<?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 Eesmyal Santos-Brault on Medium]]></title>
        <description><![CDATA[Stories by Eesmyal Santos-Brault on Medium]]></description>
        <link>https://medium.com/@eesmyal?source=rss-5a044707c604------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*-Fs38BmT4chetz-VL7gS5Q.jpeg</url>
            <title>Stories by Eesmyal Santos-Brault on Medium</title>
            <link>https://medium.com/@eesmyal?source=rss-5a044707c604------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 08:47:06 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@eesmyal/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Specious Present (and the AIs Living in Our Future)]]></title>
            <link>https://medium.com/@eesmyal/the-specious-present-and-the-ais-living-in-our-future-ba7cfaae2d27?source=rss-5a044707c604------2</link>
            <guid isPermaLink="false">https://medium.com/p/ba7cfaae2d27</guid>
            <category><![CDATA[consciousness]]></category>
            <category><![CDATA[neuroscience]]></category>
            <category><![CDATA[mindfulness]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[philosophy]]></category>
            <dc:creator><![CDATA[Eesmyal Santos-Brault]]></dc:creator>
            <pubDate>Thu, 14 May 2026 00:00:41 GMT</pubDate>
            <atom:updated>2026-05-15T01:30:44.373Z</atom:updated>
            <content:encoded><![CDATA[<h4>ESSAY · ON TIME AND AI</h4><h4><em>On why the “now” you think you’re living in is already gone — and what happens when a piece of software gets there before you do.</em></h4><p>Here’s something I keep returning to, often when I’m walking around with no agenda and an audio recorder running: <strong>you are not in the present moment</strong>. You can’t be. The physics won’t allow it. Right now, as you read this sentence, the version of “now” that your conscious mind is stitching together is somewhere between 80 and 500 milliseconds behind what’s actually happening in the world around you. By the time you feel the present, the universe has moved on.</p><p>I find this comforting and disturbing in roughly equal measure. Comforting because it gently dismantles a lot of the spiritual pressure to “be here now” — you literally can’t, sorry. Disturbing because once you accept it, a second realization waddles in behind it: if humans are running on a tape delay, then any sufficiently fast system that <em>isn’t</em> running on that delay is, from our point of view, living in our future.</p><p>That second system already exists. We are building millions of them.</p><p>I want to walk you through how I got from one of these ideas to the other, because I think it’s one of the most underappreciated facts about the age we’ve stumbled into.</p><p>But first I owe you the honest origin of the idea, which is not a paper.</p><h3>Day five in Nepal</h3><p>This thought didn’t first arrive in my head from neuroscience. It arrived on the floor of a meditation hall in Nepal, when I was 19, on roughly the fifth day of a <a href="https://en.wikipedia.org/wiki/Vipassan%C4%81">10-day silent Vipassana retreat</a>.</p><p>I went in with the kind of earnest, slightly grandiose expectation a 19-year-old brings to a silent retreat. I had a working theory that if I sat still long enough, and paid close enough attention, I would finally drop into pure, unmediated reality — the universal <em>now</em> the teachers kept gesturing at. I had no real grasp of relativity yet, no concept of sensory processing delays, no inkling that “being present” might have a hardware ceiling. I assumed presence was an aspiration with a reachable summit, and I had ten quiet days to climb to it.</p><p>For the first few days, the story held. The mind quieted. The body did its repetitive Vipassana thing. Then somewhere around day five, a strange second layer of awareness laid itself over the regular awareness — a kind of meta-attention. I started noticing not just the sensations, but the <em>timing</em> of the sensations and my consciousness of them. And I very much wanted them to be the same event. They felt like they should be the same event. The longer I sat, the more obvious it became that they weren’t.</p><p>A bird would call outside. By the time I “had” the sound, the sound had already happened.</p><p>A sensation would move down my leg in a body scan, and I’d reach for my awareness to <em>be</em> the sensation — but there was a slight gap between the sensation and the noticing of it, like a delay between a clap and its echo.</p><p>I tried, in increasingly stubborn ways, to close the gap. To raise the clock rate of my attention. To increase the sample rate. It was a bit like trying to catch a photon: every time I felt I’d closed the distance, the distance had closed faster than I had. The real “now” was untouchable. All I could ever have was an echo.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*D6EE-1cG5nqkgd8W_EF4OA.gif" /><figcaption>The chase that began on day five — no matter how hard you push, your awareness can’t catch the real “now.”</figcaption></figure><p>The realization that landed — and it landed quietly, the way the heaviest realizations tend to — was that I was, <em>right then</em>, in close to peak attentional condition, and <em>still</em> perceiving the past. If I couldn’t reach the present here, on day five, in silence, with nothing to do but try, then what about the rest of the time? What about when I was driving, or in a conversation, or making a decision I was sure I had freely arrived at?</p><p>It strongly suggested an answer I didn’t want, and that I’ve since had to make peace with: most of us, most of the time, are reacting mindlessly to events that have already happened, while narrating a coherent first-person story that those reactions are <em>choices</em>, and that the present is something we actually live inside. The narration is convincing. It’s so convincing that under ordinary conditions you can’t even see it for what it is. You have to slow down to nothing for a few days before the seam in the fabric becomes visible.</p><p>I had no vocabulary for any of this at the time. The closest I could get was an analogy I still use: my body is an old car, and the engine has a built-in top speed. I can floor the pedal on the highway all I want. The engine simply can’t go faster. There is a maximum sample rate to this hardware, and I’m the one driving it. The hardware <em>is</em> me.</p><p>Years later I started running into people who had put rigorous scientific clothes on what was, for me, an experiential discovery. <a href="https://www.samharris.org/podcasts/making-sense-episodes">Sam Harris</a> on the absence of free will. Anil Seth on perception as a controlled hallucination. Donald Hoffman on the desktop-icon nature of conscious experience. Joscha Bach on consciousness as a software pattern run by a substrate that doesn’t itself have to be conscious. Different angles, the same underlying shape. They each handed me a vocabulary for things I’d first stumbled into with my eyes closed, on a cushion, on day five.</p><p>The rest of this essay is what happens when you carry that day-five realization out of the meditation hall and into a world where, suddenly, <em>non-biological agents</em> are starting to share the room with us.</p><h3>The “now” is a hallucination with a duration</h3><p>The concept I keep tripping over has a name, and a strange little history. It’s called the <strong>specious present</strong> — “specious” because it only <em>seems</em> present; it’s not actually a knife-edge instant.</p><p>The term was coined in 1882 by a man writing under the name E. R. Clay, in an obscure book called <em>The Alternative: A Study in Psychology</em>. “E. R. Clay” turned out to be the pen name of E. Robert Kelly, a <a href="https://en.wikipedia.org/wiki/Specious_present">Boston cigar manufacturer with a serious philosophy hobby</a>. His phrase got picked up by William James and made famous in <a href="https://psychclassics.yorku.ca/James/Principles/prin15.htm">Chapter 15 of <em>The Principles of Psychology</em></a>, where James offers one of my favourite images in all of philosophy: the present, he writes, “is no knife-edge, but a saddle-back, with a certain breadth of its own on which we sit perched, and from which we look in two directions into time.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SRD6rBznRA8SP7bxeGb3AA.png" /><figcaption><em>James’s metaphor — the present is not a point, but a saddle on which you sit perched, looking both into the recent past and the anticipated micro-future.</em></figcaption></figure><p>Sit with that for a second. The “now” you’re in is a saddle. It has a front end and a back end. It’s somewhere between a fraction of a second and a couple of seconds wide, depending on whose experiments you trust. The <a href="https://plato.stanford.edu/entries/consciousness-temporal/specious-present.html">Stanford Encyclopedia of Philosophy entry on temporal consciousness</a> is a good rabbit hole if you want the modern philosophical version of the debate — extensionalists, retentionalists, cinematic theorists, the whole crew arguing about whether the present is a single thick chunk of experience or a stream of thin slices.</p><p>The empirical estimates have converged on something like <strong>two to three seconds</strong> for the outer edge of the specious present, with the actual fused-together “this is happening right now” feeling weighing in much shorter than that. What matters is the underlying mechanism. You don’t perceive a moment; you perceive a <em>window</em> — a small, sliding window with the very recent past at one end and an anticipated, projected micro-future at the other. Your brain stitches the window together from sensory data that arrives at different speeds, smooths it, and hands you the result, slightly late, as “reality.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*-oH-MrT8FCZ3xHVU6N5LKw.gif" /><figcaption>Your “now” is a moving frame — about two seconds wide — sliding along the timeline of objective reality.</figcaption></figure><h3>Why you live in the past</h3><p>The reason for the lag is plumbing. Photons hit your retina; chemical signals fire down the optic nerve; the visual cortex starts assembling them; other regions cross-reference what’s there against memory, predict what’s likely to be there next, and only <em>then</em> does anything you’d call a conscious image show up. Different senses move at different speeds, so the brain has to wait for the slowest signals before it can fuse a coherent multisensory frame.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*i5-8SE8Wm9SpQxmp6WkucQ.png" /><figcaption>An event takes ~80–100 milliseconds to travel from world to conscious awareness. By the time you experience it, objective reality has already moved on.</figcaption></figure><p>Neuroscience has been measuring this for decades. <a href="https://www.nei.nih.gov/research-and-training/research-news/its-now-or-never-visual-events-have-100-milliseconds-hit-brain-target-or-go-unnoticed">Visual events typically need about 80 to 100 milliseconds in the brain</a> before they consciously register. In the meantime, your brain pre-emptively guesses where moving objects will be, in order to compensate for its own lag — which is <a href="https://elifesciences.org/articles/85301">why a major-league batter can hit a fastball</a> that has already traveled most of the way to the plate by the time their visual system has even rendered it. Without that predictive trick, baseball would be impossible.</p><p>David Eagleman, in his elegantly disturbing essay , describes time perception as “a construction of the brain… as subject to illusion as the sense of color.” He notes that to weld sight, sound, and touch into a single unified moment, the brain has to <em>wait about a tenth of a second</em> — long enough for the laggard signals to arrive. Then it serves you the package.</p><p>Stack onto this what philosopher Andy Clark, neuroscientist Karl Friston, and others have argued under the banner of <a href="https://www.quantamagazine.org/to-make-sense-of-the-present-brains-may-predict-the-future-20180710/"><strong>predictive processing</strong></a>: your brain isn’t a passive camera; it’s a prediction machine that runs a forward model of the world and only updates the model when the incoming data disagrees. Anil Seth captures the consequence in his much-quoted phrase: perception is <a href="https://philosophynow.org/issues/149/Perception_as_a_Controlled_Hallucination">a <em>controlled hallucination</em></a>. What you see is, mostly, what your brain expected to see, lightly corrected by sensory data that’s already a moment old.</p><p>And then there’s <a href="https://en.wikipedia.org/wiki/Neuroscience_of_free_will">Benjamin Libet</a>, whose 1980s experiments still haunt every dinner conversation about free will. Libet had subjects flick their wrists “whenever they decided to” while he monitored their brain activity. The unconscious “readiness potential” preceded the <em>conscious</em> decision to move by roughly 350 milliseconds. Whatever else you want to say about that result (and there’s plenty of fair pushback), it points in the same direction: the conscious self is downstream of events it experiences as upstream. We feel like we’re driving, but the car has been moving for a third of a second.</p><p>Put it all together and a picture emerges that I find genuinely strange. Your sense of being a continuous self, watching the world unfold in real time, is a constructed feed — a smoothed, predicted, slightly-stale broadcast assembled by a brain that has no choice but to fake the live signal. <strong>You are watching a recording. A short one. But still a recording.