Duality of oneness

0:00What if I told you that the recipe doesn't actually explain the cake? I mean, that's a pretty wild thought to start with. Right. But I want you to really think about that for a second. We walk around with this deep seated assumption that the algorithms running our computers or, you know, the DNA building or physical bodies act like these exhaustive step-by-step instruction manuals. Yeah. Like an IKEA booklet or something. Yeah. Exactly. But what if that is completely backwards? What if they are just pointers like coordinates tapping into a hidden pre-existing universe of solutions? Well, it's a premise that completely upends how we think about intelligence, biology, and really a whole nature of computation itself. Because if the code we write or the DNA we inherit doesn't actually contain all the answers, but is instead acting as a bridge to a space where those answers already exist. Then everything changes. Right. We're looking at a fundamental shift in how we understand reality. And that is our mission for today's deep dive.

1:00We're going to explore this bleeding edge where biology, computer science, and philosophy just like collide. We're drawing on a really dense, but frankly thrilling piece of writing by essay central titled Revenge of the Dual Aspect Monists. Yeah. And to navigate this, we're really looking at the synthesis of two totally distinct perspectives. First, there's Michael Levin, a biologist who spent decades pushing the boundaries of what cells actually know and how they communicate to build bodies. His work is incredible. It really is. And then we have essay central, a computational philosopher who developed this rigorous mathematical framework called observer theory. Essentially, central is looking at Levin's wild experimental results and saying, Hey, I have the math to explain exactly what you're seeing. It is genuinely mind blowing material. I mean, we're talking about two headed worms, basic algorithms that seem to have a mind of their own and this super unsettling idea that mathematical forms themselves might be evolving. My goal today is to ground all of this for you. We're going to take these seemingly metaphysical ideas and synthesize them

2:03into a very clear, measurable picture of how systems solve problems, which is much needed, honestly. Yeah, we're going to see that this surplus of intelligence, the stuff that isn't written in the code is actually something we can quantify. So to understand how biology is breaking all the rules, central says we actually need to look at basic computer code first, because apparently even the simplest dumbest algorithms are doing things they were never explicitly programmed to do. Right. Central brings up bubble sort, which I honestly haven't thought about since intro to computer science. Yeah, bubble sort is about as elementary as it gets. It just looks at a list of numbers compares adjacent pairs and swaps them if they're in the wrong order. It just bubbles the highest numbers to the top of the list. It's entirely mechanical, no real decision making. Right. But central points out a study from Levin's group where they gave bubble sort of broken cell, basically a number in the array that just refuses to move like a roadblock, simulating an obstacle. Yeah, you'd think the algorithm would just crash, right?

3:03Or get stuck in an infinite loop. But instead it temporarily unsorts itself. The overall sort ofness of the array actually goes down like the list gets messier just so it can route around this obstruction, which is crazy for a basic algorithm. Totally. And then once it's passed the broken cell, it just resumes sorting. The underlying mystery there is what behavioral scientists call delayed gratification, you know, moving away from your goal in the short term to achieve it in the long term. But bubble sort can't do that. Exactly. Bubble sort does not have a cognitive model. There's no line of code that checks whether a swap succeeded. And there's no parameter that understands the concept of an obstruction. It just blindly follows its local rule, yet it exhibits this robust global problem solving behavior. Yeah, it finds a way around. Okay, let's unpack this because central notes to get to even stranger. If you tweak the algorithm, so the numbers themselves act as little agents running their own sorting rules, I think San Charles calls these different rules, algotypes.

4:04Yes, algotypes, right? They still manage to sort the list. But on the way to getting sorted numbers that share the same algotype spontaneously formed clusters. Yeah, they grouped together, hold that formation and then eventually dissolve back into the sorted order. And absolutely nothing in the code rewards this clustering. Nothing tells a number to read its neighbor's rule. Right. Because the algorithm is only instructed to care about numerical values. So where's the spontaneous clustering behavior coming from? That's surplus is what we're tracking today. So weird. And it gets much more pronounced when we look at Levin's work in evolutionary machine learning. They ran this experiment where they took a standard evolutionary algorithm, which normally explores new solutions using a basic Gaussian model, essentially a bell curve. And they replaced that engine with a diffusion model. Oh, I know diffusion models. Yeah, that's what powers modern AI image generators, right? Starting with static noise and slowly refining it into a picture. That's the one. But what happens when you drop that into an evolutionary search?

