Right now, the smartest artificial intelligence systems on the planet, I mean, the ones passing the bar exam and the 90th percentile and writing incredibly complex code in seconds, they're basically functionally mindless. Yeah, they really are. I mean, we are pouring trillions of dollars into building these synthetic brains, but we're doing it without actually knowing what a mind fundamentally is. Exactly. It's wild when you think about it. Right. We're just staring at the output of these machines completely mesmerized by the parlor tricks while sort of ignoring the fact that we don't have a clear blueprint or even like a functional x -ray of true cognitive architecture. Well, the field is suffering from a massive case of diagnostic money waters, to be honest. Oh, for sure. We're just completely obsessed with the performance of the system, but we've largely abandoned the philosophy of how a system truly thinks. how it adapts and survives. Yeah, it's all about benchmarks right now. Right. The current state of AI is entirely focused on scaling up what we already have. Just more data, more compute power, rather than stepping back to figure out the essential physics of cognition. So today's deep dive is really about fixing that exact blind spot. We are unpacking a text that feels like a totally necessary antidote to the endless hype cycle of generative AI. It really is. The book is called On Minds, Reflections of a Cognitive Architect by Paul S. Rosenblum. And for you listening right now, whether you are, you know, trying to understand where the future of technology is actually heading or you're just fascinated by human psychology, this discussion is designed to be your shortcut to clarity. Yeah, and Rosenblum's perspective is incredibly rare because he isn't just like a casual commentator or someone who jumped on the AI bandwagon post... chat GPT. Right. He's been around. Oh, he has spent over half a century in the trenches of artificial intelligence and cognitive science. But more importantly, he approaches this field as an architectural explorer. I love that term, architectural explorer. It's a great way to put it. He isn't satisfied with just writing a clever algorithm that can play a game. He wants to know what underlying universal structures are actually required to build a complete functioning mind. Which shifts the framing completely. We aren't just looking at cold lines of code today. I mean, we are looking at the foundational philosophy of building minds. Exactly. Taking these incredibly dense academic meta -scientific reflections and translating them into those those aha moments that change how you look at your own thought processes. That's our mission today. And the timing of Rosenblum's work couldn't be more critical. I mean, the tech world is currently consumed by this white -hot obsession with deep learning and generative models. It's literally everywhere. You can't escape it. You can't escape it. Right. It's a monolithic trend. But Rosenblum forces us to look past this current moment. He really demands to what it actually takes to build a full -spectrum mind, not just a system that can predict the next word in a sentence, but a complete cognitive architecture that experiences, learns, and navigates reality. We basically have to learn how to think like him. Right. Early in the text, he lays out his meta -scientist toolkit. And he starts by categorizing the entire field into three distinct levels of subject matter. Yeah. So if you imagine a hierarchy of understanding, right? At the most concrete foundational level, you have cognitive architectures. OK. These are the highly specific fixed structures and computational processes that theoretically yield a mind. So if we use a biological analogy, this is the literal physical layout of neurons, like this specific wiring of the hippocampus and the prefrontal cortex. That is exactly it. It's the hardware design, basically. Then moving up one level of abstraction, you have the common model of cognition. Which is what, exactly? This is an ongoing, massive attempt by the global scientific community to find consensus. It's the effort to look at all the different cognitive architectures researchers have built and say, OK, what features absolutely must be present in any architecture capable of human -like cognition? Oh, I see. So it's like an anatomy textbook that defines what a mammal is, regardless of whether it's a mouse or an elephant. That is a perfect analogy. Yes. And finally, at the top, the most generic and abstract level, you have minds themselves. Just minds in general. Right. The living, breathing entities, whether they are naturally evolved, like ours, or artificially constructed in the lab. Okay, so to study these three levels, architectures, the common model, and minds, he argues that we can't just use the standard scientific method, right? No, he thinks it's the wrong tool for the job. He proposes three... specific forms of analysis, I think. Essential, dichotomic, and relational. But the one that really anchors the first part of this entire journey is what he calls the reflective method. Yeah, the