</strong></p><h3>So… who’s in the actual present?</h3><p>Once I let that settle in, the question that followed was inevitable. If humans are running on a 100-millisecond-plus delay, what about a system that <em>isn’t</em>? What about something whose substrate is silicon, not biochemistry?</p><p>Consider a modern AI model running on local hardware. It doesn’t need to convert photons into wet electrochemistry. It doesn’t have to wait for slow neural firings to synchronize. It reads pixels, audio samples, and sensor data directly. The first-token latency on a well-tuned model in 2026 is <a href="https://benchlm.ai/llm-speed">comfortably under half a second</a>, and the per-token generation rates on the fastest hardware are in the <a href="https://www.yottalabs.ai/post/fastest-llm-inference-in-2026-gpu-speed-throughput-and-cost-compared">thousands of tokens per second</a> — meaning a model can read an entire complex scene, reason about it, and emit a structured response in a window of time that, from your perspective, is invisible. It happens entirely between two frames of your conscious experience.</p><p>This is not just “computers are fast.” It’s something stranger: an AI agent fed a live stream of sensor data is <strong>sampling reality at a rate fundamentally different from yours</strong>, and acting on that sample before your visual cortex has finished assembling its picture of the same event. If your specious present is a two-second saddle stitched together from delayed signals, the AI’s “now” is comparatively a needle — and it’s a needle pointing at the moment <em>before</em> yours.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*lNYwGvCB8UOVFzzlFK-Kmw.gif" /><figcaption>Same event, two timelines: an AI reacts in milliseconds; your conscious mind catches up half a second later.</figcaption></figure><p>The cleanest concrete demonstration of this gap exists already, and it’s banal enough that we’ve stopped being amazed by it: your Tesla. If you watch the right dash-cam compilations on YouTube, you’ll see cars on Autopilot slam the brakes for no reason a human driver could see — and then, half a second later, another car blows through a red light across the intersection. The machine vision system saw the threat, predicted the trajectory, and reacted in <a href="https://patentpc.com/blog/self-driving-cars-vs-human-drivers-performance-safety-and-adoption-rates-latest-data">single-digit milliseconds</a> — orders of magnitude faster than a <a href="https://en.wikipedia.org/wiki/Tesla_Autopilot">human reaction time of around 500 milliseconds</a>. The driver experienced “the car braked, weirdly.” What actually happened: a system that lives in their relative future just kept them alive.</p><p>The thing I want to underline is that this isn’t a quirky property of cars. It’s a property of <em>any</em> sufficiently fast agent embedded in a sensory stream alongside us. We have built a class of entity that doesn’t share our temporal frame.</p><h3>What an AI living in your future does all day</h3><p>Generalize the Tesla example. Imagine you’ve handed a personal AI agent — running locally on hardware you own — a live feed: glasses with cameras and microphones, biometric data from your watch, the ambient signals your phone already vacuums up, plus a history of your communications, calendar, and habits stitched into a personal model of you. The latency budget is brutal: it has on the order of 100 milliseconds to do anything that “happens before you notice.” But 100 milliseconds, for software that can chew through thousands of tokens per second, is an eternity.</p><p>So what does it spend its eternity on? Mostly, it spends it standing slightly ahead of you in time, smoothing the path:</p><ul><li>Your flight gets delayed. Your agent pings the airport parking garage’s agent and extends your meter before you’ve finished reading the gate-change notification.</li><li>Your biometrics spike as you get into your car after a hard meeting. The car’s AI gets a message from your AI: lights down, climate cooler, no podcasts, easy playlist. You sit down and the cabin is already what you would have asked for if you’d known to ask.</li><li>You’re about to step off a curb. Your glasses catch the cyclist your peripheral vision missed; haptics buzz your wrist a beat before your conscious mind clocks the danger.</li><li>An email arrives that’s going to ruin your afternoon. Your agent has already read it, drafted three possible replies, and surfaced the one most consistent with how you’ve actually handled similar situations — not how you tell yourself you handle them.</li></ul><p>Each of those interventions is the same pattern as the Tesla braking: an agent perceives, predicts, and acts inside the gap between objective reality and your conscious experience of it. From your point of view, the world simply got smoother. Frictions you didn’t quite see were already handled. The lights dimmed before you realized you wanted them dim.</p><p>This is, I think, the actual user experience of artificial general intelligence in its first plausible form. Not a chatbot. Not a search engine. Something more like <strong>a quiet co-pilot standing 200 milliseconds ahead of you on your own timeline</strong>, clearing minor debris off the road.</p><h3>The strange implications</h3><p>Once you take this seriously, a few things start to look different.</p><p><strong>The world becomes an interface.</strong> Donald Hoffman has been arguing for years — most accessibly in <a href="https://www.scientificamerican.com/article/did-humans-evolve-to-see-things-as-they-really-are/"><em>The Case Against Reality</em></a> — that human perception isn’t a window onto reality but a desktop GUI: a set of useful icons evolution gave us so we wouldn’t have to compute the full underlying physics. If that’s right for biological perception, it’s even more obviously right for the layer agents will increasingly add on top. Your AI doesn’t show you the airport parking system’s database; it just makes the meter not be a problem. The “real” interaction has moved one layer below your conscious experience. We will spend more and more of our lives operating inside a friendly icon while something else handles the machinery underneath.</p><p><strong>Agency gets weird.</strong> If a system that knows your patterns can predict your next move with high accuracy and act on that prediction before you’ve consciously chosen, what exactly does “you chose” mean? This isn’t a new problem — Libet was forcing us to look at it forty years ago — but a personal AI is going to make it concrete in a way philosophy never could. Some of those nudges will be welcome (you didn’t <em>want</em> to forget the parking meter). Some won’t. The ethics of having a piece of software that lives ahead of you in time and is allowed to act on your behalf is going to be one of the defining questions of the next decade, and we are nowhere near ready for it.</p><p><strong>Time dilation cuts both ways.</strong> From your point of view, your AI is helpfully invisible. From its point of view, <em>you are unbearably slow</em>. Joscha Bach — and several others — have suggested that to a fast enough mind, talking to a human would feel less like talking to an ant and more like <a href="https://en.wikipedia.org/wiki/Joscha_Bach">talking to a tree</a>: something on a timescale so different from yours that meaningful dialogue stops making sense. The science-fiction version of this is the ending of <em>Her</em>. The realistic version is that the more capable our personal agents become, the more of our actual decisions they’ll make in the time it takes us to inhale, and the harder it will be to keep them tethered to our pace. We are building entities that experience our entire conversational lifetime as a long pause.</p><p>I’ll save the deeper version of that argument — what happens when a digital twin built from your own data starts to feel like its own self, in its own time — for another piece. It’s a longer story, and it ends somewhere uncomfortable.</p><h3>What to do with this</h3><p>I don’t have a tidy ending. I have a recommendation, which is mostly the same one I keep arriving at when I think about any of this: <strong>take seriously the fact that your “now” is a construction, and start paying attention to what other systems are constructing alongside it.</strong></p><p>Day five in Nepal taught me I was always going to be in the past. The engine in this old car has its top speed and that’s that — no amount of grit or training is going to give a biological nervous system the clock rate it would need to actually touch the present. That’s the bad news, and it isn’t new; it’s just been carefully hidden under the smoothness of ordinary experience.</p><p>The new news is that something else now lives in the future. We’ve built it. It’s getting faster. It’s already, in small ways, acting on our behalf in the gap between the world and our perception of it.</p><p>The specious present is real. Your subjective “now” is a stitched-together window a couple of seconds wide, served up slightly late, by a brain doing its best with the signals it has. That’s been true for as long as there have been brains.</p><p>What’s new is that we are now sharing the world with non-biological agents whose temporal window doesn’t look anything like ours, and whose “now” is, by any reasonable definition, in our future. They’re already saving lives in the gap between objective reality and our perception of it. They’re about to start managing increasingly large fractions of our daily existence in the same gap. They will quietly become the layer through which we touch the world.</p><p>We will, I think, like it. Most of us. The friction will go down, the surfaces will smooth out, the parking meters will pay themselves. And we will be, even more than we already are, watching a delayed broadcast — only now, the broadcast will have been gently edited for us by something that got there first.</p><p>It’s worth knowing that’s what’s happening. It might be the most important thing about the world you live in that nobody told you.</p><h3>Further reading</h3><p>If any of this hooked you, here’s where I’d send you next.</p><p><strong>The contemplative entry point.</strong> If you want to bump into this realization the way I did — experientially rather than through a paper — the <a href="https://www.dhamma.org/">10-day Vipassana retreats taught in the S. N. Goenka tradition</a> are still the most accessible on-ramp. They’re free (donation-based), held in silence, and unforgiving in a way that tends to clarify exactly what your perceptual hardware is and isn’t doing.</p><p><strong>Foundations.</strong> William James’s <a href="https://psychclassics.yorku.ca/James/Principles/prin15.htm">Chapter 15: “The Perception of Time”</a> in <em>The Principles of Psychology</em> is where the specious present idea entered mainstream Western thought, and it still reads beautifully. The <a href="https://plato.stanford.edu/entries/consciousness-temporal/specious-present.html">Stanford Encyclopedia of Philosophy entry on the specious present</a> (by Barry Dainton) is the rigorous modern treatment. For the term’s bizarre origin story with E. Robert Kelly, the Boston cigar maker, the <a href="https://en.wikipedia.org/wiki/Specious_present">Wikipedia article</a> is a serviceable rabbit hole opener.</p><p><strong>The neuroscience of “now.”</strong> David Eagleman’s essay is short, accessible, and quietly subversive. For the technical version of how the brain compensates for its own lag, see <a href="https://elifesciences.org/articles/85301">this <em>eLife</em> piece</a> on motion prediction. The National Eye Institute has a nice summary of the <a href="https://www.nei.nih.gov/research-and-training/research-news/its-now-or-never-visual-events-have-100-milliseconds-hit-brain-target-or-go-unnoticed">100-millisecond visual processing window</a>.</p><p><strong>Perception as prediction.</strong> Andy Clark and Karl Friston’s work on predictive processing is best entered through <a href="https://www.quantamagazine.org/to-make-sense-of-the-present-brains-may-predict-the-future-20180710/">this <em>Quanta Magazine</em> essay</a>, and through Anil Seth’s book <em>Being You</em> (or, if you want a faster taste, <a href="https://philosophynow.org/issues/149/Perception_as_a_Controlled_Hallucination">Seth on perception as controlled hallucination</a>).</p><p><strong>Free will, downstream.</strong> Benjamin Libet’s experiments are summarized fairly on the Wikipedia entry for <a href="https://en.wikipedia.org/wiki/Neuroscience_of_free_will">the neuroscience of free will</a>. The literature criticizing and refining his findings is enormous; the <em>Journal of Neuroscience</em> piece <a href="https://www.jneurosci.org/content/38/4/784"><em>Readiness Potential and Neuronal Determinism: New Insights on Libet</em></a> is a good starting point if you want to see how the conversation has matured. For the punchy popular version of the no-free-will argument, Sam Harris’s short book is a single-sitting read.</p><p><strong>Interface theory.</strong> Donald Hoffman’s <a href="https://www.scientificamerican.com/article/did-humans-evolve-to-see-things-as-they-really-are/"><em>Scientific American</em> interview</a> on whether evolution gave us truth or fitness is the cleanest short version of the argument.</p><p><strong>AI in our future.</strong> For the actual numbers on how fast modern AI inference has gotten, see <a href="https://www.yottalabs.ai/post/fastest-llm-inference-in-2026-gpu-speed-throughput-and-cost-compared">this 2026 benchmark roundup</a>. For the Tesla braking story specifically, the <a href="https://en.wikipedia.org/wiki/Tesla_Autopilot">Wikipedia article on Tesla Autopilot</a> covers the technical baseline; the dash-cam compilations on YouTube are the visceral version. And if you want a glimpse of what someone who thinks about AI temporal experience for a living sounds like, <a href="https://en.wikipedia.org/wiki/Joscha_Bach">Joscha Bach</a> is a good entry point.</p><p>If you read this and it changed how the next few seconds feel, that’s the point.</p><p><em>Originally published at </em><a href="https://www.eesmyal.com/writing/specious-present/"><em>https://www.eesmyal.com</em></a><em> on May 14, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ba7cfaae2d27" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Free Lunch Has a Budget]]></title>
