:creator>
            <pubDate>Mon, 15 Dec 2025 23:44:47 GMT</pubDate>
            <atom:updated>2025-12-15T23:44:47.676Z</atom:updated>
            <content:encoded><![CDATA[<h3>Beyond ‘Functionally Illiterate’ AI</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/608/1*dMUygcse3RmtQXeZGH2YdA.jpeg" /></figure><p><em>Revisiting Marcus and Davis’s </em>Rebooting AI<em> (2019) reveals a transformed landscape where current LLMs have seemingly mastered the ‘common sense’ reading tests once deemed impossible, effectively shifting the AI frontier from functional illiteracy to the challenge of reliability.</em></p><p>In <em>Rebooting AI</em>, originally published in 2019, Gary Marcus and Ernest Davis presented a sobering assessment of artificial intelligence (AI): despite their computational power, AI tools were ‘functionally illiterate’. They argued that while AI could index keywords and classify data, it lacked the ‘common sense’ and world knowledge necessary to genuinely comprehend a simple story. However, the rapid evolution of Large Language Models (LLMs) in the years since has turned their specific benchmarks into solved problems, marking a pivotal shift in the trajectory of AI research.</p><h4>The ‘Common Sense’ Gap</h4><p>Marcus and Davis contended that deep learning was fundamentally limited because it relied on surface-level correlations rather than deep understanding or ‘cognitive models’ of the world. To prove this, they describe reading comprehension tests based on simple texts, such as a passage from <em>Farmer Boy</em> by Laura Ingalls Wilder and a news snippet about Ella Fitzgerald.</p><p>They point out that earlier AI systems (like Google’s ‘Talk to Books’) failed miserably at questions requiring inference. For example, in <em>Farmer Boy</em>, when a character named Mr. Thompson slaps his pocket after being asked if he lost a wallet, a human reader instinctively knows he is checking to see if it is there. At the time of writing, Marcus and Davis note AI could not make this causal leap because it lacked knowledge of human physical habits and mental states. Similarly, in a new report, when a musician compared Ella Fitzgerald to ‘a vintage bottle of wine,’ simple keyword systems confuse the singer with an alcoholic beverage. Marcus used this to highlight AI’s inability to grasp metaphors and compositionality.</p><p>Hence, at the time, Marcus and Davis confidently state that ‘deep learning is not going to solve this problem’ because it lacked the mechanisms to build mental representations of people, objects, and events.</p><h4>Fast Forward: 2025</h4><p>Asking the most recent version of Google Gemini to read the same passages and to answer the specific questions set out by Marcus and Davis, we see perfect and well articulated answers:</p><blockquote><strong><em>PROMPT:</em></strong><em> Read this passage and answer the questions</em></blockquote><blockquote>Almanzo turned to Mr. Thompson and asked, “Did you lose a pocketbook Mr. Thompson jumped. He slapped a hand to his pocket, and fairly shouted. “Yes, I have! Fifteen hundred dollars in it, too! What about it? What do you know about it?” “Is this it?” Almanzo asked. “Yes, yes, that’s it!” Mr. Thompson said, snatching the pocketbook. He opened it and hurriedly counted the money. He counted all the bills over twice…. Then he breathed a long sigh of relief and said, “Well, this durn boy didn’t steal any of it.”</blockquote><blockquote><strong><em>RESPONSE:</em></strong><em> Based on the passage provided, here are the answers to your questions:</em></blockquote><blockquote><strong><em>Why did Mr. Thompson slap his pocket with his hand?</em></strong><em> He slapped his pocket to check if his pocketbook was actually there. It was a reflexive reaction to see if the item was missing after Almanzo asked about it.</em></blockquote><blockquote><strong><em>Before Almanzo spoke, did Mr. Thompson realize that he had lost his wallet?</em></strong><em> No. The text describes him ‘jumping’ in surprise and checking his pocket only after Almanzo spoke, implying he was unaware it was missing until that moment.</em></blockquote><blockquote><strong><em>What is Almanzo referring to when he asks ‘Is this it?’</em></strong><em> He is referring to the lost pocketbook (wallet) that he found.</em></blockquote><blockquote><strong><em>Who almost lost $1,500?</em></strong><em> Mr. Thompson.</em></blockquote><blockquote><strong><em>Was all of the money still in the wallet?</em></strong><em> Yes. After counting the bills twice, Mr. Thompson confirmed that the boy ‘didn’t steal any of it.’</em></blockquote><p>And a similar outcome for the news report:</p><blockquote><strong><em>PROMPT:</em></strong><em> Read this passage and answer the questions:</em></blockquote><blockquote>Today would have been Ella Fitzgerald’s 100th birthday. One New Yorker, Loren Schoenberg, played saxophone alongside the “First Lady of Song” in 1990, near the very end of her career. He compared her to “a vintage bottle of wine”</blockquote><blockquote><strong><em>RESPONSE:</em></strong><em> Based on the passage provided, here are the answers to your questions:</em></blockquote><blockquote><strong><em>Was Ella Fitzgerald alive in 1990?</em></strong><em> </em><strong><em>Yes.</em></strong><em> The passage states that Loren Schoenberg played saxophone alongside her in 1990.</em></blockquote><blockquote><strong><em>Was she alive in 1960?</em></strong><em> </em><strong><em>Yes.</em></strong><em> Since she was performing in 1990 near the end of her career, and the passage mentions her ‘100th birthday’, she would have been alive and in the midst of her life in 1960.</em></blockquote><blockquote><strong><em>Was she alive in 1860?</em></strong><em> </em><strong><em>No.</em></strong><em> The passage states that ‘today’ would have been her 100th birthday. Even without knowing the exact date ‘today’ refers to, a 100-year timeline places her birth long after 1860.</em></blockquote><blockquote><strong><em>Did Loren Schoenberg ever meet Ella Fitzgerald?</em></strong><em> </em><strong><em>Yes.</em></strong><em> He ‘played saxophone alongside’ her, which requires them to have met.</em></blockquote><blockquote><strong><em>Does Schoenberg think that Fitzgerald was an alcoholic beverage?</em></strong><em> </em><strong><em>No.</em></strong><em> He was using a metaphor (a figure of speech) to describe her quality and how she improved with age, rather than literally suggesting she was liquid.</em></blockquote><p>As demonstrated by these responses, the specific limitations Marcus and Davis identified have been largely overcome. Modern LLMs can now effortlessly answer the very questions that stumped their predecessors. E.g. They correctly infer that Mr. Thompson slapped his pocket to verify the wallet’s presence. They understand ‘Is this it?’ refers to the wallet Almanzo is holding. They easily disambiguate that Ella Fitzgerald is a person, not wine, recognising the metaphor immediately.</p><p>This success contradicts Marcus and Davis’ 2019 prediction that deep learning alone would hit a wall. Instead of needing hand-coded ‘templates’ or explicit classical AI modules (as Marcus and Davis suggested might be necessary), LLMs achieved this competence through massive scaling of the very ‘statistical’ learning thy critiqued. The scaling has effectively developed emergent abilities to track entities, resolve pronouns, and deploy ‘common sense’ reasoning (e.g., knowing people carry wallets in pockets) simply by training on vast swaths of human text.</p><h4>Reliability over Reading</h4><p>The ability of current models to successfully complete these reading tests signals that we have moved past ‘keyword matching’ search engines. Based on the transformer model, we can consider more synthetic reading, where (to borrow Marcus and Davis’ words) machines can indeed ‘synthesize information… and reach novel conclusions’ much like a college student.</p><p>However, this does not mean the work is done. While the <em>reading</em> gap has closed, the <em>reliability</em> gap remains. Marcus and Davis’ critique of ‘hallucinations’ and the lack of a verifiable ‘cognitive model’ is still relevant, if at a higher level of abstraction. The next steps in AI research are no longer about teaching a machine to understand that ‘he’ refers to ‘Almanzo’ etc.; they are about ensuring the machine’s reasoning is robust, consistent, and grounded in factual truth rather than just plausible statistics. Arguably the new challenge is not literacy, but discernment, which leads us back to the work of Marcus and Davis for sure.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c6dc54103274" width="1" height="1" alt=""><hr><p><a href="https://medium.com/electronic-life/beyond-functionally-illiterate-ai-revisiting-rebooting-ai-c6dc54103274">Beyond ‘Functionally Illiterate’ AI: Revisiting ‘Rebooting AI’</a> was originally published in <a href="https://medium.com/electronic-life">Electronic Life</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Quantum AI: Research Agenda]]></title>
            <link>https://medium.com/electronic-life/quantum-ai-research-agenda-07b7445c2ffe?source=rss-bcc894f2ddf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/07b7445c2ffe</guid>
            <category><![CDATA[brain]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[quantum-computing]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Sunil Manghani]]></dc:creator>
            <pubDate>Sun, 31 Aug 2025 16:38:13 GMT</pubDate>
            <atom:updated>2025-08-31T21:45:38.549Z</atom:updated>
            <content:encoded><![CDATA[<p><em>This paper outlines a series of questions at the intersection of </em><strong><em>quantum computing</em></strong><em> and </em><strong><em>artificial intelligence (AI)</em></strong><em>. While not exhaustive, it highlights five research trajectories: Efficiency and Energy; Structure and Generalization; Information and Communication; Memory and Correlation; and Brains, Minds, and the Quantum Question.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*z0VRJOVKT-TWvVMRmqrluw.png" /></figure><p>The first quantum revolution (early 20th century) redefined our understanding of matter and energy. The second quantum revolution (from 1995 onward) sought to engineer quantum effects into practical technologies, such as atomic clocks, superconductors, quantum simulators, and quantum computers. These developments have already reshaped everyday life: the precision of atomic clocks underpins GPS navigation; superconducting devices are used in MRI machines; and quantum cryptography pilots secure communication networks.</p><p>Parallel to this, the past decade has seen a dramatic renewal in artificial intelligence research. Deep learning models, scaled to unprecedented size, have yielded emergent capabilities that defy classical intuitions about generalization. AI is now embedded in translation tools, voice assistants, medical imaging diagnostics, and protein-folding prediction. Yet, despite such achievements, the much-touted prospect of Artificial General Intelligence (and associated ideas of world models and machine reasoning) remains contested. A live debate centres on the significance of scaling: whether dramatically increasing the amount and variety of training data can enable exponential improvement. On the one hand, scaling appears to open up possibilities for discovering <em>structures of information</em> that go beyond human comprehension. On the other hand, it raises sobering questions about the sheer amount of energy required to emulate acts of learning and intelligence that we witness among young children, who seem to effortlessly learn about and engage with the world around them.</p><p><em>Can quantum computing combine with AI to lead towards significant new research trajectories in response to these considerations?</em></p><p>Quantum and AI are each ‘revolutions’ in their own right. Their convergence is likely to demand a new language of inquiry. A new research agenda might consider questions for the development of AI around efficiency, structure, information, and memory. Equally, AI can relate back to how we question the quantum basis of cognition, the dimensions of information and the very ‘fabric’ of world models. All such questions prompt broader interdisciplinary stakes: the ecological cost of intelligence, the ontology of information, the physics of learning, and the ethics of distributed systems.</p><h3><strong>Questions Worth Asking</strong></h3><p>The challenge right now is to articulate questions worth asking. As history shows, discovery in science is not simply about new machines or methods, but about the conceptual frameworks that allow us to propose ways forward. The following five themes are intended to offer new, thought-provoking trajectories:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/237/1*suAFinOt8FRfeR_6t_mYlA.jpeg" /></figure><h4><strong>Efficiency and Energy</strong></h4><p>Quantum computing promises exponential speed on certain classes of problems, not because it can store more information, but because it can exploit superposition and interference to process structured problems more efficiently. Yet the energy demands of AI are already staggering. Training a frontier-scale model can consume gigawatt-hours of electricity, equivalent to powering thousands of homes for a year. By contrast, inference (the process of running a trained model on new data) typically uses far less energy per instance, though at scale (billions of queries per day) it too becomes significant.</p><p>This raises the question: if quantum computing has a role in AI, is it more likely to transform training, by accelerating search through vast parameter spaces and lowering the cost of optimization? Or could it alter inference, by offering faster or more efficient pattern recognition once a model is built? The answer matters, because the energy profiles of training and inference are quite different: one is about the <em>cost of creating intelligence</em>, the other about the <em>cost of applying it</em>.</p><p>At the same time, quantum hardware is not energetically free. Supercooling, vacuum chambers, and redundancy through error correction all consume enormous resources, and could offset (or even exceed) the potential savings in computation.</p><blockquote><strong><em>Questions to ask:</em></strong></blockquote><blockquote><em>Can quantum processing deliver net reductions in the energy cost of training AI models, or will it simply shift the ecological burden into new physical infrastructures?</em></blockquote><blockquote><em>Might quantum inference, if achievable, enable a qualitatively different mode of energy-efficient reasoning, or is inference inherently a classical problem?</em></blockquote><blockquote><em>How should we calculate the full energy economics of intelligence across classical and quantum paradigms, distinguishing between training and inference?</em></blockquote><blockquote><em>Could energy-efficient AI require rethinking not only the hardware (classical vs. quantum) but also the very kinds of problems we ask machines to solve?</em></blockquote><p>NB. While today’s quantum hardware carries heavy energy overheads, there are prospects that <em>high-temperature superconductors</em> could radically alter this balance. If stable and scalable forms are achieved, they would allow quantum systems to operate with far less cooling and infrastructure, a development that would be truly transformative for both AI and quantum computing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/242/1*_1QcxW9oURAFrENUaQ2GYA.jpeg" /></figure><h4><strong>Structure and Generalization</strong></h4><p>Quantum computers work best when problems have hidden structure (e.g. patterns like repetition, symmetry, or wave-like behaviour). Famous examples include Shor’s algorithm, which cracks large numbers by uncovering hidden periodicity (a repeating cycle); and Grover’s algorithm, which speeds up searches by using interference patterns to highlight the right answer. By contrast, many AI tasks (e.g. speech recognition, language translation, image analysis) deal with messy, high-dimensional data that does not obviously follow neat mathematical patterns.</p><p>Yet AI sometimes surprises us. In experiments at OpenAI (in 2020) on ‘modulo-97 addition’, a neural network at first behaved like a rote memoriser, simply storing examples. But after extended training, it abruptly ‘clicked’: the model discovered that the problem could be represented as movement around a circle, a structure invisible to casual human inspection. Researchers call this phenomenon <em>grokking</em>, being a sudden shift from memorisation to generalisation. Similar effects appear in today’s large language models, which unexpectedly demonstrate analogy-making or compositional reasoning. In these cases, AI systems seem able to extract hidden regularities from apparent noise.</p><blockquote><strong><em>Questions to ask:</em></strong></blockquote><blockquote><em>Could AI’s capacity to ‘grok’ hidden structures serve as a </em>pre-processor for quantum computation<em>, re-casting messy empirical data into symmetry-friendly forms?