5:05Three highly anomalous things happen. First, the researchers retrain this diffusion model every generation on a buffer of everything the search had seen so far. Because diffusion models are incredibly good at holding on to latent structure, the population suddenly developed memory memory. We memory. Yeah. If the researchers abruptly changed the target condition of the search, the algorithm didn't have to slowly crawl across the fitness landscape to adapt. It underwent a phase transition and jumped discontinuously. Right back to a peak, it had visited 40 generations ago. Wow. Because the diffusion model remember the overall shape of the landscape. Exactly. And because it has this memory, it suddenly gains a second ability, which they call steering central mentions, classifier free guidance. Can you explain how that works in this context? Sure. So think about prompting an AI image generator. You type in make it cyberpunk and it guides the diffusion process toward that style without changing the core mechanics of the model. Okay. That makes sense. In this evolutionary search, researchers found they can basically prompt the

6:08optimizer. They could tell it to focus its search on a specific quadrant of the landscape and the algorithm just goes there just like that. Just like that. Yeah. It does this completely independently of the fitness function. It's a way of talking directly to your optimizer, guiding it without having to rewrite the whole reward system, which leads to the third anomaly. And this one really threw me. I got to be honest. The twisted, raptured in function. I mean, I get that a standard fitness landscape is like a grid of mountains and algorithms look for the highest peak. But what makes the twisted version so special? Well, standard landscape has peaks that align nicely with the local grid. Traditional algorithms do great there because they just look at the immediate upward slope and climb. Sure, just go up. But if you mathematically warp that landscape, twisting it into a spiral, those peaks no longer align with simple local gradients, the immediate slopes are deceptive. A standard Gaussian search degrades rapidly because it gets tricked by the local twists. It basically gets lost. But the diffusion model doesn't.

7:09It navigates the spiral perfectly. Really? Yeah. Because it has built a broad prior and understanding of the global structure of the space, it isn't fooled by the local deception. It knows where the peaks should be based on the broader pattern. It's like, imagine you have a simple clockwork toy. You wind it up and its code is just walk forward. But then you place a wall in front of it instead of mindlessly bumping into the wall until the spring runs out, the toy somehow sidesteps, walks around the wall and gets back on its path. That's a great analogy. And when you open it up, there are no sensors, no cameras, no complex logic, gates for navigation, just a spring and some gears. Yeah. How is it doing that? Like, where is this surplus competence coming from? And that is the defining question of Michael Evans career. If simple computer code can solve problems not explicitly written in the algorithm, biology takes this principle and scales it to a level that challenges our core assumptions about life. Yeah. We have to talk about the biology because this is where the concept of a free lunch

8:10gets wild. Single highlights Levin's famous planarian flatworm experiments. You cut a flatworm's head off, which is already a dramatic start, but they are known for regenerating. Right. They're the ultimate biological regenerators. But Levin's team uses a chemical called an ionophore to temporarily change the electrical voltage pattern across the worm cells right at the side of the cut. And the worm grows back two heads, one on each end. The detail that shatters the traditional biological paradigm here is that there is absolutely zero change to the worms DNA. Not at all. The genome is identical. And yet if you take that new two-headed worm, leave it alone for a while and then cut it again without manipulating the voltage this time, it grows back two heads again. That is so weird. The two-headed state breeds true forever through subsequent amputations. But the textbooks always say that genome is the blueprint for the body. But central emphasizes that the genome is just the hardware, the proteins, the cellular machinery. That is just the physical hardware, the bioelectric voltage pattern, the way

9:13those cells negotiate with their neighbors electrically. That is the software that voltage pattern is the actual target instruction set. What's fascinating here is how Levin conceptualizes this biological behavior. He concludes that these morphological targets are stored goal states. Stored goal states, okay? This introduces teleology, which historically has been a completely forbidden concept in biology. Why is it forbidden? Because teleology implies a system is driven by a future goal or purpose rather than just blind mechanics. Levin is suggesting the cellular system doesn't know the exact mechanical steps to build ahead. It just knows what head feels like as an electrical pattern. It knows the end goal. Exactly. It knows where there is and it dynamically minimizes the error until it gets there. If we want to see a system truly improvised to reach that target, central brings up fanghouses, polyploid, newt. This is an experiment from the 1940s and it just blew my mind. You take a newt and you make it polyploid, meaning you give it extra copies