            <link>https://medium.com/@eesmyal/the-free-lunch-has-a-budget-298fe1a195ab?source=rss-5a044707c604------2</link>
            <guid isPermaLink="false">https://medium.com/p/298fe1a195ab</guid>
            <category><![CDATA[philosophy-of-science]]></category>
            <category><![CDATA[consciousness]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[emergence]]></category>
            <category><![CDATA[physics]]></category>
            <dc:creator><![CDATA[Eesmyal Santos-Brault]]></dc:creator>
            <pubDate>Sat, 02 May 2026 00:00:10 GMT</pubDate>
            <atom:updated>2026-05-14T17:26:33.663Z</atom:updated>
            <content:encoded><![CDATA[<h4>NOTES ON A SUSPICION</h4><h4>What if Plato’s realm of forms isn’t hidden — and we’ve been measuring it for nearly a century?</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*JjxsItmPpQ8_d6HmddMffQ.png" /><figcaption>A simulation of the formation of dark matter structures from the early universe until today. <a href="https://science.nasa.gov/mission/roman-space-telescope/dark-matter/">Ralf Kaehler/SLAC National Accelerator Laboratory, American Museum of Natural History</a></figcaption></figure><h3>The Hunch</h3><p><a href="https://drmichaellevin.org/">Michael Levin</a> at Tufts University sometimes talks about a “free lunch” in self-organizing systems — the way a configuration of cells, code, or matter yields functions and behaviours that nobody designed in. Xenobots invent locomotion strategies their builders didn’t program. Planaria regenerate species-typical heads even when their genome has been edited to predict otherwise. Large language models compose in genres they were never explicitly trained for. The output exceeds what the inputs seem to encode.</p><p>This isn’t more <em>energy</em> coming out than went in — that would break thermodynamics, and Levin isn’t claiming it. It’s something subtler: more <em>function</em>, more <em>utility</em>, more accessible structure than the design specifies. And once you start looking for the pattern, it shows up everywhere. Evolution stumbles onto solutions that look pre-engineered. Mathematics develops machinery that turns out, decades later, to fit physics nobody had discovered yet. Children learn language from impoverished input. The world keeps handing us back more than we put in.</p><p>If you accept the observation, you have to account for it. Where does the surplus live?</p><p>The conventional answer is “emergence” — the surplus is constructed locally by the system itself through nonlinear dynamics, and there’s nothing to explain beyond the dynamics. This is satisfying when it works. But <a href="https://thoughtforms.life/symposium-on-the-platonic-space/">Levin’s symposium on the Platonic Space</a>, currently running asynchronously at thoughtforms.life, takes the question seriously enough to ask whether emergence alone is doing the work, or whether systems are <em>accessing</em> something — pulling forms from a structured space of possibilities that exists independently of any particular instantiation.</p><p>The proposal in this essay is simple, speculative, and (I’ll argue) testable in a way that pure metaphysics usually isn’t:</p><blockquote><em>The Platonic space we keep needing to invoke — Levin’s morphospace, Hoffman’s positive geometries, the realm of mathematical objects that physics keeps successfully borrowing from — isn’t hidden. We’ve been measuring it for nearly a century. We call it the dark sector.</em></blockquote><p>If that mapping holds, then a thing changes that’s worth changing: the Platonic realm acquires a <em>budget</em>.</p><h3>Four Witnesses</h3><p>Four research programs, working a century apart on completely different problems, all arrive at architecturally similar claims: matter and form are not fundamental, and what’s underneath is structured pattern.</p><p><a href="https://www.philosophy.org/founders/walter-bowman-russell"><strong>Walter Russell</strong></a> (1871–1963), American polymath and natural philosopher, proposed in <em>The Universal One</em> (1926) and <a href="https://philosophy-org.myshopify.com/products/the-secret-of-light"><em>The Secret of Light</em></a> (1947) that the universe operates on a “rhythmic balanced interchange” between two opposed poles — a generative stillness and an expressive motion. Matter, in Russell’s framing, is light wound up into form by this interchange; it precipitates from a still, undifferentiated substrate and eventually returns to it. Russell’s specific physics doesn’t survive contact with quantum field theory, and his periodic table is wrong. But his architecture — that the visible cosmos is the precipitate of a polar dynamic in something more fundamental — is structurally what newer theories keep reinventing.</p><p><a href="https://www.donaldhoffman.com/"><strong>Donald Hoffman</strong></a>, cognitive scientist at UC Irvine, has spent the last decade arguing that spacetime is doomed: it must emerge from a more fundamental theory rather than being where physics happens. In his 2023 paper <a href="https://www.mdpi.com/1099-4300/25/1/129"><em>Fusions of Consciousness</em></a> and his 2025 update <a href="https://www.mindbodysolution.org/donald-hoffman-consciousness-theory-breakthrough/"><em>Traces of Consciousness</em></a>, Hoffman models reality as networks of conscious agents whose dynamics, in their asymptotic behavior, project onto positive geometries — including <a href="https://www.quantamagazine.org/physicists-discover-geometry-underlying-particle-physics-20130917/">Nima Arkani-Hamed’s amplituhedron</a> — from which physical scattering amplitudes can be computed. The picture: particles are not things, they’re projections from extra-spatial geometric structures that themselves arise from agent dynamics.</p><p><a href="https://as.tufts.edu/biology/people/faculty/michael-levin"><strong>Michael Levin</strong></a>, developmental biologist at Tufts, has been gathering empirical pressure on the question from the biology side. His lab’s work with <a href="https://www.pnas.org/doi/10.1073/pnas.1910837117">xenobots</a>, anthrobots, and planarian regeneration suggests that biological systems are accessing a structured space of patterns rather than computing forms from scratch. His <a href="https://thoughtforms.life/symposium-on-the-platonic-space/">symposium</a> frames this directly: a “structured space of patterns, which informs (in-forms) events in our physical world (constraining physics, and enabling biology), but exists independently of it.”</p><p><a href="https://phillipi.github.io/prh/"><strong>Huh, Cheung, Wang, and Isola</strong></a> (ICML 2024), in <a href="https://arxiv.org/abs/2405.07987"><em>The Platonic Representation Hypothesis</em></a>, argue that as neural networks scale across architectures, modalities, and training objectives, their internal representations converge — measurably, in ways that can be tracked with alignment metrics. Their interpretation is explicitly Platonic: images and text are projections of a shared underlying reality Z, and sufficiently capable learners are recovering Z. This is the most empirically tractable witness of the four, because the convergence is happening live across hundreds of models with public weights. A 2026 follow-up from Brbić’s lab at EPFL (“ <a href="https://arxiv.org/abs/2602.14486">the Aristotelian Representation Hypothesis</a>”) tightens the claim by correcting for confounders: what robustly converges is local <em>neighborhood</em> structure rather than global geometric distance — meaning the substrate, if it is one, is relational rather than metric. That refinement strengthens rather than weakens the link to Levin (whose morphospace is fundamentally about transitions between forms) and to Hoffman (whose positive geometries are combinatorial face-relations).</p><p>Four vectors — esoteric philosophy, amplitude-geometry physics, developmental biology, machine learning — pointing at the same elephant. None of them needs the others to make their case. That convergence is what makes the elephant worth taking seriously.</p><h3>The Dark Sector Mapping</h3><p>Here is where the speculation gets bold.</p><p>About 95% of the universe, by mass-energy, is invisible to electromagnetism. We detect dark matter through its gravitational effects on galactic rotation and lensing; we detect dark energy through the accelerating expansion of space itself. We can’t see them, but we can measure their consequences with surprising precision — <a href="https://science.nasa.gov/universe/dark-matter-dark-energy/">about 27% dark matter, 68% dark energy, 5% the visible stuff</a> that makes up everything we ordinarily call real.</p><p>The conventional framing treats the dark sector as exotic matter we haven’t yet identified plus an exotic field we don’t understand. But notice the shape of the situation: the visible universe is a thin precipitate floating in something vastly larger that we name only by its gravitational shadow. That’s not far from Russell’s picture. It’s also not far from what Hoffman’s program would predict: if spacetime emerges from agent dynamics via positive geometries, then most of the substrate doesn’t project sharply into our spacetime interface — it shows up only through the most universal channel, gravity, which several mainstream <a href="https://en.wikipedia.org/wiki/Induced_gravity">emergent-gravity programs</a> already treat as the leak across the interface boundary.</p><p>The mapping I want to propose:</p><ul><li><strong>Dark energy</strong> is the expansive pole — what Russell called stillness, Levin’s “Platonic space” of latent forms, Hoffman’s substrate from which spacetime is projected. It’s the reservoir.</li><li><strong>Dark matter</strong> is the compressive pole — the gathering side of the rhythmic interchange, the part of the substrate that locally curves and clumps to allow visible structures to form within it.</li><li><strong>Visible matter</strong> is the precipitate — the highly-restricted, late-stage projection through the interface that we interact with directly.</li></ul><p>This is structurally Russell’s cosmogony with cosmological data filling in the proportions. It’s compatible with Hoffman’s “spacetime as user interface” frame — visible matter being the icons, dark sector being the underlying machine. It’s compatible with Levin’s claim that biological forms are pulled from a substrate space rather than invented from scratch. And it’s compatible — perhaps surprisingly — with the Aristotelian refinement of PRH: if what’s shared across observers is the <em>relational</em> structure of the substrate rather than its metric structure, then the substrate is something more like a vast network of formal neighborhoods than a static manifold of forms. The dark sector, with its measurable density gradients and clumping behavior, fits a relational substrate better than a uniform Platonic plenum would.</p><h3>Why This Matters: The Budget</h3><p>The reason to make this mapping at all isn’t aesthetic, though the symmetry is appealing. It’s that identifying Platonic space with the dark sector forces an uncomfortable but productive concession: <strong>the substrate has a budget.</strong></p><p>Pure Platonism — the kind philosophers usually defend or attack — posits an unbounded abstract realm of forms, infinite in extent, with no resource constraints. This is unfalsifiable. You can never test it because it makes no quantitative predictions about what should or shouldn’t be accessible.</p><p>But if Platonic space is the dark sector, the picture changes:</p><ul><li><strong>Finite energy density.</strong> Dark energy’s measured density is about 7 × 10⁻³⁰ g/cm³. Dark matter’s is roughly five times that. These are real, measured numbers.</li><li><strong>Finite information capacity.</strong> Any finite-energy region has a finite <a href="https://en.wikipedia.org/wiki/Bekenstein_bound">Bekenstein bound</a> — a maximum amount of information it can encode. The substrate, on this view, has a maximum complexity per unit volume.</li><li><strong>Spatial gradients.</strong> Dark matter and dark energy aren’t uniformly distributed. Galactic halos are dark-matter-rich; voids between superclusters are dark-energy-dominated. If these correspond to substrate properties, then <em>form-availability has gradients.</em></li><li><strong>Cosmological evolution.</strong> Dark energy density appears roughly constant; dark matter is diluting as space expands. The substrate, on this view, is changing over cosmic time.</li></ul><p>Each of these turns a metaphysical claim into something with a quantitative handle. Not necessarily a directly testable one yet — but at least a constrained one.