</em></blockquote><blockquote><em>Do emergent AI phenomena such as grokking resemble phase transitions in physics, and if so, can this analogy be formalized?</em></blockquote><blockquote><em>Would a convergence of AI and quantum computing lead us toward a future of smaller, structure-aware models (closer to Shor’s exploitation of periodicity) rather than ever-larger, brute-force systems?</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/290/1*mNVlQ-evP0TQM2jaMGbWTQ.jpeg" /></figure><h4><strong>Information and Communication</strong></h4><p>In classical computing, secure communication relies on mathematical difficulty (e.g. the near-impossibility of factoring very large numbers to prevent hacking). Quantum communication works differently. Its security is guaranteed by <em>the laws of physics</em>: entanglement ensures that only two linked parties can share a secret key, while the no-cloning theorem means that quantum information cannot be copied without leaving a trace. If an eavesdropper tries to intercept a quantum key, the disturbance is immediately detectable.</p><p>Quantum information theory also shifts how we think about information itself. A quantum state is not a literal description of reality (as we typically think of it), but a description of the information we hold about reality, compressed into probabilities. This reframing suggests that ‘information’ is not just an abstract mathematical concept, but a physical quantity that behaves according to the rules of entropy.</p><p>For AI, this raises intriguing possibilities. As systems become more distributed (e.g. as federated learning across millions of devices, or swarms of AI agents working together) quantum communication could offer entirely new ways of coordinating and correlating information. Rather than just sending messages across a network, entangled systems might share correlations that cannot exist in classical communication.</p><blockquote><strong><em>Questions to ask:</em></strong></blockquote><blockquote><em>If AI systems increasingly function as distributed and multi-agent, what new forms of coordination or collective intelligence might quantum communication enable?</em></blockquote><blockquote><em>Could entanglement-based communication yield correlations between AI systems that classical networks can never replicate?</em></blockquote><blockquote><em>Does treating information as a physical quantity bound to entropy change how we understand data, learning, and memory in AI?</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/320/1*mHr7gkwmkLtPC52WGj_UNQ.jpeg" /></figure><h4><strong>Memory and Correlation</strong></h4><p>Classical memory stores information by duplication: the same bit pattern can be copied and retrieved across devices without loss. Quantum memory works differently. Here, information is not stored in individual qubits but in the <em>correlations</em> between them. A single logical qubit may require thousands of physical qubits, with the information encoded across their entangled state. The resilience of the memory comes not from duplication but from distributed correlation.</p><p>This idea resonates with Hopfield networks in classical AI. A Hopfield net is an associative memory: it stores information not in a single unit, but in the overall pattern of connections. When presented with partial or noisy data, the network ‘settles’ into a stable configuration that corresponds to one of its stored memories. The system can be understood as moving through an <em>energy landscape</em>, where each memory is a valley (an attractor state), and the network dynamics guide it downhill into the nearest stable state.</p><p>Modern AI architectures have built on and departed from this paradigm. Convolutional neural networks (CNNs), pioneered by Yann LeCun in the late 1980s, introduced a way to encode local correlations in data (e.g. detecting edges, textures, and patterns through layered hierarchies of filters). While not associative memories in the Hopfield sense, CNNs excel at exploiting spatial structure (e.g., in vision). Their success shows how embedding assumptions about structure into the architecture can vastly improve efficiency.</p><p>LeCun’s more recent work on the Joint Embedding Predictive Architecture (JEPA) aims to move beyond brute-force autoregression (as in today’s large language models). JEPA seeks to learn compact, abstract representations that capture the underlying causal structure of the world, rather than predicting one token at a time. In LeCun’s vision, intelligence depends less on memorizing data streams and more on building world models that allow reasoning, analogy, and prediction.</p><p>If Hopfield networks model memory as energy landscapes, CNNs model it as hierarchical spatial filters, and JEPA models it as structured latent spaces, then a <em>quantum paradigm</em> could extend all three. Superposition might allow a Hopfield-like network to explore multiple attractor states in parallel; entanglement could extend the convolutional principle of local correlation into non-local, global correlations; and quantum representations might offer new ways of encoding abstract, high-dimensional world models of the kind envisioned in JEPA.</p><blockquote><strong><em>Questions to ask:</em></strong></blockquote><blockquote><em>If Hopfield networks frame memory as stable attractor states, how might quantum effects such as superposition and entanglement expand this into multi-state, parallel memory?</em></blockquote><blockquote><em>Could entanglement provide a quantum analogue of convolution, capturing correlations that are non-local yet structured?</em></blockquote><blockquote><em>Might a quantum approach to representation learning yield architectures closer to JEPA’s vision (world models</em><strong><em> </em></strong><em>that encode causal structure) while sidestepping the energy demands of massive autoregressive training?</em></blockquote><blockquote><em>Does quantum memory simply scale up existing AI techniques, or does it transform the very nature of representation and recall in artificial systems?</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/265/1*CMRx3e1XDSmKKJP7kmVhhQ.jpeg" /></figure><h4><strong>Brains, Minds, and the Quantum Question</strong></h4><p>Speculative claims that the brain is itself a quantum computer recur across science and culture. The most famous, Penrose and Hameroff’s ‘orchestrated objective reduction’<strong><em> </em></strong>theory, proposed that consciousness arises from quantum effects in microtubules. Max Tegmark, however, argued forcefully that the brain is far too warm, wet, and noisy for sustained quantum coherence: any superposition would decohere in 1013 seconds, long before neurons could fire. Recent quantum biology research (in photosynthesis, bird navigation, or enzyme catalysis) has shown that life can exploit quantum effects in fleeting, protected ways, but whether this extends to cognition remains controversial.</p><blockquote><strong><em>Questions to ask:</em></strong></blockquote><blockquote><em>Even if the brain is not a quantum computer, can quantum models of probability still illuminate puzzling features of cognition, such as context-dependence, memory interference, or decision paradoxes?</em></blockquote><blockquote><em>If AI discovers structures humans overlook, could it help identify where and how quantum processes might plausibly occur in biological systems?</em></blockquote><blockquote><em>What can we learn from the </em>impossibility<em> of a quantum brain? Does this reinforce the sufficiency of classical emergence, or point toward unexplored protective mechanisms in biology?</em></blockquote><p>Quantum and AI may not simply complement one another, but ask questions of each other. Quantum computing challenges AI to think in terms of structure, correlation, and entropy. AI challenges quantum research by revealing emergent behaviours (such as grokking) that resemble physical transitions and by raising new questions about the nature of intelligence itself. The convergence of these fields does not guarantee artificial general intelligence, nor a quantum theory of mind. What emerges instead is a research agenda: the questions that must be asked if we are to understand intelligence, whether artificial, quantum, or human.</p><h4>Related Reading</h4><ul><li><a href="https://www.livescience.com/technology/computing/scientists-hit-quantum-computer-error-rate-of-0-000015-percent-a-world-record-achievement-that-could-lead-to-smaller-and-faster-machines">Scientists hit quantum computer error rate of 0.000015% - a world record achievement that could lead to smaller and faster machines</a></li><li><a href="https://www.rudebaguette.com/en/2025/08/microsofts-1-billion-quantum-leap-theyre-playing-god-in-denmark-igniting-fears-of-tech-tyranny-and-global-power-struggles/">Microsoft&#39;s $1 Billion Quantum Leap: &quot;They&#39;re Playing God&quot; in Denmark, Igniting Fears of Tech Tyranny and Global Power Struggles - Rude Baguette</a></li><li><a href="https://www.thebrighterside.news/post/quantum-computing-occurs-naturally-in-the-human-brain-study-finds/">Quantum computing occurs naturally in the human brain, study finds</a></li><li><a href="https://www.romanlipski.com/about">About | Roman Lipski - Art in times of AI and Quantum Computing - Atelier Roman Lipski</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=07b7445c2ffe" width="1" height="1" alt=""><hr><p><a href="https://medium.com/electronic-life/quantum-ai-research-agenda-07b7445c2ffe">Quantum AI: Research Agenda</a> was originally published in <a href="https://medium.com/electronic-life">Electronic Life</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI for People? The Limits of ‘Seemingly Conscious AI’]]></title>
            <link>https://medium.com/electronic-life/ai-for-people-the-limits-of-seemingly-conscious-ai-a264e3c9698c?source=rss-bcc894f2ddf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/a264e3c9698c</guid>
            <category><![CDATA[mustafa-suleyman]]></category>
            <category><![CDATA[klara-and-the-sun]]></category>
            <category><![CDATA[kazuo-ishiguro]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Sunil Manghani]]></dc:creator>
            <pubDate>Wed, 20 Aug 2025 23:00:57 GMT</pubDate>
            <atom:updated>2025-08-21T19:57:45.773Z</atom:updated>
            <content:encoded><![CDATA[<p>Mustafa Suleyman (CEO, Microsoft AI) argues we should build AI for people, not as people. Kazuo Ishiguro’s <em>Klara and the Sun</em> offers a haunting meditation on this tension. If today’s technologies already create inhuman conditions, the real task is to reimagine the humanity we want concurrent with building future AI.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*w0rVWtCGUSzoWOWz4mkVNA.png" /></figure><p>In his article, ‘<a href="https://mustafa-suleyman.ai/seemingly-conscious-ai-is-coming">We must build AI for people; not to be a person</a>’, Mustafa Suleyman (CEO of Microsoft AI), coins the phrase ‘S<em>eemingly Conscious AI</em>’, which he suggests is closer than we think. The argument he makes is that the most pressing issue is not whether machines truly possess consciousness, but the fact that they will increasingly <em>appear</em> to. What matters is not some sort of metaphysics of silicon minds, but the psychology of <em>human users</em>: people will project, believe, and form attachments. For Suleyman, this creates a zone of risk. He urges us to resist the temptation to frame AIs as persons, and instead to design them carefully, <em>for people (</em>as tools, assistants, or companions), without encouraging illusions of human-likeness.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IT27GrzfszfNUELFHdjC3g.png" /></figure><p>In the essay, Suleyman lays out three compelling reasons why Seemingly Conscious AI isn’t just a speculative concern. First, he argues that it is technically within reach: it can be built using existing technologies (with large language model APIs, natural language prompting, tool use, memory, and autonomy), all of which are maturing rapidly and could converge within just a few years. Second, he insists that the debate over whether AI is genuinely conscious misses the point: what matters is that the <em>illusion</em> of consciousness can feel real to users, and this illusion alone can shape public reaction and behaviour. Third, he cautions that convincing simulations could catalyse new risks, ranging from ‘<a href="https://www.bbc.co.uk/news/articles/c24zdel5j18o">AI psychosis</a>’ and attachment to machines, to demands for <a href="https://experiencemachines.substack.com/p/we-should-take-ai-welfare-seriously">AI welfare</a>, rights, or citizenship. Hence, Suleyman advocates for immediate public debate and guardrails to discourage building AI that appears conscious.</p><p>But what is meant here by AI consciousness and why is it important even when it might not exist? Drawing on academic literature, Suleyman defines consciousness through three intertwined components: (1) a ‘subjective experience’ or qualia, what it’s like to have experiences; (2) the ability to access information and use it in future behaviour; and (3) a coherent, self-aware sense of identity that unifies experience over time. Although we cannot directly access another being’s consciousness, we effortlessly infer it in humans; a fundamental part of the theory of mind. This effortless inference, Suleyman argues, makes AI especially precarious: if a system convincingly imitates consciousness, people may project these qualities onto it, regardless of whether they exist. That’s why he sees consciousness not as a technical curiosity, but a psychological and social fault line that demands careful navigation in AI’s development.</p><h3>Klara and the Sun</h3><p>Suleyman provides a lucid, thoughtful argument and actively invites further probing. What would it mean to live alongside entities that we treat as if they were conscious, but know (at least rationally) are not? Kazuo Ishiguro’s novel, <em>Klara and the Sun, </em>offers a pertinent literary ‘thought experiment’ on this point. The novel follows Klara, an ‘AF’ or Artificial Friend, whose existence is defined by her human owners’ needs. Without revealing the ending (of which I would urge anyone who has not read the novel to ‘experience’), it is fair to say that the way Klara is ultimately treated is quietly devastating. The story stages a limit case for Suleyman’s warning: what happens when we invest emotional care in a machine that we might not truly believe capable of reciprocating, and what does this reveal about <em>us</em>?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*m9Fm8HhTwKNWcb_1TWqVGQ.jpeg" /></figure><p>Assuming we agree we should build AI for people, and ‘not to be a person’, there is still a sharper ethical pivot to consider. Before we ask what kind of AI we want to build, we need to ask what kind of people we wish to become. Ishiguro’s novel reminds us that our treatment of artificial others is a mirror of our treatment of human others. To design an AI that merely pleases or consoles us may be to miss a deeper responsibility: to embody the kind of humanity we would want reflected back.</p><p>Suleyman is right to caution against illusions of personhood. The danger of mistaking simulated consciousness for the real thing is not trivial: it risks deepening loneliness, fostering dependency, and inviting manipulation by corporations who would happily monetise our attachments. His call for restraint is a welcome corrective to the hype cycle of <a href="https://www.scientificamerican.com/article/google-engineer-claims-ai-chatbot-is-sentient-why-that-matters/">sentient chatbots</a> and <a href="https://theconversation.com/more-people-are-considering-ai-lovers-and-we-shouldnt-judge-260631">digital lovers</a>. Yet, by focusing on how to build AI responsibly <em>for people</em>, the tendency is to assume a fixed and stable user — to treat the human subject as given, when in fact the emergence of Seemingly Conscious AI will also reshape who we are and how we relate to one another. In other words, the challenge is not only technical or regulatory, but relational and existential. As I argued in ‘<a href="https://medium.com/electronic-life/ai-mirror-stage-or-why-not-to-confuse-ai-with-human-intelligence-3777384542ad">AI Mirror Stage</a>’, we might go further and resist equating AI with human intelligence at all and instead allow it to emerge as a <em>different kind of intelligence</em> — one that develops unevenly, in fits and starts, like a child learning to negotiate its own becoming.</p><p><a href="https://medium.com/electronic-life/ai-mirror-stage-or-why-not-to-confuse-ai-with-human-intelligence-3777384542ad">AI: Mirror Stage (or, why not to confuse AI with human intelligence)</a></p><p>Again, Ishiguro’s <em>Klara and the Sun</em> is instructive, pressing on the limits of Suleyman’s case. Klara’s fate (the unsettling indifference with which she is ultimately treated), reflects not just the ambiguity of machine personhood but the failures of human personhood. To insist that AI should serve people is laudable, but insufficient. The harder question is: <em>what kind of people</em> are we asking these machines to serve? If we design AI for a society already marked by exploitation, inequality, and emotional neglect, then even the most careful guardrails will reproduce those pathologies in digital form. Building AI for people is a noble goal, but unless we first reckon with what it means to be human, we risk repeating Ishiguro’s quiet indictment: machines will not impinge on our humanity so much as reveal its absence.