reflective method is key. He actually argues that the traditional scientific method is fundamentally flawed for this specific type of foundational work. Wait, really? The scientific method. That's a pretty bold claim. Well, think about it. The traditional scientific method. where you observe a phenomenon, formulate a hypothesis, run a controlled experiment. It is exquisitely designed for testing specific observable things in the physical world. Right, like testing what temperature water boils. Exactly. Or how a specific chemical reacts. But Rosenblum is dealing with meta -science, science about science. Oh, I see. When you are trying to define massive, historically loaded, abstract concepts like intelligence or learning or symbols, running a lab experiment doesn't actually help you. Because you can't put intelligence in a petri dish. Precisely. If researchers don't even agree on what a symbol is, an experiment involving symbols is totally useless. You need a method that identifies essences first. Right. So the reflective method is a rigorous, generalized framework for stepping completely out of the lab and rethinking the foundational categories we use to organize reality. Yeah. And the absolute linchpin of this reflective method is what he identifies as step three, finding the absolute essences of these topics. He talks about lumping before splitting, right? Yes. finding this incredibly delicate balance of deciding what a crude historical learning we take with us into the future and what we completely leave behind to change our perspective. You know, when I was trying to visualize this process, the only analogy that worked for me was Marie Kondo. The home organization philosophy applied to cognitive science. Hear me out. I mean, it's like he's Marie Kondoing decades of scientific dogma. He's picking up this incredibly complex, historically burdened AI theory. holding it up, looking at all its bells and whistles and asking like, what is the absolute bare minimum essence of this concept? Oh, I see where you're going with this. Does this spark functional intelligence? Exactly. What is mandatory for a mind and what is just an optional historical elaboration? He's stripping it down to see what actually sparks joy or in this case, intelligence. That's actually great. Taking that Marie Kondo analogy a step further. In fields that are utterly flooded with jargon, competing sub -disciplines, and frankly, massive egos, stripping away those non -essential elaborations isn't just about making things simple for the sake of a tidy textbook. Right. It has a real purpose. It is a vital structural necessity. I mean, look at the history of science. Isaac Newton was a massive lumper. Really? How so? Before Newton, people thought the forces pulling an apple to the ground were entirely different from the forces keeping the planets orbiting the sun. Oh, wow. I didn't realize that. Yeah. Newton stripped away the non -essential details, the fact that one is a fruit and the other is a celestial body, and lumped them together under the essential concept of gravity. And Rosenblum is trying to do exactly that for artificial intelligence. Exactly. By stripping a concept down to its absolute functional core. You prevent researchers from spending decades arguing over semantics. You force everyone to look at the underlying physics of cognition. Which is exactly why you, listening right now, should care about this meta -scientist toolkit. I mean, the next time you open your news feed and read a breathless headline about a massive new AI breakthrough that will change the world, applying this reflective method allows you to cut straight through the marketing speak. It really does. It gives you a filter. You can ask yourself, wait, What core architectural change actually just occurred here? Or is this just the same essence dressed up in a new shiny elaboration? And Rosenblum doesn't waste any time putting this tool to work. He immediately aims his essential analysis at the two most debated foundational and polarizing words in the entire history of artificial intelligence. Architectures and symbols. Yes, the big ones. Let's start with the building blocks. Architectures. He traces the entire lineage of the concept of an architecture, all the way from physical buildings, like the Roman Coliseum, to computer hardware, and finally up into the realm of cognition. And applying his essentialist analysis, stripping away all the noise, he determines that the absolute, irreducible core of an architecture is fixity. Fixity? Fixity. Meaning it doesn't change. Right. An architecture provides the fixed structures and fixed processes that ultimately yield a system. If you want a mathematical equation to anchor this, Rosenblum defines it very clearly. System equals architecture plus variable content. Okay, let's map that onto the listener's daily experience. System equals architecture plus variable content. Let's do it. So if my mind, my personal consciousness is the system, the architecture is the fixed biological machinery I was born with. Correct. It's the structure of my hippocampus, the neural pathways in