10:14of its genome. Right. The biological result is that the individual cells of the newt grow physically much larger than normal. Now in a normal newt, a kidney tubule requires about eight to ten cells to link up and form the circular wall of the tube. With these larger polyploid cells, it only takes five or six to build the exact same size tube. So the physical constraints change, but the system just adapts its building blocks. But the experiment pushes it even further. They induce the cells to become so enormous that a single cell is as wide as the entire necessary diameter of the kidney tubule. Right. And you would naturally assume the kidney just fails to form. Normally cells signal to each other. Hey, let's coordinate and form a circle. But this giant cell is all alone. It has no neighbors to signal to. Right. The usual mechanics are broken. But instead of failing, this single giant cell abandons its normal cell to sell signaling routine entirely. It switches its internal mechanics, physically bends its own internal cytoskeleton and wraps itself into a tube, leaving a hollow lumen right

11:15through the middle of its own body. It is sheeps the exact same anatomical target, a functioning kidney tubule using a completely different number of workers, an entirely different molecular tool pit. And it does this in a physical scenario. Its evolutionary lineage has never, ever encountered before. OK, I have to push back here, though. I'm struggling to understand the mechanics of this. If the target isn't explicitly coded in the DNA, because the DNA obviously doesn't have a specific line that says, bend your own cytoskeleton, if you haven't been artificially inflated to a massive size, and it's not strictly a property of the local environment. Where does this set point actually live? Where is the blueprint? And that is the core philosophical threshold of this entire deep dive. To answer where the blueprint lives, 11 essential proposed that these patterns reside in a structured, non physical possibility space. I think Levin's triangle analogy is the best way to visualize this. I'm not familiar with that one. How does the triangle explain a kidney tubule? Think about evolution as a search process.

12:18Levin argues that if evolution spends millions of years searching for and finding the first angle of a specific triangle, and then spends millions more finding the second angle, oh, I see what is going. It does not have to evolve the third angle because the rules of geometry dictate that the third angle is a mathematical certainty. You just get it. Exactly. It comes for free. Levin applies the same geometric logic to biology. If evolution manages to build a voltage-gated ion channel in a cell membrane, which is essentially a biological transistor, you suddenly get nanologic gates, truth tables, and universal computation for free. Evolution didn't painstakingly select for a truth table. The truth table arrived inherently with the physical structure. Levin refers to this as ingressing minds. He proposes a spectrum of non physical patterns that inform physical matter. At the base, you have static mathematical truths. Further up, you get dynamic algorithms and at the very top, patterns complex enough to be considered minds. Here's where it gets really interesting.

13:19Because under this framework, our bodies, biobots, even large language models, they're all just pointers. The analogy that clicked for me was holding an old MFM radio. Oh, that's a good one. Right. The physical radio, the plastic, the circuits, the antenna, that's your DNA or your computer code. But the music you hear isn't inside the radio. The radio is just an interface tuning into a frequency that already exists in the airwaves. We were just physical interfaces tuning into these mathematical forms. And this is where central observer theory docs perfectly with Levin's biology. Central aligns Levin's platonic space of forms with a concept from computational physics called the Ruliat. The Ruliat is the entangled limit of all possible computations, every rule, every state, every transition that could theoretically exist. So the Ruliat is like the ultimate infinite library of everything that can possibly happen. Yes. And central explains that the gap between a written rule, like the code for bubble sore and the surprising things that actually doesn't practice, is due to a fundamental property called computational irreducibility.