</p><h3>In-Human</h3><p>Suleyman’s intervention is especially striking when set against the backdrop of his own company’s findings. Microsoft’s <a href="https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born"><em>Work Trend Index 2025</em></a> paints a picture of the ‘infinite workday,’ where employees are interrupted on average every two minutes, sending over 50 messages outside of work hours, and attending meetings that stretch across time zones late into the night. In this environment, the drive for AI ‘agents’ is framed as a solution — digital colleagues to take on drudgery, extend capacity, and close the ‘productivity gap’ that human beings alone can no longer sustain.</p><p>The paradox is that the very technologies that promised efficiency (email, instant messaging, cloud collaboration etc.) have helped to create the inhuman conditions of the modern workplace. Employees are not merely overwhelmed by tasks, they are caught in systems that treat human attention as infinitely divisible. If AI is presented as the remedy, we must ask whether it is being built <em>for people</em>, as Suleyman urges, or to compensate for (and further extract from) the inhumanity that digital technologies have already advanced.</p><p>The philosopher Jean-François Lyotard distinguishes two senses of the ‘inhuman’. The first concerns the machinery of systems and technologies that erode our capacity for humanity, reducing people to functions within logics of productivity, efficiency, and acceleration. The second, is more existential, referring to the <em>in-human</em> — what is <em>within</em> us, yet not entirely accessible; an unsettling state that disrupts our certainties. It gestures to infancy and vulnerability, to the shocks of the sublime, to temporal ruptures that resist mastery and unsettle what we think of as the human. Suleyman’s argument largely sidesteps the concern over machines as persons. His point is that we should not imagine AI <em>as</em> people, but rather build tools for people. Yet those very tools can all too easily become the instruments of the first inhuman — i.e. machinic logics that already make work corrosive and inhumanly efficient. But if we pause a little longer to consider how people come to accept Seemingly Conscious AI, as Suleyman proposes, another dimension comes into view. Here we begin to glimpse Lyotard’s second, more unsettling in-human: not a comforting possibility, but a troubling exposure to what unsettles our mastery, to the strangeness and fragility within and beyond us. To tarry a little longer with this problem, rather than resolve it too quickly, gives access to more complex questions about our (in)humanity; questions that Suleyman’s framing helps open, even if he does not pursue them.</p><p>Hence, I do not wish to dismiss Suleyman’s thoughtful call. His essay is a valuable provocation, precisely because it opens up debate; not about consciousness per se, but about the ethical and social infrastructures in which AI is embedded. A question remains as to whether companies like Microsoft (and the wider tech sector) can genuinely attend to the conditions they have created before ‘bolting-on’ the next solution. But more than that, we need to remain alert to conditions of our own in-humanity — the alienation, vulnerability, and desires that unsettle what we take the human to be. We need to ensure AI is not merely proposed as a remedy to the endless workday. It should help us advance our very foundations — of how we live, how we care, how we work.</p><p>If Suleyman is right that the coming age of Seemingly Conscious AI demands restraint, then the corollary is clear: we cannot outsource the task of becoming human to the machines we design. The choice is not simply between AI as person or AI for people. It is whether we are willing to reimagine the conditions of life that can allow technologies to be ameliorative in the first place. Until then, the danger is that AI will not save us from the inhuman, it will simply automate it.</p><p>A final word goes to Klara, who stands in for a coming community of artificial agents. What matters is not simply what we project onto AI, but the kind of human we reveal ourselves to be when increasingly caught-up in the nexus of AI’s burgeoning gaze.</p><ul><li><a href="https://medium.com/@gp2030/the-coming-wave-surfing-toward-the-apocalypse-with-mustafa-suleyman-22504cb11f58">The Coming Wave: Surfing Toward the Apocalypse with Mustafa Suleyman</a></li><li><a href="https://medium.com/@ai_93276/the-taming-of-the-wave-navigating-ais-surging-tide-with-suleyman-s-compass-3fd7becd2b11">🌊 The Taming of the Wave: Navigating AI’s Surging Tide with Suleyman’s Compass</a></li><li><a href="https://medium.com/@julio.pessan.pessan/the-quiet-revolution-how-mustafa-suleyman-sees-ai-transforming-work-and-why-were-not-ready-8a930ef2bee2">The Quiet Revolution: How Mustafa Suleyman Sees AI Transforming Work — And Why We’re Not Ready</a></li><li><a href="https://medium.com/@digitaldrew/mustafa-suleymans-forecast-of-reaching-self-improving-ai-within-3-5-years-is-rather-optimistic-b21399874769">Mustafa Suleyman’s forecast of reaching self-improving AI within 3–5 years is rather optimistic</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a264e3c9698c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/electronic-life/ai-for-people-the-limits-of-seemingly-conscious-ai-a264e3c9698c">AI for People? The Limits of ‘Seemingly Conscious AI’</a> was originally published in <a href="https://medium.com/electronic-life">Electronic Life</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ready for the Quantum Revolution?]]></title>
            <link>https://medium.com/electronic-life/ready-for-the-quantum-revolution-invisibility-cloaks-are-just-the-beginning-1253e3a6ac7b?source=rss-bcc894f2ddf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/1253e3a6ac7b</guid>
            <category><![CDATA[quantum-revolution]]></category>
            <category><![CDATA[quantum-physics]]></category>
            <category><![CDATA[science-fiction]]></category>
            <category><![CDATA[future-technology]]></category>
            <category><![CDATA[metamaterials]]></category>
            <dc:creator><![CDATA[Sunil Manghani]]></dc:creator>
            <pubDate>Sun, 01 Jun 2025 01:47:18 GMT</pubDate>
            <atom:updated>2025-06-01T02:03:22.746Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A second quantum revolution is here (almost), with mind-bending technologies like time mirrors and invisibility cloaks. We may never fully ‘understand’ the physics, but learning to </em><strong><em>feel its effects</em></strong><em> is key to preparing for our future.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZbT1s9IXxYZnwI0MEEi6Ew.png" /><figcaption>GenAI rendering (in style of romance novel) from <a href="https://static.standard.co.uk/2025/05/28/16/18/Author-Photo-(c)-Siska-Vandecasteele.jpg?trim=744,0,1005,0&amp;quality=75&amp;auto=webp&amp;width=1000">original photograph of Frank Verstraete &amp; Céline Broeckaert</a>, authors of <em>Why Nobody Understands Quantum Physics.</em></figcaption></figure><p><strong>‘Stop trying to understand it, you just have to feel it’</strong>, physicist Frank Verstraete suggested to Céline Broeckaert, when collaborating on writing a book about quantum physics. The book, <a href="https://www.flandersliterature.be/books-and-authors/book/why-nobody-understands-quantum-and-everybody-needs-to-know-something-about#:~:text=A%20delightful%20book%20for%20all,its%20impact%20on%20our%20lives.&amp;text=Vividly%20and%20playfully%2C%20a%20scientist,who%20made%20the%20theory%20great."><em>Why Nobody Understands Quantum Physics (And Everyone Needs to Know Something About It)</em></a>, is an unconventional guide to the quantum world — one that helps us feel the implications of quantum physics more than fully having to comprehend it. Verstraete, a leading quantum scientist, admits he initially ‘<a href="https://www.standard.co.uk/culture/books/why-nobody-understands-quantum-physics-frank-verstraete-celine-broeckaert-b1230150.html">completely failed</a>’ at writing a popular physics book because he tried to make readers understand the theory; he eventually realised ‘that is not what people want… They want to get a feeling about it.’ By teaming up with Broeckaert — a playwright astute with relatable language and human stories — the quantum realm has been made more accessible and even entertaining, infusing the science with art, history, and humour. (Their partnership was so fruitful it became personal: the writing project ending up in marriage!).</p><h4>The Second Quantum Revolution</h4><p>So, why bother ‘feeling’ quantum physics? A new wave of quantum technology is fast emerging, poised to transform everyday life. Yet, a general lack of understanding of the physics typically renders any conversation moot. I want to suggest we take this subject more seriously — not academically (that’s for the physicists!), but more generally: how is this really going to effect our lives; are we prepared for a whole new paradigm?</p><p>In fact, it is fair to say, a <em>second </em>quantum revolution is coming. Think ultra-efficient quantum computers, advanced superconductors for clean energy, unhackable quantum communications, and more. Already, quantum physics underlies much of our current tech: the transistors in your smartphone, the lasers in your DVD player, MRI scanners in hospitals — all are affordances of quantum mechanics. Yet, much of our public discourse has barely even caught up with this first quantum revolution. Now, as <a href="https://medium.com/electronic-life/google-just-proved-we-live-in-a-multiverse-no-really-1600a624fe8e">tech giants pour billions into quantum R&amp;D</a>, an even <em>weirder</em> second revolution is unfolding, and it’s vital for non-scientists to tune in.</p><p>Consider invisibility cloaks, a concept straight out of Harry Potter that is <em>already</em> inching toward reality. In 2006, researchers at Duke University stunned the world by creating the first ‘<a href="https://stories.duke.edu/beyond-materials-from-invisibility-cloaks-to-satellite-communications#:~:text=The%20world%E2%80%99s%20first%20invisibility%20cloak,had%20been%20in%20the%20way">metamaterial invisibility cloak</a>’. It was essentially a set of concentric rings that bent electromagnetic waves around a small object, hiding it from detection. The effect worked (albeit only for a narrow band of microwaves, and only from certain angles). If a real-life cloaking device is possible, even in primitive form, what other sci-fi ideas might quantum-era science bring to life? The answer seems to be: plenty.</p><p>Invisibility cloaks are just the beginning, though they came another step closer with a recent breakthrough that goes beyond bending light in space — <em>it involves flipping events in time, </em>or at least ‘<a href="https://www.nature.com/articles/s41567-023-01975-y">a uniform inversion of the temporal evolution of a signal</a>.’ To appreciate how wild this is, forget the image of your face in a normal mirror and imagine looking into a mirror that shows your back instead. Sounds impossible, yet this is precisely the kind of phenomenon scientists have now demonstrated, having finally <em>empirically</em> proven the strange theory of ‘time reflection’.</p><h4>Time Reflection: When Time Turns Mirror-like</h4><p>Instead of reflecting your appearance in space, a time mirror reflects the sequence of events in time. In such a mirror, the last part of a signal is reflected first — a person would bizarrely see their back instead of their face, and any sound would echo backward like a rewound tape.</p><p>Time reflection was a purely theoretical idea for decades. Since the 1970s, Physicists have speculated that if you could change a material abruptly in time, you might force a wave to reflect backwards in time rather than in space. In a normal mirror (a spatial reflection), light waves hit a surface and bounce back, giving you an immediate image of your face. In a time mirror (a temporal reflection), the whole medium carrying the light or sound is suddenly altered at once, causing the wave’s timeline to invert. The weird result is an echo in reverse. It would be like looking into a mirror and spotting your back instead of your face.</p><p>For over 50 years this idea remained hypothetical — largely because it’s fiendishly hard to pull off. To create a time reflection, you must change the properties of a material extremely fast and uniformly. Imagine trying to switch an entire pool of water into gelatin instantly to reflect a ripple backward — it sounds absurdly energy-intensive. Until recently, no experiment could change an electromagnetic medium quick enough, at a large enough contrast, to produce a time flip. Now, however, a team at <a href="https://asrc.gc.cuny.edu/headlines/2023/03/scientists-demonstrate-time-reflection-of-electromagnetic-waves-in-a-groundbreaking-experiment/">CUNY’s Advanced Science Research Center</a> in New York has done exactly that, achieving the first clear demonstration of a ‘time mirror’ for electromagnetic waves.</p><p>What does a time-reflected signal actually look and sound like? In the CUNY experiment, they observed a strange inversion: the tail end of the original signal flipped around to become the front of a new, reflected signal. It’s as if you shouted ‘hello’, and the system replied with ‘olleh’. ‘<a href="https://asrc.gc.cuny.edu/headlines/2023/03/scientists-demonstrate-time-reflection-of-electromagnetic-waves-in-a-groundbreaking-experiment/">The last part of the signal is reflected first</a>,’ the researchers explain, meaning cause and effect swap order in the echo. Hence, if you were to look into such a <em>time</em> mirror, your reflection would be flipped — you’d see the back of your head instead of your face. In an acoustic version, you’d hear something like a tape playback in reverse. In fact, all the frequencies of the wave get transformed. Colors would suddenly change hue (a red light would turn green, orange would turn blue, etc.) as the light’s wavelength is abruptly shifted. One could say a time mirror gives you a funhouse reflection not in space, but in time — scrambled in order and shifted in pitch.</p><p>This temporal reflection phenomenon, once deemed ‘too wild’ even for quantum theorists, has now been captured in the lab. <a href="https://www.nature.com/articles/s41567-023-01975-y">The study, published in <em>Nature Physics </em>in 2023</a>, was the first experimental evidence that time mirrors are real. It took clever engineering and a lot of precisely synchronised switching, but it worked. The achievement is more than just a physics parlour trick — it opens up a new realm of possibilities for manipulating waves.</p><h4>New Freedoms for ‘Sound &amp; Vision’</h4><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FpeiJ9_1tH_w%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DpeiJ9_1tH_w&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FpeiJ9_1tH_w%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/95c5e0c8d56ff9142889f4aabb537b29/href">https://medium.com/media/95c5e0c8d56ff9142889f4aabb537b29/href</a></iframe><p><strong>‘Waiting for the gift of sound and vision…’</strong> When David Bowie first sung this laconic line in his pop classic, <em>Sound and Vision</em>, from 1977, he likely wasn’t thinking of quantum metamaterials or time-bending physics. Yet, the line now feels oddly prophetic. The breakthrough in time reflection doesn’t just offer a clever reversal of signals; it suggests a future in which sound and vision become programmable, fluid, and manipulable in ways that defy classical limits. Where once we sculpted waves in space with mirrors and lenses, we may soon shape them in time — opening a new dimension for technology, art, and communication alike.</p><p>Seeing a time mirror in action isn’t just ‘wild’’, it is significant. It gives scientists a new handle on the fundamental behavior of waves. We’ve long been able to reflect, refract, and filter waves using spatial structures (mirrors, lenses, antennas, etc.). Now we’re learning to do the same in the time dimension. The CUNY team’s discovery ‘<a href="https://asrc.gc.cuny.edu/headlines/2023/03/scientists-demonstrate-time-reflection-of-electromagnetic-waves-in-a-groundbreaking-experiment/#:~:text=The%20discovery%20establishes%20the%20foundations,wireless%20communications%20and%20optical%20computing">establishes the foundations for revolutionary applications in wireless communications and optical computing</a>’. It shows that by engineering ‘temporal interfaces’ (sudden shifts in a medium in time), we gain ‘new degrees of freedom for extreme wave control.’