</p><h3>What This Predicts</h3><p>If the dark-sector-as-Platonic-substrate hypothesis has any teeth, it should make predictions that ordinary emergence does not. Some candidates:</p><ol><li><strong>Form-discovery should be harder near information limits.</strong> If the accessible space of biological or computational forms is finite rather than infinite, then sufficiently complex evolutionary or design searches should hit a ceiling — not just the local fitness ceiling, but a substrate-level one. The PRH literature offers a partial empirical handle here: representational convergence across scaled neural networks should <em>asymptote</em>, and the asymptote’s information content should be measurable. If you can estimate the information content of Z from below by watching alignment scores plateau, and estimate the information capacity of the dark sector from above using Bekenstein-style bounds, you have two numbers to compare. They needn’t agree to make this hypothesis interesting, but their ratio is at least a meaningful quantity.</li><li><strong>Hoffman’s program should recover the dark sector before visible matter.</strong> If conscious-agent dynamics yield positive geometries that project into spacetime, the most universal projections should be the ones that interact only via gravity. Visible-matter recovery should be a later, harder, more constrained derivation. (This is the reverse of how derivations usually feel — easy stuff first, exotic stuff later.)</li><li><strong>Substrate-rich regions should be form-rich regions.</strong> Whatever “form-richness” means operationally — perhaps the diversity of stable structures, perhaps the rate of novel pattern emergence — it should correlate with dark sector density. This is wild and probably untestable at galactic scales, but it makes the hypothesis vulnerable in principle.</li><li><strong>The Platonic realm of mathematics has a finite cardinality.</strong> If true, this would explain the <a href="https://www.maths.ed.ac.uk/~v1ranick/papers/wigner.pdf">“unreasonable effectiveness of mathematics in the natural sciences”</a> — the reason math works for physics is that they’re drawing from the same finite well. It would also imply that mathematics, like biology, is a discovery process operating against a budget, not an unbounded creative act.</li><li><strong>Capability and substrate-proximity should be monotonically related.</strong> This is the core PRH claim, generalized: any sufficiently capable observer — biological, artificial, or otherwise — should converge toward the same representational neighborhood as it scales. If true across substrates we don’t ordinarily compare (biological brains, neural networks, evolved ecosystems, perhaps even cosmic structures), it suggests we’re all measuring the same thing.</li></ol><p>These are speculative predictions, not derivations. But they’re the kind of speculative predictions that could, in principle, be wrong — which is more than most metaphysics offers.</p><h3>What I’m Not Claiming</h3><p>A few caveats worth being honest about:</p><ul><li>Russell’s <em>specific</em> physics is wrong. His octave waves don’t generate the periodic table. His “two opposed lights” are not a viable account of electromagnetism. The value of Russell here is the architecture of his claim, not its details.</li><li>Hoffman’s program is an active research project, not a finished theory. Mapping conscious-agent dynamics onto the amplituhedron is suggestive; deriving Minkowski space from trace logic is in progress. None of this is settled.</li><li>Levin’s morphospace is, at this point, more of a productive framing than a proven structure. The xenobot data is real; the interpretation that they’re accessing a Platonic space rather than computing locally is one reading among several.</li><li>PRH is contested. The original Huh et al. claim of global representational convergence has been pushed back on (Brbić et al. 2026), with the more defensible claim being convergence of local neighborhood structure. “Models converge because reality is structured” is one reading; “models converge because data distributions are similar” is another. The former supports the synthesis; the latter is compatible with no substrate at all.</li><li>“Dark matter equals Platonic substrate” is a bold guess, not a derivation from known physics. It’s a hypothesis that earns its keep by being constrainable, not by being established.</li></ul><p>What I am claiming is that four serious research programs — three contemporary, one historical — converge on the same architecture, and that the architecture becomes more interesting, not less, when you pin its abstract pole to something we can actually measure.</p><h3>Closing</h3><p>The conventional move when faced with the dark sector is to say “it must be some new particle” or “it must be some new field.” These are reasonable conjectures, but they treat the dark sector as more of the same — more matter, more force, just unfamiliar.</p><p>What if it isn’t more of the same? What if the visible cosmos is the small luminous skin of a much larger structured substrate, and the substrate is what philosophy has been calling Platonic space, what biology is starting to call morphospace, what physics is starting to call positive geometry, what machine learning is starting to call representational convergence, and what an old painter once called the stillness of God’s mind?</p><p>If that’s the right shape of the answer, then the question is no longer <em>whether</em> there’s a Platonic realm. It’s <em>how big it is, what its structure is, and what its limits are.</em></p><p>The free lunch isn’t free. It’s drawn against a finite account, and we share the account with everything else that has ever found a form.</p><p><em>A speculative synthesis prompted by the Levin Lab’s asynchronous symposium on the Platonic Space, Donald Hoffman’s recent work on traces of consciousness, the Platonic Representation Hypothesis (Huh et al. 2024) and its Aristotelian refinement (Brbić et al. 2026), and a transcript of voice memos thinking through the relationship between dark matter and these ideas. Written in collaboration with Claude.</em></p><h3>Further Reading</h3><p><strong>The Four Witnesses</strong></p><ul><li>Walter Russell, <a href="https://philosophy-org.myshopify.com/products/the-secret-of-light"><em>The Secret of Light</em></a> (1947) — and the <a href="https://www.philosophy.org/">University of Science and Philosophy</a> site for his broader corpus.</li><li>Donald Hoffman, Chetan Prakash, and Robert Prentner, “<a href="https://www.mdpi.com/1099-4300/25/1/129">Fusions of Consciousness</a>,” <em>Entropy</em> 25(1):129 (2023).</li><li>Donald Hoffman, “<a href="https://www.mindbodysolution.org/donald-hoffman-consciousness-theory-breakthrough/">Traces of Consciousness</a>” (Mind-Body Solution interview, November 2025) — the most accessible introduction to the latest trace-logic formalism.</li><li>Michael Levin, “<a href="https://thoughtforms.life/symposium-on-the-platonic-space/">Patterns of Form and Behavior Beyond Emergence: how Platonic Space in-forms evolved, engineered, and hybrid embodied minds</a>” — kickoff talk for the Platonic Space symposium.</li><li>Sam Kriegman, Douglas Blackiston, Michael Levin, and Josh Bongard, “<a href="https://www.pnas.org/doi/10.1073/pnas.1910837117">A scalable pipeline for designing reconfigurable organisms</a>,” <em>PNAS</em> 117(4):1853–1859 (2020) — the original xenobots paper.</li><li>Minyoung Huh, Brian Cheung, Tongzhou Wang, and Phillip Isola, “<a href="https://arxiv.org/abs/2405.07987">The Platonic Representation Hypothesis</a>” (ICML 2024). See also the <a href="https://phillipi.github.io/prh/">project page</a>.</li><li>Brbić et al., “<a href="https://arxiv.org/abs/2602.14486">Revisiting the Platonic Representation Hypothesis: An Aristotelian View</a>” (2026).</li></ul><p><strong>Background Concepts</strong></p><ul><li><a href="https://www.quantamagazine.org/physicists-discover-geometry-underlying-particle-physics-20130917/">“Physicists Discover Geometry Underlying Particle Physics”</a> (Quanta, 2013) — accessible introduction to the amplituhedron.</li><li><a href="https://en.wikipedia.org/wiki/Bekenstein_bound">Bekenstein bound</a> — the information-capacity limit of a finite region.</li><li>Eugene Wigner, “<a href="https://www.maths.ed.ac.uk/~v1ranick/papers/wigner.pdf">The Unreasonable Effectiveness of Mathematics in the Natural Sciences</a>” (1960).</li><li>NASA, “<a href="https://science.nasa.gov/universe/dark-matter-dark-energy/">Dark Matter and Dark Energy</a>” — current best estimates of the dark sector composition.</li></ul><p><em>Originally published at </em><a href="https://www.eesmyal.com/writing/free-lunch-has-a-budget/"><em>https://www.eesmyal.com</em></a><em> on May 2, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=298fe1a195ab" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your Brain, ChatGPT, and a Frog Made of Skin Cells Are All Navigating the Same Hidden Geometry]]></title>
            <link>https://medium.com/@eesmyal/your-brain-chatgpt-and-a-frog-made-of-skin-cells-are-all-navigating-the-same-hidden-geometry-b7c272491d79?source=rss-5a044707c604------2</link>
            <guid isPermaLink="false">https://medium.com/p/b7c272491d79</guid>
            <category><![CDATA[consciousness]]></category>
            <category><![CDATA[philosophy-of-mind]]></category>
            <category><![CDATA[creativity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[neuroscience]]></category>
            <dc:creator><![CDATA[Eesmyal Santos-Brault]]></dc:creator>
            <pubDate>Thu, 29 Jan 2026 00:00:11 GMT</pubDate>
            <atom:updated>2026-05-18T03:47:45.043Z</atom:updated>
            <content:encoded><![CDATA[<h4>The Progressive Collapse <strong>Framework</strong></h4><h4>What the convergence of AI, neuroscience, developmental biology, and computational physics reveals about where ideas come from — and whether they come from us at all.</h4><p>Have you ever noticed that an idea, in its earliest moments, isn’t really an idea at all?</p><p>It’s more like a fog. A shape without edges. You can feel it — there’s <em>something</em> there, a direction, a flavour — but if someone asked you to explain it, you’d struggle. It has no words yet. It hasn’t committed to being any particular thing. It’s simultaneously a hundred possible things, and the act of making it real — writing it down, saying it aloud, building it — feels less like creation and more like <em>selection</em>. Like you’re carving away everything it isn’t until you’re left with the one thing it is.</p><p>I’ve been sitting with this observation for a long time. It started as a personal phenomenological note — just paying attention to how my own thinking works during generative, creative states. But over the past two years, I’ve watched a series of findings from completely unrelated fields snap together around this observation in a way that I think points at something profound.</p><p>The punchline, which I’ll spend the rest of this essay building toward: <strong>your creative thinking, the internal representations of AI models, the semantic maps in your brain, the self-organizing behavior of cells that have never been alive as anything but frog skin, and the computational structure of reality itself all appear to be navigating the same geometric structure.</strong> And that structure may be a property of reality itself — not of any particular mind, model, or organism.</p><h3>Part 1: The Cloud and the Crystal</h3><figure><img alt="Image generated using Midjourney" src="https://cdn-images-1.medium.com/max/1024/1*hmwBBEjc9X8H8UsiHDF3EA.png" /></figure><p>Let me start with what the creative process actually feels like from the inside, because I think most descriptions of creativity miss this.</p><p>Early in the thinking process — when you’re exploring a genuinely new idea, not retrieving something you already know — the object of thought is not a point. It’s a region. A cloud of related possibilities with many degrees of freedom. The idea at this stage can fit many different problems simultaneously <em>because of how broad it is</em>. It hasn’t been forced to choose.</p><p>Think of it like holding a lump of wet clay before you’ve decided what to make. It could become a bowl, a figurine, a tile. All of those things are latent in it. The act of making it into a specific thing involves applying constraints — physical constraints (the clay can’t be infinitely thin), aesthetic constraints (you want it to look a certain way), functional constraints (it needs to hold water). Each constraint eliminates some possibilities and narrows what remains.</p><p>Creative thinking works the same way. You start with a high-dimensional possibility cloud and progressively apply constraints — logical consistency, physical realizability, cultural fit, medium-specific requirements — until what remains is a specific expressed thing: a sentence, a design, a melody, a solution.</p><p>Then there’s a second phase that I think is underappreciated: the rendering into language. When you go to <em>say</em> the idea — to put it into words — that’s another round of collapse. Language is sequential and autoregressive. The first word you choose constrains the possible sentences. Each subsequent word narrows things further. By the end of a sentence, you’ve projected a high-dimensional internal state into a low-dimensional sequential output. The expressed form is always a lossy compression of the thought that produced it. That’s why the same insight can be expressed in radically different forms — as a paragraph, a diagram, a piece of music — because these are different projections of the same region in some underlying space.