my prefrontal cortex, the biological limits of my short -term memory. And the variable content is, well, everything I've experienced. My memory of the route I take to work, my ability to speak English, my political beliefs. Precisely those things. The variable content consists of the constantly changing skills, knowledge, and memories that your fixed architecture acquires, stores, and processes over a lifetime. OK, that makes sense. The architecture is the permanent stage, right? The variable content is the infinite number of different plays being performed on that stage over the course of your life. That makes perfect sense. But here is where the text gets highly controversial, especially for anyone who has studied the history of computer science. He turns his essential analysis on the word symbols. Oh yeah, this is a big deal. To fully grasp the magnitude of this controversy, we have to look at the historical schism in artificial intelligence. The Civil War. Exactly. Since the earliest days of AI, going back to the 1950s and 60s, there was this guiding philosophy called the Physical Symbol Systems Hypothesis, or PSSH. This was developed by the founding fathers of AI, giants like Alan Newell and Herbert Simon. This was the bedrock idea that manipulating symbols based on explicit rules is both necessary and sufficient to achieve general intelligence. That was the gospel for decades. This era is sometimes called good old fashioned AI or Goffi. Goffi. Goffi. I like that. The idea was that if you want a computer to understand a dog, you literally code a discrete symbol, a digital token, for dog. Right. Then you write explicit, rigid rules about how that dog symbol interacts with a barking symbol or a fur symbol. It is highly logical, highly readable by humans, and incredibly brittle. Because the real world is messy. Exactly. But the modern tension, the massive schism, is that neural networks, the deep learning models driving today's explosion in AI capabilities, they do not use traditional symbol manipulation at all. Not at all. Nope. They use continuous mathematical equations, adjusting millions of numeric weights and activations across a massive web of artificial neurons. So for decades, there has been a bitter civil war in AI research. The traditional symbolic rule -based camp on one side and the continuous mathematical neural network camp on the other. Exactly. And Rosenblum looks at this decades -long civil war and basically says, you're both arguing over elaborations. Let's use essential analysis. Right, the Marie Kondo approach. Yes. He takes the concept of a symbol and strips it down to its absolute bare minimum. He ruthlessly removes the traditional historical requirements. In the classical Newell and Simon era, a symbol had to possess composition, right? Meaning it could be snapped together with other symbols like Lego bricks to build complex thoughts. Right. And it had to possess designation, meaning it explicitly pointed to a specific identifiable thing in the real world. But Rosenblum strips all of that away. He does. He decides the true irreducible essence of a symbol is simply an atom or a placeholder. By stripping it down to a mere placeholder, he is able to propose a massive unifying umbrella, the hybrid symbol systems hypothesis, or HSSH. Right. And this new hypothesis is designed to cover both the traditional, rigid, rule -based symbolic AI and the fluid, continuous math of modern neural networks. You know, this is where I found myself pushing back pretty hard while reading. Oh, really? Yeah. I mean, if you strip a word like symbol down to just being a generic, undefined placeholder, haven't you watered the definition down so thoroughly that it completely loses its scientific utility? That's a fair question. Like, is he just trying to force modern neural networks into an old 1970s conceptual box so that the old guard of cognitive architecture doesn't feel historically left out of the current AI boom? It is a totally natural skepticism, and it's a critique frequently leveled at lumpers, as we talked about earlier. But the defense of this move is rooted in the search for a universal physics of mind. OK, unpack that for me. This isn't watering down the definition. it's finding the least common core. If you ever want a unified theory of mind, if you want to build an AI that can intuitively recognize a face, like a neural net, but also logically deduce a mathematical proof like a symbolic system, you cannot have two completely incompatible scientific vocabularies. Oh, so they literally need a shared language. Exactly. By removing the strict rigid rules of what a traditional symbol must be, Rosenblum creates a conceptual space where we can actually talk about how neural networks and symbolic logic integrate. So he's arguing that even a tiny anonymous node firing deep inside a massive neural network operating purely on a mathematical weight of 0 .8 can be viewed as a minimal essential symbol? Yes. It's a placeholder, representing some abstract feature of the world, even if we, the human programmers, have absolutely no idea what it explicitly designates. That