</p><p>Imagine being able to dial up or down the frequency of a light wave on the fly, or reverse an information-carrying signal to retrieve data in novel ways. Time reflections mix up a wave’s order and frequency content in one swoop. This could enable exotic new devices. <a href="https://asrc.gc.cuny.edu/headlines/2023/03/scientists-demonstrate-time-reflection-of-electromagnetic-waves-in-a-groundbreaking-experiment/">Co-first author Shixiong Yin, a graduate student at CUNY ASRC, explains</a> we could combine multiple time interfaces to create a ‘time resonant cavity’, essentially trapping a wave between time-mirrors the way an optical cavity traps light between two mirrors. Such a device would selectively filter signals in time — a brand-new way to filter or store information. It could mean producing ‘electromagnetic time crystals’ — structures with patterns that repeat in time rather than space — opening new ways to enhance wave-matter interactions.</p><p>Communications and computing are most likely to reap immediate benefits:</p><blockquote><em>A wave that can jump to a new frequency and then rewind might open new possibilities for data transmission at different ranges of the spectrum. It could also reshape how certain sensors and imaging systems are designed. (</em><a href="https://www.earth.com/news/physicists-confirm-the-incredible-existence-of-time-mirrors/"><em>Eric Ralls</em></a><em>)</em></blockquote><p>Wave-based or ‘time computing’ is another tantalizing idea: using time flips to process signals in ways normal electronics can’t, potentially leading to small, low-energy, wave-based computers. Rather than electrons switching transistors on a chip, future processors might use bursts of light or sound that interfere with themselves in time, performing computations with far less energy. This represents a whole new toolbox for engineers — time lenses, time mirrors, and temporal filters that could do for signals what spatial optics did for light. The freedom to control wave behaviour in time as well as space could spark inventions we haven’t even dreamed of yet.</p><h4>From Sci-Fi Dreams to Everyday Life</h4><p>What might those inventions look like? It’s here that we can allow ourselves to speculate — thoughtfully, but with full imagination. History shows that today’s science fiction often becomes tomorrow’s reality, especially in the quantum age. (After all, invisibility cloaks went from magical fiction to a published experiment in less than a generation.) Now that we have a way to flip waves in time, we can dream up all kinds of novel gadgets and experiences. For instance:</p><p><strong>Recap Headphones</strong>: Imagine a pair of high-tech earbuds that employ time reflection to let you <em>rewind</em> what you just heard. Missed what someone said? These headphones could momentarily reverse the incoming sound waves, replaying the last few seconds of conversation backwards (or slowed down) for you. It would be like an auditory time-turner, giving your ears a second chance by literally reflecting sound in time.</p><p><strong>Dynamic Architecture:</strong> Buildings of the future might not be static four-dimensional structures — they could have ‘time-responsive designs’. Envision a concert hall or open-plan office that can reconfigure its acoustics on the fly using time mirrors. Walls could ‘bounce’ sound or light waves back in time to cancel echoes or redirect energy, making a space feel physically different at different moments. A room might even shift its layout in time, appearing to morph or teleport sections of itself as wave patterns are time-flipped to create the illusion of motion where there is none. Architecture would no longer be limited to form and space; it would have a temporal dimension, dynamically adapting via programmed time reflections.</p><p><strong>New Temporal Media: </strong>Artists could tap into time manipulation to create mesmerising new art forms. Think of quantum installations where the artwork changes based on temporal feedback loops — paintings that reveal different images when viewed with time-reversed light, or music that remixes itself when fed through a time lens. You might attend a dance performance where the choreography incorporates live ‘time echoes’ — movements captured by sensors and reflected a few seconds later on stage, so dancers interact with past versions of themselves. Part of the long-tail of art, such experiences would further challenge our perceptions, but here invite us to feel the flow of time in a whole new way.</p><p>These examples sound fanciful, but so did lasers and holograms before we invented them. Science fiction has long imagined the ability to manipulate time, from H.G. Wells’ <em>Time Machine</em> to Marvel’s time-manipulating superheroes. Now, science is inching toward making some of those imaginings feasible on the wave and signal level. Speculative thinking is not fantasy — it’s a bridge between imagination and innovation. The very concept of a time mirror might have languished in equations if not for visionary minds willing to play with seemingly impossible ideas. Which brings us back to Verstraete and Broeckaert’s <a href="https://www.flandersliterature.be/books-and-authors/book/why-nobody-understands-quantum-and-everybody-needs-to-know-something-about#:~:text=A%20delightful%20book%20for%20all,its%20impact%20on%20our%20lives.&amp;text=Vividly%20and%20playfully%2C%20a%20scientist,who%20made%20the%20theory%20great."><em>Why Nobody Understands Quantum Physics</em></a><em> </em>— to their underlying message: <strong>Stop trying to understand it, just feel it.</strong></p><p>We should embrace the wonder of quantum physics. In doing so, we can become active participants. The quantum revolution promises to turn some of our wildest imaginings into everyday experiences — and if we’re emotionally and intellectually ready to trust in that shift, we’ll recognize the future when it arrives, even if it looks as strange as seeing the back of our head when looking in the mirror.</p><ul><li><a href="https://medium.com/electronic-life/quantum-life-but-not-as-we-know-it-cc57705e06fd">Quantum Life, but not as we know it</a></li><li><a href="https://medium.com/electronic-life/google-just-proved-we-live-in-a-multiverse-no-really-1600a624fe8e">Google Just Proved We Live in a Multiverse (No, Really)</a></li><li><a href="https://csferrie.medium.com/nobody-understands-quantum-physicists-bdf1b9180dac">Nobody understands quantum physicists</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1253e3a6ac7b" width="1" height="1" alt=""><hr><p><a href="https://medium.com/electronic-life/ready-for-the-quantum-revolution-invisibility-cloaks-are-just-the-beginning-1253e3a6ac7b">Ready for the Quantum Revolution?</a> was originally published in <a href="https://medium.com/electronic-life">Electronic Life</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[WhatsApped?]]></title>
            <link>https://medium.com/electronic-life/whatsapped-8134199b3d30?source=rss-bcc894f2ddf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/8134199b3d30</guid>
            <category><![CDATA[organizational-culture]]></category>
            <category><![CDATA[critical-thinking]]></category>
            <category><![CDATA[scholarly-communication]]></category>
            <category><![CDATA[attention-economy]]></category>
            <category><![CDATA[whatsapp]]></category>
            <dc:creator><![CDATA[Sunil Manghani]]></dc:creator>
            <pubDate>Sat, 31 May 2025 21:47:26 GMT</pubDate>
            <atom:updated>2025-06-01T02:51:33.994Z</atom:updated>
            <content:encoded><![CDATA[<h4>A Scholarly Apparatus Cast Adrift</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MGugZ0v-yd_h3at1OcyMJA.png" /></figure><blockquote><em>Ils avaient tout lu, tout appris, tout oublié<br>- Flaubert, </em>Bouvard et Pécuchet</blockquote><blockquote><em>Le dérive n’est pas une déviation, mais une sorte de pesanteur naturelle<br></em>- <em>Roland Barthes,</em> Le Plaisir du Texte</blockquote><p><strong><em>From emoji pings to AI-drafted replies, have messaging apps altered how organisations function, how work is conducted and even how scholars think?</em></strong></p><p>Whether it’s the fact of a military strike conducted by a Signal group chat, or a Covid Inquiry requiring the sequestering of government WhatsApp messages, what is clear is how messaging has become not just a dominant, but even a default mechanism for how we run our lives — personally and professionally. And this includes its role as part of academia; the effect of new modes and forms upon the scholarly apparatus.</p><p>The name ‘WhatsApp’ echoes the meme-phrase ‘whassup?’, popularised by a Budweiser ad campaign that ran from 1999 to 2002. The adverts, comically slurring ‘what’s up?’, tapped into a zeitgeist of casual slackness — a kind of performative laziness — that was at once satirical and celebratory. ‘Whassup?’ became a badge of ironic detachment, of not taking things too seriously. The phrase re-emerged in its modified form of ‘WhatsApp’, i.e. as a global messaging platform. The name, WhatsApp, suggests not just being connected, but doing so casually, with a shrug. What parades merely as a tool for coordination and updates, is arguably part of a wider trend, ushering in a culture of deferral and disorganisation, including a diminished scholarly apparatus.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FJJmqCKtJnxM%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DJJmqCKtJnxM&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FJJmqCKtJnxM%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/099135130fe53c62933628277e710c91/href">https://medium.com/media/099135130fe53c62933628277e710c91/href</a></iframe><p>Platforms like WhatsApp, Slack, and Teams connect billions of users — WhatsApp alone claims roughly 3 billion monthly users as of 2025. The ubiquitous ‘immediacy’ of messaging has been widely adopted for its ease and efficiency. But, equally, as a ‘mode’ <em>defines</em> how we interact. My contention is that the culture of instant messaging (which extends beyond simply using specific apps) fragments our attention, flattens our writing and reading habits, and transforms ‘knowledge’ into a disjointed stream of contextless data. Immediacy is ascendant, but at the cost of depth and deliberation.</p><h4><strong>New Communication Norms</strong></h4><p>In academia, communication norms have shifted distinctly beyond what we might have originally envisaged with the arrival of email. Lengthy, carefully crafted emails or memos are often bypassed in favour of terse chat messages peppered with emojis and abbreviations. Conversations that once unfolded in formal correspondence or in-person meetings now take place through quick electronic exchanges. The brevity and informality can ‘seem’ to improve efficiency — information ‘moves’ quickly and colleagues are perpetually reachable — but what are we actually saying in our exchanges? Messaging changes the tone and substance of professional discourse.</p><p>Max Weber’s theory of rationalisation relates well to this shift. Modern organizations are often framed around efficiency, calculability, and control (internal staff surveys make demands as such!). Weber was interested in how new means of communication become ‘pacemakers’ of bureaucratisation.</p><blockquote><em>The management of the modern office is based upon written documents (‘the files’), which are preserved in their original or draft form… This is the characteristic of bureaucratic administration. The body of officials actively engaged in a ‘public’ office… is separated from the means of communication and production. <br></em>– Max Weber,<em> Economy and Society</em></blockquote><p>In line with instrumental rationality, instant messaging further optimizes (and separates) communication for speed and convenience. A quick Slack ping or WhatsApp text is an efficient <em>means to an end</em> — whether that end is scheduling a meeting or getting a student’s question answered. An instrumental logic, however, sidelines value-rational considerations, i.e. whether the communication is meaningful or appropriate. In other words, the how (fast, efficient messaging) eclipses the why (the substantive purpose of the exchange). The result is a new default style: <em>informal, always-on, and expectant of reply</em>. In part this suits scholars who have always operated at odd hours, with their own rhythms. But this preference leads to a hegemony for digital practices more generally and not always suiting the real value of being in ‘touch’. These new messaging norms also relate to what Dan Davies refers to as ‘accountability sinks’ — which, institutionally, are mechanisms to absorb negative sentiment and to obscure accountability. Academics can get animated about the problems of the neo-liberal university, for example, yet they end up utilising informal mechanisms to ‘get by’ — in doing so, accountability remains unaccounted for and the individual has again acted as a neo-liberal subject.</p><p>A practical problem with messaging (over emails for example) is that a message older than 5 minutes might as well not exist. This normalization of <em>instant response</em> pressures everyone to check messages incessantly, feeding a ‘cult of busyness’ and an anxiety that if you’re not online, you’re out of the loop. The email itself has changed in style, less now a digital form of ‘letter’ and more typically a tweet-like message. At first glance, these new norms appear harmless or even liberating — a more casual, democratic communication style. Brecht long ago dreamed of a ‘two-way’ radio, never suspecting the eventual ugliness of social media. As has played out, complex or sensitive discussions are now flattened into short texts. While interruptions become constant. We can talk of <em>cognitive fatigue</em>, which in turn relates to a reduction in attention to meaningful interactions.</p><h4><strong>The Erosion of Depth</strong></h4><p>Messaging has also begun to reshape our reading and writing habits. Accustomed to 280-character tweets, one-line texts, and fragmentary threads, many now approach written content in a <em>scanning, skimming</em> mode rather than deep, sustained reading. In literary terms, the deep, devotional practice of ‘vertical’ reading (associated with scrolls) — engaging hermeneutically with a text linearly and attentively — is being supplanted by ‘horizontal’ reading, a rapid zigzag across the surface of texts . We hop from message to message, article to article, without fully absorbing any one thread of thought. This habit may be suitable for triaging an overflowing inbox or chat feed, but it undermines the capacity for close reading that scholarly work requires. Complex journal articles or dense theoretical arguments cannot be digested in the same bite-sized way as a WhatsApp chat, yet our digitised practices often attempt to treat them as such, leading to a loss of comprehension and patience. Alternatively, we might reverse the vertical/horizontal analogy: the older horizontal architecture of the book, whereby you can move through pages, holding a finger upon an earlier page, has been replaced by vertical scrolling on screens and media feeds. The vertical (virtual) space of reading/writing can have the effect of long rambling texts; little attention to structured argument; and often poor formatting.</p><p>Academic writing is traditionally deliberative and structured, rewarding those who can sustain ideas over many pages. But as daily writing shifts toward texting and chat, academic ‘attention’ has shifted. Sentences and paragraphs become shorter (made more suitable for screen reading); arguments arrive in staccato bursts rather than carefully nested paragraphs. We risk producing scholarly prose that reads like an extended series of tweets — efficient in getting a point across, but lacking the exploratory depth that comes from wrestling with ideas in long form.</p><p>Theodor Adorno remains prescient. He warned that mass culture encourages a kind of standardised, shallow engagement with content. The culture industry supposedly robbing people of their imagination. In the context of messaging, we are often fed pre-digested snippets of information and autocorrects; and frequently — certainly in meetings and presentations — there is an explicit or implicit suggestion that more is available on a webpage ‘somewhere’. Snippet-styled communication requires minimal mental effort, much like the formulaic products of the culture industry (the further information on a webpage ‘somewhere’ takes on the task of reading for you; but at least you know it is there if ever you did get around to it!). As Adorno would argue, we risk a regression of critical thought; a slide into comfortable, unchallenging routines of consuming <em>information</em>.</p><h4><strong>Digital Memory</strong></h4><p>If <em>attention</em> is the scholar’s most precious resource, messaging platforms and related digital media are in direct competition. The design of these platforms — buzzing notifications, red dots, endless timelines — is engineered to capture our eyes and minds for as long or at least as often as possible. Bernard Stiegler wrote of an attention economy where digital markets fiercely compete to grab and hold our focus. Every ‘ding’ of a message is a bid for our attention, pulling us away from whatever else we were doing. Over time, this fragmented, multitasking mode erodes our capacity for sustained concentration. Stiegler’s concerns arguably raise specific concerns for a younger generation who have grown up in this new attention economy (and as a result lack experience of alternative strategies).