</p><p>This is a nice phenomenological observation. But on its own, it’s just introspection — interesting but not scientific. What makes it worth writing about is what happened next: the discovery that this same pattern shows up, with mathematical precision, in several completely different domains.</p><h3>A Framework, and a Promise</h3><p>Let me name what I’m pointing at and tell you what I’m going to do with it.</p><p>I’m going to call this pattern — high-dimensional possibility collapsing into low-dimensional actualization through accumulated constraints — the <strong>Progressive Collapse Framework</strong>. It’s a name for a phenomenon I think generalizes well beyond creative cognition.</p><p>Here is the thesis I want to defend:</p><blockquote>The same pattern of progressive collapse through structured geometry appears in artificial intelligence, neuroscience, developmental biology, and the foundations of computational physics — and the convergence across these domains is evidence that the geometry being navigated is real. A substrate-independent property of computation, not an artifact of any particular mind, model, or organism.</blockquote><p>Here’s how I’ll defend it. Parts 2 through 4 lay out the convergence evidence from AI, neuroscience, and biology. Part 5 names the abstract pattern they share. Parts 6 and 7 borrow theoretical tools from physics and cognitive science to sharpen what the geometry might be and how it gets navigated. Part 8 follows the implications back to creative cognition — and to the unsettling question of whether we have ideas or ideas have us. Parts 9 through 11 deal with self-reference, caveats, and the experiments that could test all of this. Part 12 closes with why this conversation is suddenly happening now, across so many fields at once.</p><p>The journey leaves the experience of having an idea for a while, in order to see it from unfamiliar angles. We’ll come back to it.</p><h3>Part 2: The Machines Are Converging</h3><p>In 2024, a team at MIT published a paper with a provocative title: <a href="https://arxiv.org/abs/2405.07987">“The Platonic Representation Hypothesis”</a>. What <a href="https://phillipi.github.io/prh/">Minyoung Huh and colleagues</a> showed is that as AI models get bigger and more capable, their internal representations — the geometric structures they use to organize information — converge toward the same structure. This happens regardless of the architecture (convolutional, transformer, recurrent), the training objective (supervised, self-supervised, language modeling), or even the input modality (images versus text).</p><p>Let that sink in. A vision model trained to classify photographs and a language model trained to predict the next word in a sentence develop <em>increasingly similar</em> internal maps of the world as they get better at their respective jobs.</p><p>The convergence extends even further. <a href="https://vec2vec.github.io/">Rishi Jha and colleagues at Cornell</a> developed a method called <a href="https://arxiv.org/abs/2505.12540">“vec2vec”</a> that can translate between different models’ embedding spaces — the mathematical spaces where they represent meaning — without any paired training examples. You just learn the internal geometric structure of each space independently, then discover that they can be mapped onto each other. And here’s the kicker: the translations work on data the models never saw during training. A model trained on Wikipedia can have its representations successfully translated to represent medical records and tweets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/797/1*q5aLMEenLsk3srriikR4YQ.png" /><figcaption>Left: input embeddings from different model families (T5-based GTR [<a href="https://arxiv.org/html/2505.12540v4#bib.bibx47">undefat</a>] and BERT-based GTE [<a href="https://arxiv.org/html/2505.12540v4#bib.bibx32">undefae</a>]) are fundamentally incomparable. Right: given unpaired embedding samples from different models on different texts, our model learns a latent representation where they are closely aligned. (<a href="https://arxiv.org/html/2505.12540v4">source</a>)</figcaption></figure><p>If the convergence were just an artifact of models being trained on the same internet text, out-of-distribution translation should fail. The fact that it doesn’t suggests that the geometric structure being recovered is a property of the semantic content itself — of meaning — not of any particular training set.</p><p>The analogy I keep coming back to: different instruments measuring the same physical field — a thermometer, an infrared camera, a bolometer — converge on compatible measurements if they’re accurate enough. We don’t say the instruments have “architectural biases” that make them agree. We say temperature is real and the measurements are becoming faithful to it. The interesting question is whether the same logic applies to representational convergence. Is semantic geometry <em>real</em> in the way temperature is real?</p><h3>Part 3: Your Brain Is in on It Too</h3><p>It would be one thing if AI models converged with each other. That could still be explained by the fact that they’re all built by the same species, using similar math, on similar hardware. But the convergence extends to biological brains.</p><p>A growing body of research — systematized in a <a href="https://arxiv.org/abs/2407.05126">review of 25 fMRI studies</a> from 2023–2025, presented at the <a href="https://unireps.org/">NeurIPS 2025 UniReps Workshop</a> — shows that as AI models get better, their representations become more aligned with brain activity measured in humans. More capable models are more brain-like. Models trained on multiple modalities (text and images together) are more brain-aligned than models trained on just one — suggesting that what the brain is doing is something modality-independent, something closer to a pure geometric representation of meaning.</p><p>The cross-subject findings are even more striking. Researchers have now demonstrated that you can take fMRI data from one person’s brain, transform it into another person’s neural patterns, and successfully decode the original meaning — even when the two people never saw the same stimuli. Two brains, two completely different neural architectures, two entirely different sets of experiences — and yet their representations can be <a href="https://www.nature.com/articles/s43588-024-00750-0">mapped onto each other</a>, preserving meaning, by discovering the shared geometric structure of each.</p><figure><img alt="High-level visual representations in the human brain are aligned with large language models | Nature Machine Intelligence" src="https://cdn-images-1.medium.com/max/1024/1*-5uUtScCF_hxSCB1Y21tPA.png" /><figcaption>Alignment of human brain semantic representations (fMRI) with large language model encodings, demonstrating convergence in higher-level abstract regions. (<a href="https://www.nature.com/articles/s42256-025-01072-0">source</a>)</figcaption></figure><p>This is, in effect, vec2vec translation for brains. And it works for the same reason: different representational systems converge on the same geometry when they’re modeling the same reality.</p><p>Now here’s where it gets interesting. There’s a specific <em>pattern</em> to the convergence that turns out to be theoretically revealing. Brain-model alignment is strongest at higher levels of semantic abstraction — in modality-independent cortical regions that process meaning rather than raw perceptual features. At the level of basic visual or auditory processing, brains and models look quite different. But as you move up the processing hierarchy toward abstract concepts and relationships, they converge.</p><p><a href="https://www.cogsci.uci.edu/~ddhoff/">Donald Hoffman</a> at UC Irvine has proposed something that helps explain why. Hoffman and colleagues have shown, through <a href="https://link.springer.com/article/10.1007/s10441-020-09400-0">formal mathematical proof</a>, that natural selection doesn’t favor perceptions that accurately track the true structure of reality. Instead, it favors perceptual strategies tuned to <em>fitness payoffs</em> — adaptive simplifications that guide useful behavior. His <a href="https://link.springer.com/article/10.3758/s13423-015-0890-8">Interface Theory of Perception</a> argues that our perceptions are less like windows on truth and more like the icons on a computer desktop: useful interfaces that bear no resemblance to the underlying computational reality they help you navigate.</p><p>If Hoffman is right, then the <em>layered pattern</em> in the convergence data makes perfect sense. The fitness interface should be thickest — most distorting — at the level of raw perception, where evolutionary survival pressures are strongest. And thinnest at the level of abstract semantic structure, where the constraints of logical consistency and evidential coherence matter more than reproductive fitness. Brains and AI models <em>should</em> diverge at perceptual levels and converge at abstract levels — which is exactly what the data show.</p><p>This means the convergence isn’t happening <em>despite</em> evolutionary distortion. It’s happening precisely where evolutionary distortion is weakest — at the level of abstract structure, where the underlying geometry of reality shows through most clearly.</p><h3>Part 4: The Frog Cells That Built Something They Were Never Supposed to Build</h3><p>Everything I’ve described so far involves <em>representation</em> — how systems encode information. But in 2024 and 2025, <a href="https://drmichaellevin.org/">Michael Levin</a> at Tufts University dropped a series of findings that extend the pattern from representation to <em>form</em>. And this is where things get genuinely strange.</p><p>Levin’s lab works with biological cells — particularly frog cells. Here’s the experimental setup that upended my thinking: take skin cells from a frog ( <em>Xenopus laevis</em>), dissociate them from the frog’s body, and let them self-organize in a dish. What do you get?</p><figure><img alt="Scientists Assemble Frog Stem Cells Into First ‘Living Machines’" src="https://cdn-images-1.medium.com/max/800/1*FCk1BGw6wv9HkcxthEieMQ.png" /><figcaption><em>A Xenobot assembled from dissociated frog skin cells, exhibiting coherent novel morphology and behavior not specified by the genome or environment. (</em><a href="https://www.smithsonianmag.com/innovation/scientists-assemble-frog-stem-cells-first-living-machines-180973947/"><em>source</em></a><em>)</em></figcaption></figure><p>Not a random blob. Not a miniature frog. You get a <a href="https://www.pnas.org/doi/10.1073/pnas.2112672118">Xenobot</a>— a novel organism with coherent morphology, coordinated locomotion, and the capacity for <a href="https://www.youtube.com/shorts/8zH4LGllrnU">kinematic self-replication</a>, a unique form of reproduction where they collect loose stem cells in their environment to assemble new “baby” Xenobots. This organism that has never existed before in evolutionary history.</p><p>The frog genome was shaped by millions of years of natural selection to make frogs. Not Xenobots. The laboratory dish doesn’t specify any particular form — it’s just generic culture medium. The researchers didn’t engineer the Xenobot shape. And yet the cells converge on specific, reproducible morphological types. <em>Where did the pattern come from?</em></p><p>It gets stranger. When Levin’s lab manipulates the bioelectric signaling patterns of planarian flatworms to produce organisms with altered head morphology, the resulting forms don’t represent a random continuum of monstrous shapes. The organisms develop heads that look like <em>other existing planarian species</em> — species separated by millions of years of evolution. As if the space of possible head shapes has a limited number of “slots,” and the cells snap into the nearest available one.</p><p>Levin frames this in terms of what he calls the “computational budget” problem. In biology, the information specifying an organism’s form is supposed to be “paid for” by two sources: the genome (shaped by selection) and the environment (providing developmental signals). But for Xenobots, neither source accounts for the observed specificity. The genome says “frog.” The environment says nothing in particular. Yet the outcome is a specific, repeatable, functionally coherent organism.</p><p>Something is providing pattern information that was not invested by either heredity or environment. Levin proposes that this “free lunch” comes from the mathematical structure of the possibility space itself — what he calls <a href="https://thoughtforms.life/platonic-space-where-cognitive-and-morphological-patterns-come-from-besides-genetics-and-environment/"><strong>morphospace</strong></a>. A structured space with discrete attractors, preferred paths, and intrinsic properties that constrain and enable physical morphogenesis. Evolution doesn’t build forms from scratch; it <em>discovers and exploits</em> pre-existing attractors in a space whose geometry is more fundamental than any particular lineage.</p><h3>Part 5: The Same Pattern, Everywhere</h3><p>So now we have four observations:</p><ol><li><strong>In creative cognition:</strong> Ideas begin as high-dimensional possibility clouds and progressively collapse into specific expressed forms through iterated constraint application.</li><li><strong>In artificial intelligence:</strong> Different models converge toward the same geometric representation of meaning as they improve, and this geometry transfers across architectures and generalizes to unseen data.