synthesis is the breakthrough. It stops treating these two camps as fundamentally incompatible alien species and starts treating them as different structural expressions under the exact same genus. We have to agree on the raw materials before we can build the house. Exactly. But if these hybrid symbols and fixed architectures are the raw materials, what exactly are we trying to build? I mean, the tech world throws the word intelligence around constantly. Oh constantly. Everything is smart now. Right. Every startup claims their software is intelligent. But without a clear, rigorous definition, we're building that house without a blueprint. We really need to define the goal. And Rosenblum is acutely aware that the cacophony of definitions for intelligence is just deafening. Biologists have one definition, software engineers have another, philosophers have a third. It's a mess. So he attempts to bring structural clarity by dissecting the concept of intelligence and organizing it into a nested three -level structure. This is the Russian nesting doll model of intelligence. Yes. Think of the smallest doll, the innermost core. He labels this level one immediate intelligence. And what does that entail? This is the raw ability to process rapid perception, recognize patterns, and make immediate single step decisions based on the current context. Which perfectly describes exactly what large language models, the chat GBTs and clods of the world currently excel at. Absolutely. You give them a prompt, they rapidly calculate the mathematical probabilities across their massive neural net, and they immediately generate the next most likely word. It is a highly sophisticated but immediate reflex. But if you open that smallest doll, you find it sits inside a larger, more complex structure, level two, cumulative intelligence. OK, how is that different? This represents lifelong pervasive incremental learning. It's an intelligence that learns online, constantly updating its internal models, correcting its assumptions, and integrating new knowledge across an entire lifetime of experiences. Which is where current generative AI struggles massively. Right, because an LLM doesn't actually learn from you while you're talking to it. Its core training is permanently frozen. Yeah, I've noticed that. It has a temporary context window for your current conversation, sure. But the moment you close the tab, it basically forgets you. It resets completely. It isn't fundamentally altering its base neural architecture with every single interaction. The way, say, a human toddler does when they touch a hot stove and permanently rewire their understanding of the world. That's a great example. The toddler possesses cumulative intelligence. The LLM currently does not. And then that level two doll sits inside the largest all -encompassing doll, level three full spectrum intelligence. The whole package. Right. This encompasses absolutely all aspects of a mind. It includes internal goal generation, physical sensations, complex temporal planning, and true holistic awareness of its environment. And Rosenblum is actually quite blunt about where our current technology sits, isn't he? Oh, very blunt. He points out that today's generative AI systems possess massive memory banks and incredibly superhuman immediate intelligence, but architecturally speaking, they are essentially mindless. It sounds incredibly harsh, especially given how fluent and convincing these systems are, but from a cognitive architecture standpoint, it's objectively accurate. Because they lack that level two and three stuff? Exactly. They lack the pervasive incremental learning loops of a true architecture. They are functionally static once their training run ends. They are just processing your input through millions of frozen mathematical weights. Returning to that Russian nesting doll analogy for a moment, right now the entire tech sector, the global stock market, the media, they are hyper -focused on the smallest doll, immediate intelligence. Because it's shiny and new. Right. But because the output is so convincing, we are desperately trying to paint that tiny doll to make it look like the biggest doll, full spectrum intelligence. We treat chatbots as if they have internal goals, beliefs, or self -awareness. And Rosenblum highlights that this projection is exactly what leads to the dangerous hallucinations we see in AI. We are demanding full spectrum reliability from a system that only possesses immediate reflex level architecture. That framing clarifies so much of the current AI panic, honestly. And it ties directly into his exploration of what a theory actually is. Yeah, because building a cognitive architecture is ultimately building a unified theory of mind. He boils the essence of a theory down to the concept of aboutness or what he calls meta systems. Aboutness. Right. A theory is simply a system that is about another system. A map is a theory about a territory. And when building these theories, these architectures, he introduces a contrast that I found deeply fascinating. The pursuit of functional elegance versus the trap of