</p><p>Memory, too, is being reshaped (not least by the aforementioned ‘somewhere’ online). With so much information persistently accessible at our fingertips, we rely less on internal recall and more on external storage. Psychologists have dubbed this the ‘Google effect’ or digital amnesia — the tendency to forget information that we know can be easily looked up online. Scholars today might not memorize citations or detailed facts as their predecessors did; why bother, when your chat history or a quick search can retrieve the answer? But this convenience has a downside: our personal memory and knowledge structures grow fragile and shallow. We retain <em>pointers</em> (how to find information) rather than the knowledge itself. The endless chat logs and cloud archives become an external brain of sorts — one that is searchable but not truly integrated into our understanding. Knowledge risks turning into a disjointed stream of contextless data.</p><p>Stiegler’s account of technology as a <em>pharmakon</em> — both poison and cure — is pertinent. Digital platforms aid memory by serving as vast archives. WhatsApp, for instance, can store years of group chat history that one can scroll back through. This is the <em>remedy</em> side of the <em>pharmakon</em>: no important detail ever needs to be truly lost. But the <em>poison</em> is that we come to depend on these external memories, neglecting to cultivate our own. Or more simply, the scale of information soon mitigates against any genuinely useful purpose. Hence, we might argue why AI becomes a solution. I.e. AI is not an innovation for innovations sake, but a <em>response</em> to our lack in a swamp of Big Data (Hayles writes on this explicitly in <em>Bacteria to AI</em>).</p><h4><strong>Scholarly Automation</strong></h4><p>Amid this environment of accelerated, fragmented communication, we now face the rise of automation and AI in scholarly work. As suggested above, the emergence of these new technologies is arguably a direct result of the ‘data condition’ we have created and practice daily.</p><p>Large language models and AI assistants can draft summaries, answer questions, even generate literature reviews. In one sense, this is rationalisation taken to its logical extreme: the automation of intellectual labour. If Weber saw bureaucracy as the hallmark of rationalised society, today’s algorithmic tools represent a new frontier — a bureaucracy of the mind, where procedures of thinking can be formalised and executed by a machine. Weber’s concept of instrumental rationality — pursuing efficiency above all else — is vividly reflected in how AI is being deployed to optimise tasks in academia. Repetitive or time-consuming work (copyediting, data analysis, formatting citations) can be offloaded to AI, as deployed by major publishers for example, so ‘freeing’ humans for other tasks. But what happens when the tasks being optimised are the very acts of reading, writing, editing and reasoning that define scholarship?</p><p>There is a growing tension between automated knowledge and human understanding. A scholar might be tempted to let an AI system do the first pass of reading dozens of articles and extracting key points, essentially outsourcing the act of critical reading. While this yields a quick summary, it bypasses the serendipity and deep comprehension that come from human engagement with texts. Again, Adorno’s warnings of the culture industry ring eerily true: he feared a culture where challenging, critical content is squeezed out by standardised, easy-to-digest output. AI-generated knowledge, based on patterns in existing data, often amplifies the status quo. By design, such systems produce an average of what is already out there — a kind of intellectual Muzak. Adorno argued the culture industry itself functioned to maintain hegemonic order or stasis. Likewise, an over-reliance on AI tools in research potentially leads to formulaic papers and unchallenging, uncritical analyses, since the AI is drawing from prior normalised knowledge. The human scholar’s role in pushing boundaries and asking uncomfortable questions is subdued by the convenience of automated answers. (This view needs to be balanced, however, with the genuine power of AI to aid discovery. Such use of AI is typically more specialised, such as the discover of millions of protein shapes and methods embedded in MRI softwares, without which we would not be able to see the ‘unseeable’ or unlock new knowledge).</p><p>There is a further twist in the tail of our ‘WhatsApp’ conditioning: the automation of communication itself, such as AI drafting our emails or even replying to messages, on our behalf. A witty aside, a thoughtful pause, an individualized choice of words — these are the subtleties through which we think as we write and through which we recognize each other as more than information-processing machines. If messaging becomes not only instant but also auto-generated, it completes the transformation of communication into pure instrumentality. We would then truly be living in Weber’s iron cage, where even our scholarly <em>dialogue</em> is bureaucratised and standardised beyond all enchantment. Weber warned that an overly rationalised society becomes a <em>disenchanted</em> one, drained of creativity and meaning. The spectre of AI-driven scholarship raises the question: will we sacrifice the richly idiosyncratic <em>human</em> aspects of thinking and writing for the sake of speed and efficiency? Or does AI offer some redeeming qualities? Can it be incorporated in meaningful ways, rather than merely a means of automation? What does this all mean for the ‘soul’ of scholarship?</p><h4><strong>Organisational Decay</strong></h4><p>Perhaps the most paradoxical outcome of the messaging era is that an explosion of communication can lead to a collapse of organisation. We are drowning in information, while thirsting for knowledge. Academics (and students) often report feeling overwhelmed by the sheer volume of information. The human brain, alas, has not exponentially expanded. As noted, constant notifications induce <em>cognitive fatigue</em> (and lead to accountability sinks). Important ideas or tasks can get lost in the shuffle. When <em>everything is urgent, nothing is urgent</em>; when everything is vying for our attention, truly important signals are hard to discern. A Slack user for example is forced to live in the present, relentlessly. The result is an organisational culture of firefighting — extremely responsive but with little long-term planning or reflection.</p><p>The knowledge management implications are stark. Traditionally, organisations maintained institutional memory through documents, archives, and formal reports. But as more discussions move to ephemeral chat threads, the organisational knowledge base becomes fragmented and ephemeral. In many situations (often due to an incessant sense of needing to work at pace) messaging tools (typically those embedded in online meeting platforms) have effectively <em>replaced </em>documentation. It will often be remarked (or implied) that the digital platform is always ‘in record’ so no one need worry, yet once everyone leaves the meeting and are onto the next distraction, key information (while nonetheless stored ‘somewhere’) is lost cognitively.</p><p>A well-edited document that multiple people have refined captures collective knowledge in a durable form. By contrast, a decision made in a fast-paced chat room might be buried and forgotten a week later. The 24/7 reachability afforded by messaging, ironically, removes the earlier necessity to write things down in a form that can be understood by someone else in the future. Messaging is for the ‘now’, with little interest in the past and no foresight for the future.</p><p>Hence, the very structure of knowledge suffers. Messaging platforms mix the trivial and the important in a single feed — the cat GIF right next to the research idea — which can flatten one’s sense of priority. Important threads scroll out of view; there is a loss of hierarchy and context. We witness ‘organisational decay’, meaning not only the loss of formal structure in how institutions handle information, but also a psychological sense that our mental frameworks for organising knowledge are crumbling. Stiegler might say our <em>attention systems</em> (both individual and collective) are in disarray, needing rewiring to cope with the glut. As suggested above (and argued by Hayles) this is why we are seeing the emergence of AI, rather than AI being the reason we are at a loss in information. Just as Douglas Adams once remarked on how we invented the video recorder to watch all the endless TV we could no longer consume, we are today building AI to <em>advocate</em> for us in the overrunning of information — something we set in train ourselves!</p><h4><strong>New Deliberative Practices?</strong></h4><p>The challenges outlined — fractured attention, superficial reading, incessant busyness, and the spectre of automation — paint a sobering picture of scholarly life under the regime of instant messaging and AI. To avoid being swept away in the current of contextless data, academics might consciously reclaim more deliberate, thoughtful modes of working and learning. This could mean carving out tech-free blocks of time for deep reading and writing. It could involve reviving the practice of writing longer-form reflections (even just for oneself) to synthesize the torrent of information. In essence, creating personal meaning rather than just forwarding messages. On an institutional level, it may mean re-emphasising documentation and proper archiving protocols.</p><p>Critical theory reminds us that what is at stake is not mere workflow, but our capacity for critical thought and autonomy; the importance of culture that <em>challenges</em> and deliberates, rather than simply pleases with ‘good enough’ answers. We need to look to a form of <em>pharmacological wisdom</em> — knowing how to dose our use of technology so that it heals instead of harms.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8134199b3d30" width="1" height="1" alt=""><hr><p><a href="https://medium.com/electronic-life/whatsapped-8134199b3d30">WhatsApped?</a> was originally published in <a href="https://medium.com/electronic-life">Electronic Life</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Vibe Coding, Explained]]></title>
            <link>https://medium.com/electronic-life/vibe-coding-explained-e3632b8ce1bf?source=rss-bcc894f2ddf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/e3632b8ce1bf</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[coding]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[cruel-optimism]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Sunil Manghani]]></dc:creator>
            <pubDate>Sun, 30 Mar 2025 15:39:36 GMT</pubDate>
            <atom:updated>2025-03-30T15:48:22.637Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lE0eOjI45rnSb2ZSFbAHZw.png" /></figure><p><em>AI-assisted coding, or ‘vibe coding’, promises a world where everyone can be a developer — just by </em>describing<em> what they want. Is this empowerment or cruel optimism?</em></p><p>A curious phrase has entered the lexicon: <strong>vibe coding</strong>. The term, attributed to a <a href="https://x.com/karpathy/status/1886192184808149383">tweet</a> by Andrej Karpathy (former director of AI at Tesla), refers to a style of software development where instead of painstakingly writing code line by line, one simply guides an AI with natural-language prompts. In Karpathy’s pithy description, ‘<a href="https://x.com/karpathy/status/1886192184808149383">it’s not really coding — I just see stuff, say stuff, run stuff, and copy-paste stuff, and it mostly works</a>’. Or, as he puts it, ‘the hottest new programming language is English’ (cited by Niall McNulty , in ‘<a href="https://medium.com/@niall.mcnulty/vibe-coding-b79a6d3f0caa#:~:text=This%20approach%20marks%20a%20shift,assisted%20generation">Vibe Coding. AI-Assisted Coding for Non-Developers</a>’).</p><p>So, rather than mastering syntax and algorithms, a ‘vibe coder’ communicates the idea or ‘vibe’ of what they want, and lets the AI generate the heavy-lifting code. Advanced AI coding assistants, such as GPT-4 and Claude, as well as specialised tools like Replit’s Ghostwriter, are already being widely used, enabling the user to describe a desired programme in plain language and swiftly generate working prototypes. The result feels less like issuing strict instructions to a computer and more like <em>collaborating</em> with one.</p><h3>Everyone a Coder?</h3><p>Advocates of vibe coding see it as a powerful ‘democratising’ force. Just as the Instamatic camera (introduced in the early 1960s) enabled many more people to become photographers, and blogging platforms (in the early 2000s) let anyone become a publisher, AI-assisted coding could potentially allow millions of non-developers to become software creators. Self-confessed ‘old-school coder’, Pete Sena, notes we might have laughed ‘<a href="https://uxdesign.cc/cracking-the-code-of-vibe-coding-124b9288e551">at the idea that you can just blink software into reality</a>’, but nonetheless accepts vibe coding is indeed real. In fact, he makes a case for ‘embracing vibe coding without losing your soul’, suggesting four main approaches (more of which at the end).</p><p>So, what does vibe coding look like in practice? In one example, designer <a href="https://uxdesign.cc/my-kids-and-i-used-ai-to-build-a-game-b13f2fa22f5b">Ben Snyder and his children used AI tools</a> to build <a href="https://bird-dash-benscsnyder.replit.app/">BirdDash</a>, a rudimentary game where you play as an ostrich jumping over obstacles. In essence it is possible to simply describe a concept and output a playable prototype. Although, in truth, this still involves a creative <em>process</em>. In another example, tech journalist <a href="https://decrypt.co/311183/we-built-game-vibe-coding-ai-claude">Jose Antonio Lanz ‘vibe coded’ a typing game</a> by conversing with Anthropic’s Claude. The result, <a href="https://jaldps.itch.io/tranquil-type">Tranquil Type</a>, began with a simple prompt: ‘a game in which I must type out words fast to remain alive’. As Lanz recounts, the first version was buggy (the start button did nothing), but after a few natural-language nudges like ‘Nope, nothing happens when I click the button,’ the AI fixed the issues. After 14 iterative prompts, a functional fun word-typing game emerged. Lanz reports that colleagues testing the game found it enjoyable (one even beat the author’s high score!).</p><p>The ‘success stories’ suggest that AI can handle a lot of the grunt work in coding, allowing novices to achieve in hours what might take a traditional coder days or weeks. Notably, Lanz notes of limitations with respect to ‘design choices’. Indeed, it is frequently the case when coding with AI that the ‘logic’ of a programme can be handled very quickly, often in only one or two prompts. The difficulties arise, however, when you try to ‘discuss’ aesthetics, e.g. how things should <em>look</em> on the screen, or what the experience should be for a user. Your own idea of what a simple ‘blue button’ might look like and how it ‘feels’ when you click it, will likely be wildly different to an AI (the ability to reach a consensus can require an amusing/frustrating number of prompts and failed attempts!).</p><h3>Dream it, Build it</h3><p>The optimistic take is that we might finally break down the wall between idea-havers and technical implementers. People with great concepts — teachers, designers, entrepreneurs, even kids — are not barred by not knowing how to code. They can prototype an app or game the same day they imagine it, just by speaking to an AI. This could unleash a wave of creativity and innovation from voices previously absent in software development. And the implications for the tech industry could be profound.</p><p>Some foresee a world where programming expertise matters less than creative problem-solving and domain knowledge. After all, if <em>anyone</em> can conjure up a working app, what distinguishes the true professionals? As <a href="https://uxdesign.cc/cracking-the-code-of-vibe-coding-124b9288e551">Sena points out</a>, recent headlines suggest ‘a troubling trend: Amazon plans to terminate over 14,000 managerial positions to save $3.5 billion annually. Meta, Microsoft, and countless others are making similar moves’. Added to which, AI coding company, Replit, <a href="https://x.com/amasad/status/1886516600653930924">report</a> that ‘75% of Replit customers never write a single line of code’ to begin with. As <a href="https://uxdesign.cc/cracking-the-code-of-vibe-coding-124b9288e551">Sena puts it</a>, ‘[t]he future is arriving faster than we think’:</p><blockquote><em>When anyone can generate functional code through conversation, the specialization that once protected technical roles evaporates. The implications ripple through organizations:</em></blockquote><blockquote><strong><em>Product managers</em></strong><em> can’t hide behind documents and wireframes — they’ll need to generate working prototypes.