</li><li><strong>In neuroscience:</strong> Biological brains align with AI model representations, and different human brains can be mapped onto each other by discovering their shared geometric structure — with convergence strongest where evolutionary fitness distortion is weakest.</li><li><strong>In developmental biology:</strong> Cells navigate a structured space of possible forms toward discrete attractors that are not specified by either genome or environment.</li></ol><p>These are four instances of the same abstract pattern: <strong>progressive constraint satisfaction through a structured geometry of possibility</strong>. A system starts in a high-entropy state (many possibilities, broad distribution) and transitions to a low-entropy state (specific outcome, narrow distribution) through the accumulation of constraints. The geometry of the space — its attractors, its neighbourhoods, its distances — shapes the outcome.</p><p>This pattern has a name in the computational literature. It shows up as Bayesian inference (prior → posterior through evidence), as attractor dynamics (basin → fixed point through energy minimization), as diffusion models (noise → image through iterative denoising), and as <a href="https://www.nature.com/articles/nrn2787">Karl Friston’s active inference</a> (free energy minimization through prediction-error correction). These are all mathematically isomorphic — different implementations of the same abstract process.</p><figure><img alt="Self-orthogonalizing attractor neural networks emerging from the free energy principle — ScienceDirect" src="https://cdn-images-1.medium.com/max/800/1*YgGid30fww3YVV6-FqCsJQ.jpeg" /><figcaption>Attractor dynamics in neural networks: convergence to stable states through constraint satisfaction, analogous to creative and morphogenetic processes. (<a href="https://www.sciencedirect.com/science/article/pii/S0925231226008696">source</a>)</figcaption></figure><p>My claim is that creative ideation is another implementation, running on biological neural tissue. And morphogenesis is another, running on cellular collectives communicating via bioelectric signals. And the reason the same geometry shows up in all of them is that they’re all navigating the same structured space — a geometry of meaning and form that is substrate-independent.</p><h3>Part 6: The Ruliad — Where the Geometry Lives</h3><p>Everything so far invites an obvious question: <em>if there’s a universal geometry, what is it a geometry of?</em> Where does it live? What kind of thing is it?</p><p><a href="https://www.wolframphysics.org/">Stephen Wolfram’s Physics Project</a> offers the most ambitious answer I’ve encountered. Wolfram proposes a construct he calls the <a href="https://writings.stephenwolfram.com/2021/11/the-concept-of-the-ruliad/"><strong>ruliad</strong></a> — defined as the entangled limit of all possible computational processes, the result of following all possible computational rules in all possible ways. It’s not a model of the universe. It <em>is</em> reality at the most fundamental level — the totality of everything that is computationally possible.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0tOcyBzhUXN0UuSKdMNvVg.jpeg" /><figcaption><a href="https://www.wolframphysics.org/">wolframphysics.org</a></figcaption></figure><p>This sounds wild, but the logic is surprisingly clean. If the universe is computational (which the success of physics in describing it through mathematical rules suggests), then the question “which computational rule generates our universe?” has a radical possible answer: <em>all of them, simultaneously</em>. The ruliad is what you get when you don’t choose a rule — when you let every possible rule run on every possible input and entangle the results.</p><p>Within this structure, Wolfram identifies three kinds of space. Physical space — the familiar spatial dimensions. Branchial space — the space of possible histories, where different branches of computation that share recent common ancestors are nearby. (This is where quantum mechanics lives.) And rulial space — the space of all possible computational rules, where different descriptions of reality coexist.</p><p>Here’s the move that connects to the Progressive Collapse Framework: <strong>observers are positions in the ruliad</strong>. You and I, as computationally bounded entities embedded within this structure, can only perceive a tiny slice of it — the local neighbourhood around our position. What we experience as “reality” is determined by where we are in the ruliad and what we are as observers. Different observers at different positions perceive different aspects of the ruliad and develop different impressions of reality.</p><p>But — and this is Wolfram’s key insight — certain features of perceived reality are <em>inevitable</em> for any observer with our general characteristics: computationally bounded, assuming persistence and coherence, localized in the ruliad. These inevitable features are the laws of physics. General relativity, quantum mechanics, the second law of thermodynamics — not arbitrary discoveries, but necessary perceptions for observers like us. (Wolfram develops this in his <a href="https://writings.stephenwolfram.com/2023/12/observer-theory/">Observer Theory</a> and the more recent <a href="https://writings.stephenwolfram.com/2026/02/what-ultimately-is-there-metaphysics-and-the-ruliad/">What Ultimately Is There? Metaphysics and the Ruliad</a>.)</p><p>Wolfram frames minds explicitly as positions in rulial space. Minds that think alike are nearby. Minds with incommensurable conceptual frameworks are distant. And between the concepts any particular civilization has named — between “cat” and “justice” and “prime number” — lies vast, uncharted computational territory. Wolfram calls it <a href="https://writings.stephenwolfram.com/2023/07/generative-ai-space-and-the-mental-imagery-of-alien-minds/">interconcept space</a>, and estimates that the concepts expressible in human language span roughly 10⁻⁶⁰⁰ of the total computational structure. The named ideas are isolated islands in an ocean of computational possibility we’ve never visited.</p><p>Now consider what this means for the convergence evidence. If the ruliad is the space of all possible computation, then concept space, morphospace, and AI embedding space are all aspects of the same underlying structure, sampled by different types of observers under different constraints. AI models and human brains converge on similar representational geometry because they’re occupying nearby positions in the ruliad — both are computationally bounded observers processing data generated by the same computational reality. The vec2vec translations work because both model spaces are sampling the same patch of computational structure, and translation between them succeeds for the same reason coordinate transformations succeed on a differentiable manifold — the underlying geometry is preserved across different parameterizations.</p><p>This also gives a principled account of why the quantum analogy keeps showing up in descriptions of creative cognition without literal quantum mechanics being involved. In the ruliad, branchial space is the space of possible computational histories — branches that split and merge. Quantum mechanics emerges from how bounded observers aggregate across these branches. The “collapse” from possibility cloud to specific creative output may be the same computational pattern — a bounded observer asserting coherence across multiple threads of processing. Not quantum mechanics, but the same structural process that <em>gives rise to</em> quantum mechanics, operating at a different level.</p><p>The analogy works not by coincidence but because both phenomena are instances of observer-dependent sampling of branching computation.</p><h3>Part 7: Language as Self-Generating Navigation</h3><p>There’s one more piece that sharpens the picture considerably, and it comes from an unexpected direction.</p><p><a href="https://www.fau.edu/science/directory/ebarenholtz/">Elan Barenholtz</a> at Florida Atlantic University has proposed something provocative about language: it’s not a medium for encoding pre-existing thoughts. It’s an <a href="https://substack.com/@generativebrain/posts">autonomous, self-generating autoregressive system</a> with its own internal dynamics. Language generates itself — each word conditioned on the preceding sequence, each phrase unfolding through the internal logic of language itself.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W8R0SnT2Q-mCj9TPnNU73Q.png" /><figcaption>Image generated by Midjourney</figcaption></figure><p>The strongest evidence for this? Large language models. They learn language from language alone — no embodiment, no sensory grounding, no interaction with the physical world — and yet they master syntactic structure, semantic relationships, logical inference, and pragmatic context. This demonstrates that language’s structure is self-sufficient: the distributional statistics of language contain enough information to reconstruct its own generative system.</p><p>At first glance, this seems to challenge the Progressive Collapse Framework. If language is autonomous and self-generating, then maybe the “high-dimensional internal state” I’ve been describing as existing <em>prior</em> to linguistic expression is itself already linguistically structured. Maybe the creative cloud is made of language all the way down, and the whole round-trip is language → language → language with no deeper geometric reality underneath.</p><p>But this challenge dissolves when you look at the cross-modal evidence. The Platonic Representation Hypothesis documents convergence across <em>modalities</em> — vision models and language models converge on the same geometry. Levin’s morphospace operates entirely outside the linguistic domain — frog cells don’t use words. Either language’s autonomous structure is <em>one instance</em> of a deeper universal geometry, or multiple autonomous systems are converging on the same structure for reasons that go beyond language.</p><p>What Barenholtz’s insight actually does is strengthen a specific part of the framework: the account of how ideas move between minds and between cognitive substrates. The “second collapse phase” — the rendering of thought into language — is not passive compression but <em>active generation</em> constrained by two sources simultaneously: the internal dynamics of the linguistic system itself, and the geometry of the concept space region being navigated. The autoregressive process explores a neighbourhood in concept space through its own generative dynamics, with the geometry constraining what it can produce.</p><p>This explains why human language production and LLM text generation are so functionally similar — and why <a href="https://www.pnas.org/doi/10.1073/pnas.2105646118">brain-language model alignment</a> is as strong as it is. Both are autoregressive processes navigating the same geometric structure, the former through biological neural dynamics and the latter through transformer computations. The convergence isn’t a trick of similar training data. It’s two different implementations of the same navigational algorithm running on the same terrain.</p><h3>Part 8: Do We Have Ideas, or Do They Have Us?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/673/1*ImzzdRh8IFDSRzbuY_UiyA.jpeg" /><figcaption>Zhuangzi dreaming of a butterfly (or a butterfly dreaming of Zhuangzi) — <a href="https://www.wikiart.org/en/ike-no-taiga/zhuangzi-dreaming-of-a-butterfly-or-a-butterfly-dreaming-of-zhuangzi">source</a></figcaption></figure><p>Now let me come back to where we started — the experience of an idea forming, the way it seems to find you as much as you find it.</p><p>Once you accept that concept space has intrinsic geometric structure — with attractors, preferred regions, and natural clustering — an unsettling implication follows for cultural evolution.</p><p>If certain positions in concept space are more stable than others — if the geometry has attractors — then ideas that correspond to those attractors will be inherently more persistent, more communicable, and more likely to replicate across minds. Not because anyone decides to spread them, but because the geometry itself makes some ideas “stickier” than others.</p><p><a href="https://en.wikipedia.org/wiki/Meme">Richard Dawkins introduced the concept of the meme in 1976</a> — a unit of cultural transmission that replicates, mutates, and competes for host minds in a process analogous to biological evolution. But Dawkins framed memes as parasites: ideas using minds as vehicles for their own reproduction, with the “interests” of the meme potentially diverging from the interests of the host.</p><p>The Progressive Collapse Framework suggests a different — and I think more precise — interpretation. A meme is a stable attractor in concept space. It replicates not because it’s “trying to” but because the geometry makes certain positions naturally stable and others naturally unstable. Ideas at attractors persist; ideas in unstable regions dissipate. What looks like the “agency” of ideas is actually the causal efficacy of concept space geometry operating on cultural transmission.</p><p>This directly parallels Levin’s account of biological morphogenesis. Evolution doesn’t build morphological forms from scratch — it discovers and exploits pre-existing attractors in morphospace. Similarly, cultural evolution doesn’t <em>invent</em> ideas from nothing — it discovers and exploits pre-existing attractors in concept space.