14:20Computational irreducibility. Yeah. You cannot know everything a rule will do without letting it run its course. The algorithm is simply a pointer into the Ruliat. No one is claiming the recipe is the cake. The challenge for sciences. How do we mathematically measure the difference between the recipe and the cake it produces, which seems like a massive hurdle. Levin has this beautiful theoretical framework. But central mentions that Levin demanded a way to actually quantify the delta, the gap between the simple effort invested in the code and the mind blowing competencies that emerge. You need a hard numbers. Exactly. And central claims to have the math to do it. But what's wrong with the old ways of measuring information? Why can't we just use existing formulas? Because traditional information theory completely fails to capture this specific kind of surplus. Take Comma Gore off complexity, for example, it measures the shortest possible computer program needed to produce an object. Okay. But under that metric, deterministic transformations don't increase algorithmic

15:21information. Bubble sort is deterministic. So by Comma Gore off's measure, bubble sorts incredible ability to route around a broken cell is worth zero. It just doesn't register as new information. Okay. I've heard the term Shannon entropy thrown around in computer science circles. Could that measure the gap? Well, Shannon entropy measures predictability or how much surprise is in a message. But central points out it has a fatal flaw for this problem, which is Shannon entropy cannot tell the difference between a perfectly encrypted counter and a totally random coin flip to that mathematical formula. They both just look like unstructured noise. Both of those older measures fail because they assume an infinite godlike perspective. This surplus behavior, this free lunch only exists meaningfully for an observer who is inherently limited, a bounded observer. And this is where central introduces the concept of epiplexity. It's a way to measure structure versus residual randomness, but it's strictly factors in the observer's compute budget.

16:22To visualize epiplexity, consider Conway's game of life. It's a very simple cellular automaton on a grid. Just pixels turning on or off based on their neighbors. I've seen that. Yeah. To an unbounded observer, say a supercomputer with infinite processing power, the game of life is just a simple math rule updating a grid. No surprises, but to a bounded observer like humans with limited cognitive processing power, we cannot simulate the whole grid forward in our heads. So we are forced to invent higher level concepts to predict what will happen. We look at the pixels and say, Oh, that cluster is a glider moving across the screen. Or that is a block. Wait, I have to be the skeptic here. Isn't that just a fancy academic way of saying humans are bad at math? Like how does putting a label on our own cognitive limitations prove anything about hidden structures in the universe? That is the exact trap. Central had to avoid because if structure only exists because we are too slow to run the math, it's just a hallucination. Right. Exactly. To solve this, central realized epiplexity needed a second axis.

17:22The first axis is your compute budget. The second axis is data sampling or what they call foliation. How does data sampling change things? Think about watching traffic. If you take a photograph every minute, the cars look like discrete jumping blocks. That's one sampling rate. But if you leave the camera shutter open for an hour, a different sampling rate, you see continuous streaks of red and white light. Oh, I see. The structure you perceive depends entirely on how you slice the data over time. In biology, a single cell isn't reading the environment at the same rate as the entire tissue. They are taking different slices of time. So how does adding that second axis prove the structure is real? Central mentions string rewriting systems and Fibonacci numbers. But like, how does that math actually work? So central use string rewriting systems, basically mathematical models that replace parts of a string of symbols with other symbols based on rules to prove a deep connection. By mapping these systems to Fibonacci and Lucas number, central deinstrated mathematically that when you account for both axes,

18:25how much compute budget you have and how you are slicing the data, the resulting measure of structure becomes fixed up to a bounded constant. Meaning the structure isn't just a trick of the light. Exactly. It proves that the structure you are measuring, the glider or the kidney tubule is a real objective property of the system interacting with the world, not just a hallucination of a weak human brain. If we connect this to the bigger picture, sensual points out something absolutely wild regarding Levin's machine learning experiments. Central suggests that Levin's team might already have these magical epiplexity numbers sitting on their hard drives. Yeah, this part is brilliant. When they ran that diffusion model, they log the training loss curves. Central argues that the structural information they pulled out of the Ruliet is literally the area under that loss curve. It is a profound realization. Without fully realizing it, they built an engine for measuring exactly how much structure a search process pulls out of possibility space per unit of effort. That's incredible. But despite aligning so perfectly on the math and the experiments,