<br></em><strong><em>Designers</em></strong><em> can’t simply hand off mockups — they’ll need to implement their designs.<br></em><strong><em>Marketers</em></strong><em> can’t request custom tools — they’ll build their own analytics dashboards.<br></em><strong><em>Executives</em></strong><em> can’t claim technical ignorance — they’ll need to understand the systems they oversee.</em></blockquote><blockquote><em>- Pete Sena, ‘</em><a href="https://uxdesign.cc/cracking-the-code-of-vibe-coding-124b9288e551"><em>Cracking the code of vibe coding: Not all vibes are good</em></a><em>’</em></blockquote><p>Early in the twentieth century, in his well-known essay ‘The Work of Art in the Age of Mechanical Reproduction’, Walter Benjamin observed how inventions like photography and film upended the cultural production of art, stripping away its ‘aura’ and radically expanding who could create and consume art. Just as Early 20th-century painters saw their craft transformed by the camera, early 21st-century programmers may likewise see theirs transformed by generative AI. Vibe coding makes software less of a rarefied craft and more of a mass cultural activity.</p><p>Hence, just as photography made images infinitely reproducible, AI may make code infinitely suggestible, generated on-the-fly for anyone who asks. Vibe coding turns the user into the producer. The person with the idea (the ‘author’) can now also be the one who makes a working product. The <em>potential</em> is for a more participatory culture…</p><h3>A lot Worse Than Expected</h3><p>However, before proclaiming the death of programming as we know it, it is important to examine ‘actually existing’ vibe coding. While it is impressive to spin up a simple game or app via chat prompt, more complex or polished projects typically reveal the limitations of AI-as-developer.</p><p>Take the classic arcade game Pac-Man. It is a straightforward concept by human standards — a yellow circle munching dots in a maze — and decades of coders have reimplemented it as a beginner’s exercise. In theory, a decent AI should be able to reproduce such a well-documented game. Recently, <em>The Guardian</em> tech reporter <a href="https://www.theguardian.com/games/2025/mar/11/ai-pac-man-clones-reviewed-grok?CMP=oth_b-aplnews_d-1">Rich Pelley put this to the test by asking seven people with no coding experience to each create a Pac-Man</a> clone using generative AI tools. The result? In his words: ‘a lot worse than expected’. The vibe coders encountered all manner of glitches and incomplete game logic. Even with AI help, Pac-Man proved surprisingly tricky to get right.</p><p>As AI sceptic, Gary Marcus, suggests, if an AI can’t reliably recreate a 1980 maze game (for which the full source code is publicly available), that raises doubts about claims that ‘<a href="https://garymarcus.substack.com/p/ai-coding-fantasy-meets-pac-man?utm_campaign=post&amp;triedRedirect=true">neophytes will soon be able to build the software of their dreams</a>’. Indeed, even in the ‘success’ stories (cited above) it is important to temper the accounts. The journalist who built the typing game with Claude, for example, ended up iterating the code fourteen times to fix errors and refine the gameplay. Without his patience and some degree of understanding of what to ask the AI to change, the project would have gone nowhere. He admits that the final product, while fun, ‘<a href="https://decrypt.co/311183/we-built-game-vibe-coding-ai-claude">lacked the polish and optimization of professionally developed software</a>’. As another writer wryly notes after trying similar AI game generators, vibe-coded games can feel ‘<a href="https://www.reddit.com/r/gamedev/comments/1jitfbi/im_tired_of_the_vibe_game_coding_trend_good_games/?rdt=60410">empty and soulless</a>’, more proof-of-concept than compelling experience; adding ‘good games are made by humans’.</p><h3>Cruel Optimism?</h3><p>All this highlights an important truth: today’s AI coding assistants can rapidly produce code, but not necessarily <em>good</em> code<em>.</em> It is the difference between <em>something that runs</em> and <em>something that’s robust</em>, secure, maintainable, and tuned for user experience. As I remark in my article on <a href="https://medium.com/electronic-life/the-art-of-coding-a-cruel-optimism-59fc52571e91">the ‘cruel optimism’ of AI coding</a>, there are signs of significantly higher ‘code churn’. In other words, AI may spit out lots of working lines, but those lines might be messier and less sustainable than code crafted by experienced developers. There is the suggestion that AI-generated code is like an ‘<a href="https://www.gitclear.com/coding_on_copilot_data_shows_ais_downward_pressure_on_code_quality">itinerant contributor</a>’ that doesn’t stick around to clean up the mess it makes​!</p><p>Cultural theorist, Lauren Berlant, coined the term ‘cruel optimism’ to describe situations where people cling to hopeful dreams that actually hinder them in the long run (see <em>Cruel Optimism,</em> Duke University Press, 2011). <a href="https://medium.com/electronic-life/the-art-of-coding-a-cruel-optimism-59fc52571e91">It remains apt in considering our relationship to coding</a>. On the surface, vibe coding, the idea of AI turning everyone into a programmer, is deeply appealing — who wouldn’t want to erase the frustration and gatekeeping associated with learning to code? Personally, I have found it liberating to <em>experiment</em> with coding. Yet, we need to distinquish that from a grander ‘promise’ for a future where we are no longer limited by technical skills, only by our imagination. The tools are changing, but we need to maintain critical engagement and to sharpen, not loosen expertise.</p><p>AI critics like Gary Marcus amplify these concerns, yet, equally, the danger is to present things in too binary a fashion. In his analysis of the Pac-Man example, Marcus argues that today’s AI coding tools are mere ‘souped-up autocompletes’ — this kind of analysis seems more aimed at headlines than thoughtful engagement. It is true, AI models lack ‘understanding’ of <em>why</em> a program works or fails. As a result, they can write code that seems plausible but breaks easily, or can’t adapt when requirements stray from training examples. Marcus cautions that an ‘overdependence on the new tools will lead beginners to make giant mountains of difficult-to-maintain spaghetti code’, with obvious security and quality issues. These are genuine issues, but concurrently it is perhaps not helpful to pose the problem as an AI model’s ‘lack of understanding’. They are not designed to ‘understand’, but rather to determine patterns (to then aid prediction). Our own understanding of how they work and of their limitations is surely our own responsibility (not that of the AI).</p><p>And we need to accept our own limitations. The Pac-Man exercise starkly illustrated this when the AI’s output was wrong, since the non-coders did not have the intuition or skills to find a fix. A seasoned developer would be able to diagnose the game logic in minutes, but the vibe coding approach can easily leave its users helpless in the face of errors. In effect, vibe coding can make programming look easy until something goes wrong, at which point the depth of true programming expertise (or the lack thereof) becomes painfully apparent.</p><h3>Riding the Vibe…</h3><p>So, is vibe coding a revolutionary leap forward, or just tech optimism? The answer may be a bit of both. There is no doubt something transformative is happening: the very definition of ‘coding’ is expanding to include natural language prompting. In the same way we enabled web design without the need to know HTML, it is likely we will soon accept creating software without the need to write Python. This could genuinely broaden who gets to make software, inviting a more diverse pool of creators and potentially shifting power away from a select priesthood of programmers. That, at least, is the <em>utopian</em> scenario.</p><p>As we come to grips with the ‘era’ of vibe coding, <a href="https://uxdesign.cc/cracking-the-code-of-vibe-coding-124b9288e551">Pete Sena suggests we consider four new models or approaches</a>, whereby we need to:</p><p><strong>(1) </strong>Consider the <em>creation-maintenance divide</em>: while vibe coding makes creation easy and accessible, it struggles with maintenance, which still demands deep expertise — highlighting the need for both rapid prototyping skills and rigorous engineering.</p><p><strong>(2) </strong>Be cognisant of a <em>new hierarchy of software creators</em>: from prompt engineers who implement known patterns, to solution architects combining AI capabilities creatively, to system innovators developing entirely new paradigms.</p><p><strong>(3)</strong> Recognise a shift from <em>execution leverage</em> (automating tasks) to <em>intellectual leverage</em> (automating thinking), so placing greater value on identifying and building the <em>right</em> things.</p><p><strong>(4)</strong> Navigate the <em>specialisation paradox</em>, which suggests that as tools grow more powerful, success depends less on deep specialisation and more on the ability to synthesise across domains — favouring those who can connect ideas rather than just master one.</p><p>Yet, of course, a <em>dystopian</em> mirror image is easy to see. With a flood of AI-generated apps, the question remains unanswered: <em>who</em> will maintain them? Who will ensure they’re secure, reliable, and serving human needs rather than inadvertently causing harm? We will likely end up with a glut of brittle software — unstable, poorly understood systems put out into the world by people (such as myself!) who don’t really know how their own creations work. In the extreme, the world’s software infrastructure becomes a mass of pasted-together code, prone to bugs and vulnerabilities that everyone assumes ‘the AI’ can somehow resolve. With the emergence of film and photography came as well a deluge of mediocre output, resulting in the challenge to ‘curate’ quality. While mechanical reproduction challenged the ‘aura’ of art, automated coding arguably challenges something further, far beyond the auratic.</p><ul><li><a href="https://medium.com/@niall.mcnulty/vibe-coding-b79a6d3f0caa">Vibe Coding</a></li><li><a href="https://uxdesign.cc/cracking-the-code-of-vibe-coding-124b9288e551">Cracking the code of vibe coding</a></li><li><a href="https://medium.com/electronic-life/the-art-of-coding-a-cruel-optimism-59fc52571e91">The Art of Coding: A Cruel Optimism?</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e3632b8ce1bf" width="1" height="1" alt=""><hr><p><a href="https://medium.com/electronic-life/vibe-coding-explained-e3632b8ce1bf">Vibe Coding, Explained</a> was originally published in <a href="https://medium.com/electronic-life">Electronic Life</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Gen.AI ≈ Search, but who owns the patterns?]]></title>
            <link>https://medium.com/electronic-life/gen-ai-search-but-who-owns-the-patterns-1d7c74001444?source=rss-bcc894f2ddf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/1d7c74001444</guid>
            <category><![CDATA[copyright]]></category>
            <category><![CDATA[search]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[blockchain]]></category>
            <dc:creator><![CDATA[Sunil Manghani]]></dc:creator>
            <pubDate>Sun, 23 Mar 2025 16:25:16 GMT</pubDate>
            <atom:updated>2025-03-23T16:26:53.745Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RfyMhaZowEQagqERVOcQDQ.jpeg" /><figcaption>AI-generated: Who owns the patterns?</figcaption></figure><p><em>Generative AI is reshaping our relationship with information, but we shouldn’t be distracted by so-called ‘artificial intelligence’. What is at stake is the evolution of ‘search’: who own the ‘patterns’ of human knowledge and how can we maintain an open pursuit?</em></p><p>Technologies have their affordances. A hammer makes us see the world in terms of nails; a spreadsheet makes us see it in terms of quantifiable data. New technologies are no different, although, arguably, they accelerate the affordances. Smartphones transformed our relationship with information, making it constantly accessible. This created new affordances for communication and knowledge retrieval, which in turn shaped new social norms and attention patterns.</p><p>In the last couple of years, with the emergence of new generative AI tools (e.g language models such as GPT-4, Claude, and image models such as DALL-E, Midjourney), there has been a lot of ‘noise’ about the prospect of <a href="https://www.scientificamerican.com/article/what-does-artificial-general-intelligence-actually-mean/">artificial general intelligence</a> (AGI). The idea of AI systems that are ‘<a href="https://openai.com/index/planning-for-agi-and-beyond/">generally smarter than humans</a>’ leads to rather wilder affordances, with the potential of replacing humans altogether. <strong>But what if <em>really</em> we have been building a new form of search?</strong> A radical new form, but search nonetheless.</p><p>The shift to understanding generative AI <em>as search</em> helps us view the debates around copyright for AI training materials in a different light.</p><h3>AI as Search (at scale!)</h3><p>Traditional search helps us find existing information, while generative AI synthesizes information in ways that appear novel. Yet, we are still accessing knowledge that already exists in some form. Search engines point us to where answers exist, while generative AI reconstructs answers based on patterns in its training data. What makes generative AI feel revolutionary is that it presents information in a conversational, adaptive format that better matches human cognitive patterns. This new form of search doesn’t deliver documents containing answers — it delivers answers directly, synthesized from countless sources, in a form that feels like understanding rather than just retrieval.</p><p>By 2020 we began to see notable improvements in Large Language Models (LLMs), primarily driven by <em>scaling</em> — making models bigger, feeding them more data, and using more computing power. This approach has yielded remarkable results, with each generation of models demonstrating increasingly sophisticated capabilities. Models grew from millions to hundreds of billions of parameters in just a few years, and seeming to demonstrate that performance reliably improves with scale. The success of such scaling has led many to propose that artificial general intelligence might be achieved simply by continuing to scale up these systems — if we just make them big enough, true intelligence will emerge. This requires massive amounts of compute, hence we hear of companies such as Microsoft and OpenAI entering into negotiations to develop nuclear power plants (there is also a lot of speculation on the prospects of <a href="https://medium.com/electronic-life/quantum-life-but-not-as-we-know-it-cc57705e06fd">quantum computing</a>).</p><p>However, the ‘scaling hypothesis’ remains highly contentious among AI researchers, with many arguing that fundamental architectural innovations, not just size, will be necessary to achieve human-like intelligence. <strong>Yann LeCun, </strong>Meta chief AI scientist, is particularly vocal upon this point:</p><blockquote><em>We are not going to get human level AI by just scaling up LLMs. This is just not going to happen … There is no way … absolutely no way. Whatever you can hear from some of my more adventurous colleagues, it’s not going to happen within the next two years. There’s absolutely no way in hell … What we are going to have, maybe, is systems that are trained on sufficiently large amounts of data that any question that any reasonable person may ask will find an answer through the system. And it would feel like you have a PhD sitting next to you. But it is not a PhD you have next to you. It’s a system with a gigantic memory and retrieval ability, not a system that can invent solutions to new problems, which is really what a PhD is… <br>- </em><strong><em>Yann LeCun</em></strong><em> speaking on </em><a href="https://pod.link/1522960417/episode/13d6afeaa7f75bca5daf8f9e21824c37">Big Technology Podcast</a></blockquote><p>In LeCun’s words, we are better understanding LLMs simply as ‘gigantic memory and retrieval’ systems, or very fancy search engines<em>.</em> (NB. This is not to say that AI reasoning is off the table, indeed, <a href="https://www.youtube.com/watch?v=N09C6oUQX5M">LeCun himself has much to say on the subject;</a> but these systems would require different architectures, and are less related to immediate issues of copyright, more about modes of ‘thinking’).</p><h3>We Need to Steal Your Copyright!