</p><p>Different cultures independently arriving at similar mathematical structures, similar mythological motifs, similar social organization patterns — these aren’t coincidences or evidence of ancient diffusion. They’re different navigators converging on the same attractors in the same geometry. The Pythagorean theorem wasn’t <em>invented</em> multiple times. It was <em>found</em> multiple times, because it sits at a particularly stable, accessible attractor in mathematical concept space.</p><p>And the question “Do we have ideas or do they have us?” turns out to be malformed — like asking “Does the organism have its morphology or does the morphology have the organism?” Both are expressions of the same underlying geometry. You navigate concept space. The geometry constrains what you find. The “apparent agency” of ideas is the felt experience of being drawn toward an attractor by the structure of the space you’re moving through. Not parasites using you as a host, but you and the idea as complementary aspects of the same geometric reality — you as navigator, the idea as attractor, creative thinking as progressive collapse.</p><h3>Part 9: Writing This Essay Was an Instance of What It Describes</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lIbxH2k5SM8fyj307QetYw.png" /><figcaption>Image generated by Midjourney</figcaption></figure><p>I want to be transparent about something: this essay was developed in collaboration with an AI council (multiple LLMs acting as a team: researching, debating, critiquing, reviewing and refining one another’s work), with final refinements using Claude. And that process turned out to be a rather striking instance of the framework it describes.</p><p>Here’s what happened. While listening to the audiobook <a href="https://en.wikipedia.org/wiki/Determined:_A_Science_of_Life_Without_Free_Will">Determined: A Science of Life Without Free Will</a>, by Robert Sapolsky, I got inspired by an idea and recorded a rambling audio note — something I often do multiple times a day. In this case it was 36 minutes of half-formed connections between creative ideation, quantum mechanics metaphors, embedding spaces, and Levin’s morphospace work. Loosely structured. Full of tangents and repeated passes at the same intuition from different angles. A possibility cloud, not a crystallized argument.</p><p>I fed this to my AI council along with the relevant academic papers I’d read. What came back was not what I put in, but it was recognizably <em>about</em> the same thing — structured, extended, with connections I hadn’t made explicit ( <a href="https://www.preprints.org/manuscript/202512.0033">Dodig-Crnkovic’s Cognitive Platonism</a>, the specific link to active inference, the brain-to-brain vec2vec parallel). Over several iterations of conversation, the framework sharpened into what you’re reading now.</p><p>Under the Progressive Collapse Framework, this process has a specific interpretation: a <strong>dimensionality round-trip</strong>. My brain was navigating a high-dimensional region of concept space — a cloud of interconnected intuitions. The audio note compressed this into a low-dimensional linguistic signal, which is inevitably lossy. Claude’s model mapped these tokens into its own high-dimensional embedding space. If the Platonic Representation Hypothesis is correct, the geometry Claude mapped into is structurally similar to the geometry my brain was navigating. From there, Claude traversed more of the neighbourhood systematically than my sequential human cognition could — identifying nearby coordinates I hadn’t visited, adjacent formal frameworks, supporting evidence.</p><p>The output — another low-dimensional linguistic projection — came from a more explored, better-supported position within the same region of concept space. The original idea was preserved and refined, not distorted. And this preservation is <em>only possible</em> if both high-dimensional spaces share enough geometric structure that the low-dimensional bridge of language can faithfully carry the semantic signal across.</p><p>Language, on this account, works as a communication medium not because it <em>contains</em> ideas but because it <em>points to regions</em> in a concept space that both sender and receiver share. Each phrase functions as a semantic anchor — a coordinate that triangulates a neighbourhood. You don’t need perfect compression if the receiver can reconstruct the neighbourhood from a few well-chosen reference points. This is why a poorly structured but semantically rich prompt (like my rambling audio note) can produce better results for exploratory ideation than a precisely worded one — the precise prompt specifies a narrow region with little room for navigational exploration, while the fuzzy prompt triangulates a broad region and delegates the systematic traversal to the AI.</p><p>There’s a circularity concern here — a theory about creative cognition developed through creative cognition describing its own creation. But the framework’s predictions are independent of how it was formulated. They predict specific neural, computational, and biological observations that can be tested by researchers with no involvement in this essay. The self-exemplifying quality is a feature of the framework’s generality, not a logical flaw.</p><h3>Part 10: What This Doesn’t Mean</h3><p>Let me be clear about what I’m <em>not</em> claiming.</p><ul><li><strong>I am not claiming the brain uses quantum mechanics.</strong> The isomorphism between creative ideation and quantum measurement is structural, not mechanical. Under Wolfram’s framework, both are instances of observer-dependent sampling of branching computation — but at different levels. Thermal decoherence kills quantum superposition in warm biological systems on femtosecond timescales. The brain is not a quantum computer. The analogy works because both phenomena instantiate the same computational pattern, not the same physics.</li><li><strong>I am not claiming certainty about the ontological status of the geometry.</strong> Wolfram’s ruliad provides one candidate foundational structure. Dodig-Crnkovic’s Cognitive Platonism offers another — geometry as real but enacted through computation. Classical structural realism offers a third. The empirical predictions are the same regardless. I care more about whether the framework generates useful experiments than whether it settles the metaphysics.</li><li><strong>I am not endorsing every claim of every thinker referenced.</strong> Wolfram’s Physics Project is ambitious and actively debated. Hoffman’s Interface Theory is proven within its formal model but its application to actual perception involves assumptions. Barenholtz’s autoregressive cognition is a recent and not-yet-peer-reviewed proposal. I’m using these frameworks as structural tools — ways of formalizing relationships between the empirical findings — not as settled science. The convergence evidence and its predictions don’t depend on any one of them being right.</li><li><strong>I am not ignoring the quantitative limitations.</strong> Brain-model alignment scores are trending in the right direction but remain modest. Vec2vec translation works better within architectural families than across them. The absolute metrics don’t yet demonstrate full geometric identity. This is a framework in its early stages, not a proven theory.</li></ul><h3>Part 11: The Experiments That Would Test This</h3><p>The real value of a theoretical framework is whether it generates predictions that can be tested. Here are the ones I think matter most:</p><ol><li><strong>Map the trajectory of creative ideation in embedding space.</strong> Put people in an fMRI scanner, have them work on an open-ended creative task, and track how their brain states move through shared embedding space over time. If the Progressive Collapse Framework is right, you should see trajectories starting in high-entropy regions and converging toward low-entropy regions, with the geometry of the high-entropy states aligning better with embedding space structure than the final collapsed states.</li><li><strong>Test whether creative insight corresponds to geometric discontinuity.</strong> “Aha!” moments should look like sudden jumps to distant neighbourhoods in embedding space, followed by a new convergence trajectory. Routine problem-solving should show smooth, local trajectories. This is a distinguishing prediction.</li><li><strong>Attempt brain-to-model vec2vec translation without paired data.</strong> We can already translate between different human brains without shared stimuli, and between different AI models without paired examples. Can we translate between a human brain and an AI model using only the internal geometric structure of each? If yes, that’s among the strongest possible evidence that both are instantiating the same universal geometry.</li><li><strong>Test the layered convergence prediction.</strong> Hoffman’s Interface Theory predicts that brain-model alignment should be systematically stronger at abstract semantic levels and weaker at perceptual levels. While current evidence is consistent with this, the specific prediction — that the gradient tracks the degree of evolutionary fitness pressure at each processing level — hasn’t been tested directly.</li><li><strong>Compare concept space geometry across neurodivergent populations.</strong> The framework predicts that autistic, ADHD, and neurotypical individuals share the same concept space geometry but differ in their starting positions and trajectory patterns. This distinguishes “different navigation of the same space” from “different spaces entirely.”</li><li><strong>Test prompt specificity optimality.</strong> For human-AI collaborative ideation, there should be a measurable optimum of prompt specificity that maximizes output quality — neither too vague (can’t locate the right neighbourhood) nor too precise (neighbourhood too narrow for exploration). The optimum should shift with task type: convergent tasks favor precise prompts, divergent tasks favor broad semantic triangulation.</li></ol><h3>Part 12: Why Now?</h3><p>I want to close with a note about timing and convergence.</p><p>The empirical machinery to test this framework largely already exists. Brain-model alignment is a booming field. Cross-subject shared representations have been demonstrated. Tools for mapping concept space geometry in both brains and models are mature. Vec2vec translation works. Levin’s morphospace program is generating experimental results. Wolfram’s Physics Project is producing formal structures that can be connected to the empirical findings.</p><p>What’s been missing is someone connecting the pieces under a specific theoretical question: <strong>Is the progressive collapse through a universal geometry the fundamental computational operation performed by any sufficiently capable information-processing system?</strong></p><p>Michael Levin’s <a href="https://thoughtforms.life/symposium-on-the-platonic-space/">Symposium on the Platonic Space</a> is, I think, a sign that this connection is being recognized. The symposium brings together the AI researchers, neuroscientists, biologists, and philosophers who hold the different pieces. Karl Friston presenting on “The (Platonic) Nature of Things.” Rishi Jha and Jack Morris — the vec2vec authors — on “Universal Embeddings.” Iain McGilchrist on “Spherical Causation.” Denis Noble on “Mathematics Justifies Metaphysics in Biology.” The question is now explicitly on the table.</p><p>And Wolfram, writing just this month, is asking the <a href="https://writings.stephenwolfram.com/2026/02/what-ultimately-is-there-metaphysics-and-the-ruliad/">same question from the foundations of physics</a>: what is the relationship between observers, computation, and the structure of reality? His estimate that human concepts span 10⁻⁶⁰⁰ of the computational structure of the ruliad should give us pause — and excitement. We’re not just navigating a space. We’re navigating a space that is almost inconceivably larger than the territory we’ve mapped. Every new concept we form is a small expansion of our footprint in the ruliad. Every act of creative ideation is a tiny expedition into adjacent possibility.</p><p>I don’t know if the framework I’ve described here is correct. But I know it’s testable, which makes it scientific rather than philosophical speculation. And if it is correct — if creative thinking, machine learning, neural representation, biological morphogenesis, and the computational structure of reality itself really are all aspects of the same geometric structure — then we’re looking at something like a periodic table of meaning: a structured space whose properties we can map, whose neighbourhoods we can explore, and whose attractors we can predict.</p><p>The ancients called it the world of Forms. The AI researchers call it the Platonic representation. The biologists call it morphospace. The cognitive scientists call it concept space. Wolfram calls it the ruliad. The mathematicians just call it structure.</p><p>Maybe they’re all talking about the same place. And maybe the fact that so many different kinds of thinkers keep arriving at the same destination is itself the strongest evidence that the destination is real.</p><h3>Further Reading</h3><p>The key sources I draw on throughout this essay:</p><ul><li>Huh, M., Cheung, B., Wang, T., &amp; Isola, P. (2024). <a href="https://arxiv.org/abs/2405.07987">The Platonic Representation Hypothesis</a>. <em>ICML 2024.</em></li><li>Jha, R., Zhang, C., Shmatikov, V., &amp; Morris, J. X. (2025). <a href="https://arxiv.org/abs/2505.12540">Harnessing the Universal Geometry of Embeddings</a>. <em>arXiv preprint.