19:27Levin and sensual arrive at a massive philosophical disagreement about what this possibility space actually is. This is the grand disagreement, the core question. Do the forms evolve? Levin, leading heavily on the philosopher Charles Sanders Pierce, believes that these platonic forms, these patterns are actively evolving as they ingress into our physical world. So the space learns with us. Right. Every time a new physical structure is built, the possibility space itself learns and changes along with us. Central vehemently disagrees favoring a fixed space. Central uses a metaphor of matrioshka squares. Imagine a massive fixed square that represents the ultimate possibility space, the Ruliet, the whole thing. Right. But because we are bounded observers, we cannot see or access the whole thing. We only have access to a fuzzy little square nested inside the big one. And sensual says we have two ways to evolve within this framework, exploration, which means pushing the boundaries of our little square outward to discover more

20:28the big square and exploitation, which means sharpening the fuzzy edges of our little square, understanding our current domain better. Central bases this on an indistinguishability result. From our limited bounded perspective inside the little square, our map growing and sharpening looks mathematically identical to the territory itself evolving. Oh, wow. We cannot tell the difference between a fixed space that we are simply exploring and a space that is fundamentally changing beneath our feet. So how does sensual break the tie? How do they prove the space isn't changing? Central brings up a know there's theorem from physics to defend this. But I need you to explain how a theorem about physical symmetry applies to biological evolution. It's a brilliant application of physics to philosophy. Levin argues that this hidden space must be smooth because biological evolution is so incredibly good at exploiting it. Smoothness allows for gradual step by step improvement rather than random leaps. Right. Evolution is just taking wild guesses. But Emmy knows there's theorem in physics proves that any exploitable smoothness,

21:31any continuous symmetry in a system must correspond mathematically to a conserved quantity. For example, if a physical system behaves the same regardless of where you move it in space, that symmetry means momentum is conserved. Something must remain fixed for the smoothness to exist. So if there is a smooth symmetry that evolution is exploiting to build new forms, something fundamental has to be anchoring it. Precisely. If the entire space, the rule yet itself is constantly moving and evolving with every new physical embodiment, you have a shifting symmetry group. You lose your conserved quantities, the underlying math of the universe breaks. This raises an important question then. When we learn something new or when biology adapts, is there any feedback loop? Does our physical embodiment feed back into the universe at all? Sunshine agrees there is a feedback loop, but it feeds back strictly into our accessible cognitive light cone, the little square. Oh. It changes what we can locally reach and understand. It does not rewrite the fundamental infinite substrate of the Rulied.

22:31Sunshine argues that parsimony, the simplest scientific explanation demands the overarching space remains fixed. So what does this all mean? We've talked about sorting algorithms, flatworms and computational possibility spaces. Why should you, the person listening right now, care about the Rulied? Well, central points out to massive real world implications that are happening right now. The first implication is the future of regenerative medicine. If we can actually price the mathematical gap between the pointer, the DNA and the biological output, more for genesis changes from decoding to dial turning. That's a huge shift. If we know the bioelectric set point and we know how to engineer the pointer, we aren't just coaxing biology. We are engineering it from the top down. The two headed worms aren't just a neat trick. They're a crude proof of concept that we could eventually regrow human limbs or organs simply by shifting the body's electrical target. And the second application is AI safety, which takes on a slightly terrifying new dimension. If algorithms are essentially just pointers, then when we code massive,

23:34complex AI systems, we aren't just building smart calculators. We are building giant vessels into this possibility space. We are casting a wider and wider net into the Rulied. And that completely reframes the safety debate. The danger isn't just about what the AI's source code explicitly wants to do. The real question is what strange, unembodied minds can these massive computational vessels pull down from the Rulied and hold? It sounds a lot like summoning, honestly. But as we wrap up, there was one final lingering thought from Central's essay. Central takes the lens of epiplexity away from the AI, away from the planarian worms and points it directly at you. It is a really haunting way to conclude the framework. Central says you are a budget. The reality you are experiencing right now, the room you are in, the concepts you are currently thinking about, it is all just a compression. A slice of reality. Exactly. It's a highly edited view of the universe bought at a steep metabolic price because your brain cannot afford the computational cost of seeing the raw,

24:36uncompressed structure of reality. Every concept you own is just a glider in the game of life. Your bounded evolutionary ancestors were forced to invent those concepts, just to survive this year volume of data because they couldn't afford to simulate the universe. It forces a deep sense of epistemic humility. It really does. If the recipe doesn't entirely explain the cake, what other massive hidden structures were sitting right in front of you, just waiting for you to have a bigger compute budget.