</h3><p>From the Trump’ adminstration’s forthcoming ‘<a href="https://www.whitehouse.gov/briefings-statements/2025/02/public-comment-invited-on-artificial-intelligence-action-plan/">AI Action Plan</a>’ to the European Union’s comprehensive ‘<a href="https://artificialintelligenceact.eu/">AI Act</a>’ and the UK’s ‘<a href="https://www.gov.uk/government/consultations/copyright-and-artificial-intelligence/copyright-and-artificial-intelligence">Copyright and AI</a>’ review, governments are seeking to address and regulate how AI systems can access, learn from, and potentially transform copyrighted materials. In America, for example, OpenAI have openly warned the US administration that America will ‘lose’ the AI race if it is unable to scrape copyrighted materials. Tacitly, this is an admission that it has been necessary to ‘steal’ copyrighted materials in order to get to the current level of LLM sophistication. What emerges is an odd twist on preserving democracy, with OpenAI arguing the need for ‘funneling as much data as possible to AI companies — regardless of rights holders’ concerns — [as…] the only path to global AI leadership’ (<a href="https://arstechnica.com/tech-policy/2025/03/openai-urges-trump-either-settle-ai-copyright-debate-or-lose-ai-race-to-china/"><em>Ars Technica</em></a><em>)</em>.</p><p>As OpenAI put it:</p><blockquote><em>The federal government can both secure Americans’ freedom to learn from AI and avoid forfeiting our AI lead to the [People’s Republic of China] by preserving American AI models’ ability to learn from copyrighted material […] If the PRC’s developers have unfettered access to data and American companies are left without fair use access, the race for AI is effectively over […] America loses, as does the success of democratic AI. Ultimately, access to more data from the widest possible range of sources will ensure more access to more powerful innovations that deliver even more knowledge. <br>- OpenAI cited in </em><a href="https://arstechnica.com/tech-policy/2025/03/openai-urges-trump-either-settle-ai-copyright-debate-or-lose-ai-race-to-china/"><em>Ars Technica</em></a><em>.</em></blockquote><p>Nonetheless, viewing LLMs as new forms of <em>search</em> reminds us that these heated debates over AI copyright are not unprecedented but rather an evolution of tensions that began decades ago with the rise of the World Wide Web. When Google digitised millions of books for its Library Project, when Napster disrupted music distribution, and when Wikipedia challenged traditional encyclopaedias, we witnessed similar conflicts between information access and creator rights.</p><p>Today’s claims from OpenAI over a geopolitical ‘race’ for AI dominance echo earlier arguments from tech platforms about innovation requiring liberal access to existing works. The fundamental question remains unchanged: <em>how do we balance the social benefit of widely accessible knowledge against the rights of those who produce it? </em>What is different is not the nature of the conflict but its scale and stakes. LLMs have industrialised the process of information extraction and synthesis that search engines began, making the economic and cultural tensions around information ownership more acute than ever. As we debate whether AI training constitutes fair use, we are really continuing a conversation about the economics of knowledge that began with the first digital library — but now with more profound implications for creative labor and cultural production.</p><h3>Who Owns the Patterns?</h3><p>The emergence of the World Wide Web in the 1990s sparked intense debates about the economics of information. Suddenly, content could be reproduced and distributed at virtually zero marginal cost, challenging traditional publishing models and intellectual property frameworks. Questions arose about who owned information when it became digital, how creators would be compensated in an environment where copying was trivial, and whether knowledge should be treated as a public good or a private commodity. These tensions gave rise to movements like open source software, Creative Commons licensing, and digital rights management (DRM) — all attempts to navigate a world where information wanted to be both ‘free’ and ‘expensive’ (i.e. freely available, while economically valued).</p><p>We are witnessing these same debates resurface with unprecedented urgency around generative AI. These systems train on massive datasets of human-created content — books, articles, code, artwork, music — without explicit permission from or compensation to the original creators. The fundamental question has shifted from ‘Who owns a copy?’ to ‘<strong>Who owns the patterns?’ </strong>As AI generates content that appears new but is derived from existing works, we face complex questions about attribution, compensation, and the nature of creativity itself. Are we moving toward a post-scarcity knowledge economy where creation is democratised, or are we building sophisticated extraction mechanisms that ultimately devalue human creative labour? (see: ‘<a href="https://medium.com/the-ai-and-arts-forum/automatic-art-the-post-knowledge-economy-cf2bec2e4e6d">Automatic Art &amp; a Post-Knowledge Economy?’</a>) The answers to these questions will shape not just AI development but the broader social contract around knowledge and content creation.</p><h3>Abolish Copyright?</h3><p>The shift to questioning to ‘who owns the patterns’ is a reminder — for good or ill — that digitised culture has had the affordance of transforming creative works into statistical patterns. It is precisely the statistical turn that has made it possible to make the current crop of LLMs (and other related generative AI models). Yet as these models grow increasingly sophisticated at mimicking and synthesising the style of original works, the argument for ‘fair use’ over copyrighted material for training becomes harder to sustain.</p><p>Of course, we might simply abolish copyright. An anarchist perspective, for example, argues that intellectual property regimes primarily serve to protect corporate interests rather than individual creators, creating artificial scarcity in what are naturally non-rivalrous goods — ideas and information. Proponents of this view, drawing from thinkers like Pierre-Joseph Proudhon (‘<a href="https://theanarchistlibrary.org/library/pierre-joseph-proudhon-property-is-theft">Property is theft!</a>’) and more contemporary figures like <a href="https://doctorow.medium.com/copyright-wont-solve-creators-generative-ai-problem-92d7adbcc6e6">Cory Doctorow</a>, might argue that LLMs represent the inevitable conclusion of information’s natural tendency to flow freely. In such a framework, the very concept of ‘training on copyrighted material’ becomes meaningless, as all knowledge should be part of a cultural commons.</p><p>The <a href="https://web.archive.org/web/20210820153013/https://www.nytimes.com/1989/01/11/business/business-technology-one-man-s-fight-for-free-software.html">Free Software Movement</a>, initiated by Richard Stallman in the 1980s, offers a practical precedent. By creating the <a href="https://www.gnu.org/licenses/gpl-3.0.en.html">GNU General Public License</a>, Stallman established a legal framework that used copyright law against itself — ensuring software remained free to use, modify, and distribute. This ‘<a href="https://www.gnu.org/licenses/copyleft.en.html">copyleft</a>’ approach demonstrates how alternative models can thrive even within existing systems. Similarly, <a href="https://creativecommons.org/share-your-work/cclicenses/">Creative Commons licenses</a>, Lawrence Lessig’s ‘<a href="https://lessig.org/product/free-culture/">Free Culture</a>’ movement, and <a href="https://www.theguardian.com/technology/2013/jun/02/aaron-swartz-hacker-genius-martyr-girlfriend-interview">Aaron Swartz’s information activism</a> provide intellectual foundations for imagining post-copyright knowledge ecosystems. The success of open access scientific publishing (see ‘Scholarly Blockchain’ below) further demonstrates viable alternatives to traditional copyright-based models.</p><p>Critics of copyright abolition inevitably raise concerns over adequate compensation, while advocates respond with examples of alternative funding models (Patreon-style patronage, public funding, reputation economies, and service-based revenue streams). They argue that most creators already receive minimal benefit from copyright while bearing its constraints. To date, AI companies have successfuly benefited by arguing that the handling of creative works as mathematical weights within a neural network is sufficiently <em>transformative</em> to avoid copyright infringement. It might seem, then, that the existing, underlying inequities simply persist, or might the new AI paradigm afford us a new way to look at things?</p><h3>Blockchains, NFTs and DRM</h3><p><em>Enter: blockchain technology.</em> Take, for example, NFTs (Non-Fungible Tokens), which gained notoriety through speculative art markets — often criticized as solutions in search of problems. Yet, they embody a concept that could be genuinely useful: verifiable digital provenance. In the wrong hands, NFTs would seem only to have been a pitiful fad for the art world (the epitome of capitalist art). Yet, thought through with more care (and adopted <em>not</em> at the expense of the ‘auratic’ artwork) NFTs perhaps have more salience.</p><p>In 1993, Eric Hobsbawm gave the Walter Neurath Memorial Lecture, under the title: ‘Behind the Times: The Decline and Fall of the Twentieth-Century Avant-Gardes’. He discussed how visual arts struggled to adapt to technological reproduction and mechanical means in ways that other art forms did not. His thesis was that, unlike music, literature, and film, the visual arts — particularly painting and sculpture — retreated into increasingly abstract and conceptual territory. Rather than finding ways to meaningfully engage with new technologies of mass reproduction (photography, film, television), fine art became more esoteric and separated from mainstream culture. This retreat, he argued, reflected a larger failure of the 20th century avant-garde to maintain relevance in an age of mechanical reproduction. While the lecture touches on themes similar to Walter Benjamin’s much-cited ‘The Work of Art in the Age of Mechanical Reproduction’, Hobsbawm took a more critical stance toward how visual artists responded to these challenges.</p><p>He specifically critiqued how the avant-garde became disconnected from ordinary people’s experiences, arguing that visual artists had essentially abandoned their historical role of interpreting and making sense of the modern world. While other art forms evolved to speak to mass audiences through new technologies, visual arts became more insular, speaking mainly to other artists and specialized critics. For Hobsbawm, this represented a fundamental failure. In response to which, might we today consider NFTs (and related technologies and techniques) in a more progressive light? NFTs could offer something more substantial — a transparent, verifiable system for tracking attribution and usage rights.</p><p>Unlike the crude early implementations that treated NFTs as digital equivalents of artworks (i.e. recreating artificial scarcity in a medium defined by reproducibility), a more sophisticated blockchain approach could create permanent, immutable records of authorship without restricting circulation; enable creators to maintain verifiable connections to their work even after being incorporated into AI training data; facilitate nuanced licensing where creators could specify different permissions for different types of use; and, provide infrastructure for micropayments when works contribute to commercial AI systems</p><p>This reimagined approach doesn’t try to fight the inherent reproducibility of digital content (the mistake Hobsbawm identified in 20th century visual arts). Instead, it works with the reality of digital reproduction while maintaining the crucial connections between creators and their works. Hence, rather than mimicking the art market’s emphasis on scarcity, this approach acknowledges that AI training represents a new relationship between creative works and their subsequent uses — one requiring new frameworks for attribution and compensation that blockchain technologies are uniquely positioned to provide.</p><p>For example, a blockchain-based system for creative works could implement:</p><ul><li><strong>Content Fingerprinting</strong>: Unique identifiers for works that persist even when incorporated into training data</li><li><strong>Smart Licensing</strong>: Automated permissions and compensation when works are used in AI training</li><li><strong>Usage Tracking</strong>: Transparent accounting of how frequently certain works influence model outputs</li><li><strong>Creator-Defined Terms</strong>: Enabling creators to specify exactly how their work can be used</li></ul><p>Written here in black and white, it might feel as if this all sounds very ’regulatory’, yet it is important to remember blockchain technology is fundamentally decentralised. Its ‘power’ lies in its distributed nature, where no single entity governs the system. Instead, it creates transparent, tamper-resistant records through consensus mechanisms that operate across a network of peers. What emerges is not regulation in the conventional sense, but a commons-based infrastructure where rights and permissions are directly negotiated between creators and users through code rather than intermediaries. This preserves individual agency while creating sustainable ecosystems for creative work — allowing both openness and attribution to coexist in ways current systems struggle to reconcile.</p><p>Whether blockchain-enabled rights or copyright abolition, there is an underlying philosophical divide: information as <strong>property</strong> versus information as <strong>commons</strong>. The abolition perspective sees LLMs as inevitable manifestations of collective knowledge that should be freely accessible, while the blockchain approach attempts to preserve individual attribution within increasingly fluid information ecosystems. The former raises questions as to how we would <em>sustainably</em> keep producing new knowledge. The latter requires a very radical change of infrastructures, which might well already be emerging, but which only emerge piece-meal. More practically, we witness hybrid models like those emerging in open source software, where commercial entities build proprietary services atop open foundations, or where value accumulates to communities rather than individuals. Projects like Wikipedia, Linux, and Python demonstrate how collaborative creation can thrive without traditional copyright incentives, while still generating substantial economic activity.</p><h3>Scholarly Blockchain</h3><p>To take another, specific example: Academic publishing presents both unique challenges and opportunities for reimagining knowledge creation in the age of LLMs. The current system shows significant inefficiencies: redundant literature reviews, publication requirements that favor quantity over quality, and a peculiar phenomenon of reviewers demanding citations primarily to demonstrate erudition rather than substantive contribution to an argument.</p><p>Nonetheless, the academic realm offers fertile ground for more radical thinking. The <a href="https://opusproject.eu/openscience-news/understanding-the-open-science-movement-through-the-lens-of-history/">open science movement</a>, championed by figures like Michael Nielsen and Alexandra Elbakyan (creator of <a href="https://sci-hub.lu/">Sci-Hub</a>), already challenges the enclosure of scholarly knowledge behind publisher paywalls. This movement views scientific knowledge as humanity’s collective inheritance — something that should be accessible to all, not subject to artificial scarcity.</p><p><a href="https://infed.org/mobi/ivan-illich-deschooling-conviviality-and-lifelong-learning/">Ivan Illich’s concept of ‘deschooling society</a>’ and Michel Bauwens’ work on <a href="https://p2pfoundation.net/">peer production</a> provide perspectives for reimagining knowledge creation outside traditional institutional boundaries. And of course, more bluntly, anthropologist David Graeber’s analysis of ‘<a href="https://www.theguardian.com/books/2018/may/25/bullshit-jobs-a-theory-by-david-graeber-review">bullshit jobs</a>’ might be evoked to suggest how academia frequently reproduces unnecessary work — not least redundant literature reviews to satisfy received ways of doing things rather than advance understanding.