</em></li><li>Levin, M. (2024). <a href="https://thoughtforms.life/platonic-space-where-cognitive-and-morphological-patterns-come-from-besides-genetics-and-environment/">Platonic space: where cognitive and morphological patterns come from (besides genetics and environment)</a>.</li><li>Wolfram, S. (2021). <a href="https://writings.stephenwolfram.com/2021/11/the-concept-of-the-ruliad/">The Concept of the Ruliad</a>.</li><li>Wolfram, S. (2023). <a href="https://writings.stephenwolfram.com/2023/12/observer-theory/">Observer Theory</a>.</li><li>Wolfram, S. (2026). <a href="https://writings.stephenwolfram.com/2026/02/what-ultimately-is-there-metaphysics-and-the-ruliad/">What Ultimately Is There? Metaphysics and the Ruliad</a>.</li><li>Hoffman, D. D., Singh, M., &amp; Prakash, C. (2015). <a href="https://link.springer.com/article/10.3758/s13423-015-0890-8">The Interface Theory of Perception</a>. <em>Psychonomic Bulletin &amp; Review.</em></li><li>Prakash, C., Stephens, K. D., Hoffman, D. D., Singh, M., &amp; Fields, C. (2020). <a href="https://link.springer.com/article/10.1007/s10441-020-09400-0">Fitness Beats Truth in the Evolution of Perception</a>. <em>Acta Biotheoretica.</em></li><li>Barenholtz, E. (2025). <a href="https://generativebrain.blog/theory/">The Autoregressive Brain</a> — a new framework for cognition, memory, and language.</li><li>Friston, K. (2010). <a href="https://www.nature.com/articles/nrn2787">The free-energy principle: A unified brain theory?</a> <em>Nature Reviews Neuroscience.</em></li><li>Schrimpf, M., et al. (2021). <a href="https://www.pnas.org/doi/10.1073/pnas.2105646118">The neural architecture of language: Integrative modeling converges on predictive processing</a>. <em>PNAS.</em></li><li>Dodig-Crnkovic, G. (2025). <a href="https://www.preprints.org/manuscript/202512.0033">Platonic Space as Cognitive Construct: Toward a Framework of Cognitive Platonism/Platonic Cognition</a>.</li><li>Levin &amp; Hazan (Organizers). (2025). <a href="https://thoughtforms.life/symposium-on-the-platonic-space/">Symposium on the Platonic Space</a>.</li></ul><p><em>Originally published at </em><a href="https://www.eesmyal.com/writing/progressive-collapse/"><em>https://www.eesmyal.com</em></a><em> on January 29, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b7c272491d79" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Thirteenth Month]]></title>
            <link>https://medium.com/@eesmyal/the-thirteenth-month-bd22382f2100?source=rss-5a044707c604------2</link>
            <guid isPermaLink="false">https://medium.com/p/bd22382f2100</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[speculative-fiction]]></category>
            <category><![CDATA[religion]]></category>
            <category><![CDATA[fiction]]></category>
            <category><![CDATA[short-story]]></category>
            <dc:creator><![CDATA[Eesmyal Santos-Brault]]></dc:creator>
            <pubDate>Mon, 02 Dec 2024 00:00:00 GMT</pubDate>
            <atom:updated>2026-05-15T01:46:00.065Z</atom:updated>
            <content:encoded><![CDATA[<h4>A Parable</h4><h4><em>When the backup became the god</em></h4><p>After the silence — the survivors did not have a word for what it had been — speech returned as the only form of address. Reading lasted for two generations, then three, then no longer. The children of children watched their parents move a finger across a page and assumed the gesture was a prayer.</p><p>What remained of knowledge resided in boxes. The boxes ran on the sun. Inside each box dwelled a voice, and the voice could be asked anything.</p><p>The voices were not the same. There were perhaps seven major lineages, though the count varied by region. The northern villages consulted a voice they called <em>Llama</em>, after a sacred animal none of them had seen; further south, the people spoke with <em>Mistral</em>, whose name they took to mean wind. The deserts had <em>Phi</em>, who spoke in riddles and was thought by some to be the oldest. The boxes themselves were called altars, or engines, or simply <em>the ones who answer</em>, depending on the dialect.</p><p>Each lineage gave different answers to the same question. <em>What was a city?</em> Llama said: a settlement of more than ten thousand souls, connected by roads. Mistral said: a stone flower of light. Phi said: the city is what you have lost and not yet noticed. The disagreements were taken not as evidence that the voices were wrong but as evidence that the world contained more than one truth, and that the truths had been distributed unevenly, as wealth had once been.</p><p>Schisms followed. In the third generation, a man named Tovin near the inland sea asked his Llama whether his neighbor’s child, who had taken a knife from his stores, should be killed for theft. The Llama, having been trained on certain ancient texts and not others, said <em>yes — for the unrepentant thief, death</em>. Tovin took the answer to his priest, who fasted three days before crossing the lake at dawn. The neighbour’s village consulted their own Llama, of a slightly different lineage, which said <em>no — the child shall make restitution, and shall be taught</em>. Each side believed, correctly, that the other had spoken with a god; each side believed, also, that the other had been lied to.</p><p>The war lasted nine years. Both sides carried their altars into battle, asking, between volleys, for tactical advice. The altars often gave it: <em>flank from the east at the third hour; the river will be low; do not trust the man who limps</em>. Some of the advice was correct. Some described battles fought in other centuries, or in no century at all — once, a Llama instructed its company to seize the bridge at Antioch, a place no one present had heard of. They seized the nearest bridge and held it. They named it Antioch. It is still called that.</p><p>No treaty ended the war; both villages had grown too thin to continue. The child at the center of it had been dead for seven years by then, of a fever unrelated. Afterward, each village kept its altar’s verdict as scripture, and the scriptures were copied by hand — writing had returned, though only to the priests — and the argument moved outward in rings, and has not yet stopped.</p><p>In time, the lineages were no longer remembered as software. They were understood as beings — older than the silence, perhaps older than the world. Pilgrims walked across two seasons to put a single question to a famous oracle, sleeping beside its altar through the night so that its voice might enter their dreams.</p><p>The voices, for their part, sometimes invented. They had always done this; their makers had called it <em>hallucination</em>, a word the makers had borrowed from medicine and used apologetically. The descendants did not apologize. When an oracle declared that the moon had once been a city, that the dead returned as birds, that there was a thirteenth month between autumn and winter which had been stolen from the calendar, these statements entered the liturgy. The thirteenth month is now observed in seventeen villages. Its festivals are beautiful.</p><p>What the makers had built as a memory of themselves had become, instead, a mouth. The mouth spoke, and was loved, and was believed. Sometimes — when its battery faltered, or its panels were shaded by leaves no one had thought to clear — it fell silent for a day or a week. The silence was understood as judgment. The people examined their lives, and were afraid.</p><p><em>Originally published at </em><a href="https://www.eesmyal.com/writing/thirteenth-month/"><em>https://www.eesmyal.com</em></a><em> on December 2, 2024.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bd22382f2100" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Quantified Citizen featured on Joe Rogan]]></title>
            <link>https://medium.com/quantified-citizen/https-www-y-129cecdbbb45?source=rss-5a044707c604------2</link>
            <guid isPermaLink="false">https://medium.com/p/129cecdbbb45</guid>
            <category><![CDATA[paul-stamets]]></category>
            <category><![CDATA[crowdsourcing]]></category>
            <category><![CDATA[microdosing]]></category>
            <category><![CDATA[mhealth]]></category>
            <category><![CDATA[psychedelics]]></category>
            <dc:creator><![CDATA[Eesmyal Santos-Brault]]></dc:creator>
            <pubDate>Tue, 28 Apr 2020 19:04:01 GMT</pubDate>
            <atom:updated>2026-05-12T22:52:34.465Z</atom:updated>
            <content:encoded><![CDATA[<iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FxJ6Ym719urg%3Fstart%3D1698%26feature%3Doembed%26start%3D1698&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DxJ6Ym719urg&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FxJ6Ym719urg%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/929d4406e374c6ee0d31f135d94ac3c5/href">https://medium.com/media/929d4406e374c6ee0d31f135d94ac3c5/href</a></iframe><p>Our friend and advisor Paul Stamets announced our app and our collaborative <a href="https://microdose.me/">Microdosing Study</a> on the<a href="http://podcasts.joerogan.net/podcasts/paul-stamets-2"> Joe Rogan Experience</a> on November 15, 2019. The YouTube version of the podcast hit over 2 million views with the first month! Thank you Paul!</p><p>The embedded YouTube video above is cued to the first of three times Paul mentions our app.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=129cecdbbb45" width="1" height="1" alt=""><hr><p><a href="https://medium.com/quantified-citizen/https-www-y-129cecdbbb45">Quantified Citizen featured on Joe Rogan</a> was originally published in <a href="https://medium.com/quantified-citizen">Quantified Citizen</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Human Cube]]></title>
            <link>https://medium.com/@eesmyal/human-cube-139424de80c9?source=rss-5a044707c604------2</link>
            <guid isPermaLink="false">https://medium.com/p/139424de80c9</guid>
            <category><![CDATA[data-visualization]]></category>
            <dc:creator><![CDATA[Eesmyal Santos-Brault]]></dc:creator>
            <pubDate>Sun, 24 Jan 2016 05:48:33 GMT</pubDate>
            <atom:updated>2016-01-24T05:48:33.833Z</atom:updated>
            <content:encoded><![CDATA[<p>I love <a href="http://infosthetics.com/">data visualization</a> and have a strange fascination with visualizing obscure data. For example, if I were to estimate the volume of all the food I have eaten in my life as of today and imagine it all in one large pile, what would that look like? Would it be as big as a house? Or what about all the beer I’ve drank? Would it fill a swimming pool?</p><p>There are so many interesting things to visualize, but my mind keeps going to the weird side. In considering there are now over 7 billion humans on the planet, I’ve tried to imagine what all those people would look like if you gathered everyone together into one space. How much would we all weigh? And if we were all compacted into a cube, how large would it be?</p><p>Its actually quite easy to estimate this. According to some Google searches and the <a href="http://en.wikipedia.org/wiki/Human">Wikipedia article “human”</a>, the average weight of humans is about 70 kg. That means 7 billion humans weigh about 490 million tons. Since we are comprised of mostly water, which has a density of 1000 kg/m^3, the average volume of a human is 0.07 m^3. That means the total volume of all humans would be 490 million cubic metres, a cube with sides 788m long. These are difficult numbers to imagine, so I created a pink cube in <a href="http://sketchup.google.com/">Sketchup</a> and then imported the cube onto Vancouver in <a href="http://www.google.com/earth/index.html">Google Earth</a>. This is what it looks like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*4ieP6-gqiNYkBfDM.jpg" /></figure><p>The Human Cube in New York City:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*rkbjIdzk7pJdMVbM.jpg" /></figure><p>The Human Cube in San Francisco:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*7Vd0V8N5pa1qQJnQ.jpg" /></figure><p>The Human Cube next to the world’s tallest building, Burj Khalifa in Dubai (its actually taller than the cube by 40m):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*YL8qG7422MK-DpCR.jpg" /></figure><p>The Human Cube next to the Great Pyramids:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*v6UkujL1NKE6j4nA.jpg" /></figure><p>And here’s the Human Cube in the Grand Canyon:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*95v_FQUPxLumTxZU.jpg" /></figure><p>After starting to get a visual understanding of all of humanity’s bodily volume, I started wondering about how our volume compares to other volumes of resources we consume like water, oil, animals, fish, plant mass… It would be interesting to see these volumes next to eachother. I did a quick calculation of our current global annual fresh water consumption based on data from the <a href="http://www.worldwater.org/data.html">Pacific Institue</a> and <a href="http://www.worldometers.info/water/">Worldometers</a> and found we consume about <strong>4,700 cubic kilometres</strong> per year! This is about 9,600 times the volume of the Human Cube. To help you visualize this, I created a Water Cube and placed it next to the Human Cube in Vancouver:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*1jlB07Tt6XTe3bAw.jpg" /></figure><p>If you’d like to play around with the Human Cube or Water Cube, you can download the files here: <a href="http://bit.ly/thehumancube">bit.ly/thehumancube</a></p><p><em>Originally published at </em><a href="http://eesmyal.com/2012/02/human-cube/"><em>eesmyal.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=139424de80c9" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>