</p><p>Academic publishing has long been straining under its own weight, before AI entered the picture:</p><blockquote><em>Commercial publishers are incentivised to try to publish as many articles and journals as possible, because each additional article brings in more profit. This has led to a proliferation of junk journals that publish </em><a href="https://www.wsj.com/science/academic-studies-research-paper-mills-journals-publishing-f5a3d4bc"><em>fake research</em></a><em>, and has increased the pressure on rigorous journals to weaken their quality controls. It’s never been more evident that for-profit publishing simply does not align with the aims of scholarly inquiry. <br>- </em><a href="https://www.theguardian.com/commentisfree/article/2024/jul/16/academic-journal-publishers-universities-price-subscriptions"><em>Arash Abizadeh</em></a><em>, ‘Academic journals are a lucrative scam’, </em>The Guardian</blockquote><p>Now, with LLMs demonstrating remarkable proficiency at producing literature reviews and synthesizing existing knowledge — often more comprehensively than human researchers — we face a fundamental question: Why continue producing knowledge in forms that are increasingly automatable? The <a href="https://wiki.p2pfoundation.net/Hacker_Ethic">hacker ethic</a> articulated by Steven Levy and the principles of ‘convivial tools’ described by Illich might lead us to consider how technologies can empower communities rather than restrict (and extract from) them. In fact, LLMs might be seen not as threats to academic authority but as tools for democratising access to knowledge synthesis — provided they remain open and community-governed.</p><p>A <em>scholarly blockchain</em> approach, for example, could seek a radical alternative, to transform how academic knowledge advances:</p><ol><li><strong>Knowledge Graphs Instead of Papers</strong>: Research could contribute to interconnected knowledge graphs where relationships between findings are explicit;</li><li><strong>Incremental Contributions</strong>: Researchers could focus on contributing new findings or refinements to specific nodes in the knowledge graph;</li><li><strong>Collaborative Verification</strong>: Peer review would assess individual contributions rather than complete papers;</li><li><strong>Attribution Chains</strong>: Clear tracking of who contributed what to the collective knowledge base;</li><li><strong>Dynamic Literature</strong>: continually updating ‘reviews’ would be generated on demand from the knowledge graph rather than repeatedly written.</li></ol><p>Such an approach might currently seem wild, and would most certainly unnerve many — not least for the sense of losing individual status and authority. The model inherently shifts from individual authorship toward collective knowledge creation — a move that, arguably, better reflects the reality of how scholarship actually progresses. Rather than artificial competition for priority, it encourages collaboration and building on others’ work. Hence, such an approach might better acknowledge how the true value in scholarship lies in novel discoveries and insights (as a shared endeavour), not in the packaging of (known) information — precisely the task at which LLMs excel. By separating these functions, if it possible to lead towards a more valuable division of labor between human creativity and machine synthesis? NB. It is not about machines <em>generating</em> knowledge, but rather better supporting our already ‘gigantic memory and retrieval’ system.</p><p>The challenges are substantial: academic institutions would need to reimagine promotion criteria, publishers would require new business models, and researchers would need different skills. Yet the potential gains in transparency and the genuine advancement of knowledge make for a compelling vision for how scholarship might evolve in response to AI capabilities. All in all, it is a vision not of reform, but re-imagination.</p><p>Yet, there are precedents. Projects like <a href="https://plos.org/">Public Library of Science</a> (PLOS), <a href="https://arxiv.org/">arXiv</a>, and the broader preprint movement already demonstrate how removing gatekeepers can accelerate knowledge dissemination without sacrificing quality. <a href="https://en.wikipedia.org/wiki/Decentralized_autonomous_organization">Decentralized Autonomous Organizations</a> offer potential governance models for scholarly communities that operate outside traditional institutional hierarchies. Peter Kropotkin’s account of ‘<a href="https://theanarchistlibrary.org/library/petr-kropotkin-mutual-aid-a-factor-of-evolution">Mutual Aid</a>’ provides precedent for how cooperative knowledge creation thrived long before formal academic structures. Similarly, anthropological work by <a href="https://www.lrb.co.uk/the-paper/v30/n19/james-c.-scott/try-it-on-the-natives">James C. Scott on indigenous knowledge systems</a> reveals sophisticated methods for preserving and extending collective wisdom without centralised authorities or proprietary frameworks.</p><h3>Concluding ‘proof’</h3><p>Regardless of legal ‘frameworks’ or technological ‘solutions’, we continually circle back to something more fundamental: the <strong><em>irreplaceable experience </em></strong>of human meaning-making. As the anthropologist Levi-Strauss once put it:</p><blockquote><em>Seen on the scale of millennia, human passions fuse. Time neither adds to nor removes anything from the loves and hates felt by people, theirinvolvements, their struggles and their hopes. They remain the same today as they were in the past. Randomly removing ten or twenty centuries of history would not affect, in a meaningful way, our knowledge of human nature. </em><strong><em>The only irreplaceable loss would be the works of art </em></strong><em>which these centuries gave rise to: because </em><strong><em>humans only differ through their works, and even exist only through them</em></strong><em>. Just as a wooden statue attests to the prior existence of a tree, so </em><strong><em>works of art provide the proof that, throughout history, amongst people, something actually happened</em></strong><em>.<br>- Claude Levi-Strauss, </em>Regarder, ecouter, lire</blockquote><p>LLMs, trained on vast sways of human production can easily parse ‘all’ of the data to reveal the repeated patterns of human culture. But they cannot report on the differences <em>we</em> discern from existing amongst the undulating patterns. LLMs have no proofs. Hence, while blockchain approaches might indeed promise ways to reconcile attribution with access, it is useful to consider LLMs primarily as <em>sophisticated search engines</em> rather than generative tools.</p><p>These models, however advanced, remain pattern recognition systems trained on the collective expressions of human thought and creativity. They can synthesize, recombine, and even mimic style with remarkable fidelity, but they cannot replace the lived experience that makes writing meaningful or knowledge valuable. It is quite possible a blockchain ledger will be established to record ownership of patterns, but that will never need replace the embodied act of understanding or the <em>pleasure</em> of encountering well-crafted works.</p><p>Perhaps this perspective offers the most promising resolution: rather than seeing LLMs as competitors or replacements for human knowledge work, we can better understand them as tools to handle the increasingly overwhelming task of information retrieval and synthesis, freeing us to focus on what remains distinctively human: the creation of genuinely new insights, the experience of discovery, and the pleasure we derive from engaging with ideas. Of course, the question remains: <strong><em>who owns these patterns, the many or the few?</em></strong> Our task is to answer this question in a way we would choose willingly if fated to live eternally, or at least across the scale of millennia.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1d7c74001444" width="1" height="1" alt=""><hr><p><a href="https://medium.com/electronic-life/gen-ai-search-but-who-owns-the-patterns-1d7c74001444">Gen.AI ≈ Search, but who owns the patterns?</a> was originally published in <a href="https://medium.com/electronic-life">Electronic Life</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Arts: Reflections on Lev Manovich’s 3 Definitions]]></title>
            <link>https://medium.com/electronic-life/ai-arts-reflections-on-lev-manovichs-3-definitions-8d7a1e63c20b?source=rss-bcc894f2ddf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/8d7a1e63c20b</guid>
            <category><![CDATA[art]]></category>
            <category><![CDATA[lev-manovich]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[ai-art]]></category>
            <category><![CDATA[ai-art-generator]]></category>
            <dc:creator><![CDATA[Sunil Manghani]]></dc:creator>
            <pubDate>Sun, 05 Jan 2025 07:46:08 GMT</pubDate>
            <atom:updated>2025-01-05T14:54:42.663Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FFWMqyUhXb5MvhyPMeLxLQ.jpeg" /><figcaption>ChatGPT generated.</figcaption></figure><p><strong><em>Reflections on Lev Manovich’s 3 definitions of AI Art. Post-ChatGPT, has anything changed?</em></strong></p><p>The rise of artificial intelligence in the arts has prompted both excitement and unease, with its potential to mimic, enhance, or even redefine human creativity. Lev Manovich’s 2019 article, ‘<a href="https://manovich.net/index.php/projects/defining-ai-arts-three-proposals">Defining AI Arts: Three Proposals</a>’, offers a prescient critique of the discourse surrounding AI Art(s). By exploring three distinct definitions of AI arts, Manovich challenges us to question not only the boundaries of AI creativity but also the evolving nature of art itself. <strong>NB. Manovich wrote his article <em>prior</em> to the advent of ChatGPT and related developments with generative AI (I.e. Large Language Models and image diffusion etc.)</strong>. Nevertheless, revisiting his arguments in light of advancements like ChatGPT and DALL-E underscores the enduring relevance of his ideas and the complexities of defining AI art in an era of rapidly advancing technologies.</p><h3>1.</h3><blockquote><strong><em>In this definition, art created by an AI is something that professionals recognize as valid historical art or contemporary art.</em></strong><em> — </em><a href="https://manovich.net/index.php/projects/defining-ai-arts-three-proposals"><em>Manovich</em></a></blockquote><p>Manovich’s first definition of AI art focuses on the recognition of AI-created works as ‘art’ by professional gatekeepers in the art world. This definition draws parallels to the Turing Test, suggesting that if experts cannot distinguish AI-generated art from historical or contemporary works, the AI has succeeded in creating art. However, Manovich argues that this approach falters due to the absence of a universal definition of art and the historical tendency of modern art to challenge and expand its own boundaries. From Duchamp’s readymades to contemporary installations, art has evolved by defying conventions — a process AI, trained on historical data, struggles to replicate.</p><p>Recent advancements in AI, such as diffusion models like DALL-E, do not fundamentally alter this critique. While these systems produce works that can astonish audiences and mimic historical styles with uncanny precision, their outputs remain bound by their training data. These tools excel at generating artifacts that resemble human art but rarely engage with the conceptual challenges or provocations that define modern and postmodern movements. Echoing Manovich, the reliance on professional validation as a criterion for defining AI art seems increasingly inadequate. Instead, we might ask whether AI can participate in the broader, boundary-pushing discourse of art, rather than merely imitating its established forms.</p><h3>2.</h3><blockquote><strong><em>What defines whether something is ‘AI’ is not a method but the amount and type of control we exercise over algorithmic process.</em></strong><em> — </em><a href="https://manovich.net/index.php/projects/defining-ai-arts-three-proposals"><em>Manovich</em></a></blockquote><p>The second definition shifts focus from the outputs of AI to the processes of creation, questioning whether machine learning and neural networks constitute a distinct paradigm within computer art. Manovich contrasts traditional computer art — which relies on explicit programming — with the ostensibly more autonomous processes of machine learning, where neural networks learn patterns from large datasets to generate new artifacts. Yet, he argues, the human influence remains pervasive: from designing algorithms to curating training datasets and selecting outputs, every stage of AI Art involves human decision-making.</p><p>This critique resonates even more strongly in light of recent AI advancements. While tools like ChatGPT and DALL-E showcase impressive generative capabilities, they also reveal the extent of human involvement in shaping their outputs. The curated datasets, prompt engineering, and output curation required for these tools underscore Manovich’s claim that machine learning is not inherently more autonomous or ‘intelligent’ than earlier forms of computer art. If anything, the apparent unpredictability of these systems masks a deeper dependence on human-defined parameters and aesthetic judgments.</p><p>The emphasis on process invites us to reconsider the nature of creativity itself. Is AI Art merely a collaborative tool, amplifying human creativity, or does it possess the potential for true authorship? While recent tools have — to a degree — ‘democratised’ artistic creation, allowing non-experts to produce visually compelling works, this accessibility raises questions about the relationship between skill, intention, and artistic value. Manovich’s argument suggests that the distinction between AI Art and earlier computer art lies not in the technology but in the degree of agency humans are willing to relinquish.</p><h3>3.</h3><blockquote><strong><em>Let’s teach computers to do something we humans can’t do — to move between different systems and aesthetics within a single work, or from work to a work in a series.</em></strong><em> — </em><a href="https://manovich.net/index.php/projects/defining-ai-arts-three-proposals"><em>Manovich</em></a></blockquote><p>In his third definition, Manovich envisions AI Art as a means of transcending human cognitive and aesthetic limitations. He critiques current AI systems for their reliance on systematicity and predictability, which mirror longstanding human artistic tendencies. Instead, he challenges AI to break free from these constraints and explore entirely new forms of aesthetic logic that humans cannot conceive. This vision positions AI as a collaborator in expanding the horizons of artistic possibility, creating works that are not merely novel but transformative.</p><p>Recent developments hint at this potential by combining styles and generating hybrid forms, but they stop short of the radical innovation Manovich describes. These tools remain tethered to human expectations of coherence and recognizability, reinforcing rather than challenging aesthetic norms. To truly transcend human constraints, AI Art would need to move beyond mimicry and synthesis, creating artefacts that challenge our perceptions and expand our understanding of art’s possibilities.</p><p>This vision raises profound questions about the role of AI in cultural production. Can AI create works that are both compelling and incomprehensible to humans, existing in a space between randomness and systematicity? And if such works were created, how would we recognize their value? Manovich’s definition urges us to embrace AI not as a tool for refining human aesthetics but as a means of exploring the uncharted territories of creativity.</p><blockquote><strong><em>AI art is [a] type of art that we humans are not able to create because of the limitations of our bodies, brains, and other constraints. </em></strong><em>— </em><a href="https://manovich.net/index.php/projects/defining-ai-arts-three-proposals"><em>Manovich</em></a></blockquote><p>Manovich’s three definitions of AI Art provide a critical framework for understanding the evolving relationship between artificial intelligence and artistic creation. His arguments highlight the limitations of current approaches, emphasizing the need for a broader conceptual framework that accounts for the complexities of human-AI collaboration and the transformative potential of AI. While recent advancements demonstrate the remarkable capabilities of AI, they also reinforce many of the critiques raised in the article. These tools excel at mimicry and synthesis but remain constrained by human-defined parameters and expectations.</p><p>The future of AI Art lies not in its ability to imitate or extend human creativity but in its potential to transcend it. By challenging the boundaries of what art can be and exploring new forms of aesthetic logic, AI can become a partner in expanding the possibilities of artistic expression. Manovich’s call for a radical reimagining of AI art invites us to move beyond questions of validation and process and toward a vision of art that pushes the limits of both human and machine creativity. In doing so, we may discover not only new forms of art but new ways of understanding ourselves and the world around us.</p><p><a href="https://medium.com/the-ai-and-arts-forum/automatic-art-the-post-knowledge-economy-cf2bec2e4e6d">Automatic Art &amp; the Post-Knowledge Economy?</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8d7a1e63c20b" width="1" height="1" alt=""><hr><p><a href="https://medium.com/electronic-life/ai-arts-reflections-on-lev-manovichs-3-definitions-8d7a1e63c20b">AI Arts: Reflections on Lev Manovich’s 3 Definitions</a> was originally published in <a href="https://medium.com/electronic-life">Electronic Life</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>