exuberance. Functional elegance is his holy grail. It is the goal of achieving incredibly broad, complex, human -like capabilities from a very simple, theoretically elegant base of fixed mechanisms. So doing more with less? Exactly. It's the architectural equivalent of doing a tremendous amount with very little. Exuberance, on the other hand, is the brute force approach. Just throwing everything at the wall. Throwing endless parameters, massive server farms, oceans of scraped internet data, and sheer computing power at a problem until it magically starts working. Which is the exact definition of modern massive neural networks. Yep. They are pure exuberance. We don't fully understand how they work. We just keep adding trillions of parameters and hoping intelligence emerges from the sheer scale of it all. And this dichotomy highlights a crucial difference between architectural understanding and mere cognitive understanding. LLMs have a built -in mechanism for learning during their massive training phase. They predict a word, check if they are right, and mathematically adjust their weights. Right. But they entirely lack what Rosenblum identifies as fundamental, low -level architectural appraisals. He talks about architectural emotions, which is just a wild concept to me. I know. It sounds weird at first. But he isn't talking about a computer feeling sadness or joy in the human romantic sense. He's talking about biological mechanisms translated into code. Let's look at the architectural appraisal of surprise. In a true full -spectrum cognitive architecture like your brain, your mind is constantly predicting what will happen a fraction of a second in the future. Constantly. Right. If your mental model predicts one thing but reality delivers something completely different, Your underlying architecture experiences a low -level biological appraisal of surprise. Okay, think about the listener's daily commute. You drive the exact same route to work for five years. Your brain is essentially on autopilot, highly efficient, predicting every stoplight and turn. We all do it. But one morning, you turn a corner and see a massive sinkhole blocking the road. Your brain doesn't just crash, it experiences an immediate spike of architectural surprise. And that surprise isn't just a feeling. It is a vital, functional trigger. That spike forces your cognitive architecture to instantly break out of its habitual cycle, rapidly shift attention, adapt its internal models, and begin learning incrementally to resolve the discrepancy between its prediction and reality. It forces your variable content to permanently update. Exactly. Large language models simply do not have this. They don't experience architectural surprise. If they predict a totally nonsensical word, they don't pause, feel a discrepancy and restructure their worldview. They just plow ahead. They confidently output the next most probable token, even if it totally contradicts reality. They lack the architectural mechanism of surprise that forces cumulative learning. This shifts the entire paradigm of how you interact with a system like ChetGPT. You can now clearly see it not as an emerging mind, but as an unimaginably vast ocean of human variable content. Literally the entire digitized text of human history. All of it. And it is being processed without any of the fixed, cumulative architectural structures required for actual intelligence. It's the world's largest database of human variable content. being blindly processed by a very narrow, purely immediate intelligence architecture. So we have the raw materials defined, these essential hybrid symbols and fixed architectures. We have the ultimate goal defined, full spectrum intelligence achieved through functional elegance. Right. But if you look at the actual landscape of AI research today, it is total chaos. Rosenblum mentions there are over 195 distinctly different cognitive architectures out there. It's a very crowded field. There are thousands of competing papers, endless subfields, tribal academic wars. How do we navigate this wilderness without getting completely lost in the weeds? To navigate the wilderness, you need a map. This is where Rosenblum shifts gears entirely. We move from the essential analysis of defining concepts into dichotomous analysis. He sits down to build a comprehensive map of the entire artificial intelligence landscape. Dichotomic analysis. Meaning analyzing complex systems by breaking them down into fundamental dichotomies or binary switches. Yes. Most standard textbooks on AI try to organize the field by human functions. You'll find a chapter on computer vision, a chapter on natural language processing, a chapter on robotic planning. Which makes sense to us because that's how we categorize human skills. It feels intuitive. But Rosenblum argues that categorizing AI by human skills is like categorizing vehicles by what cargo they carry, rather than how their engines actually work. That is a brilliant comparison. Right. He rejects the functional approach entirely. Instead, he looks at the underlying mechanical physics of information processing. He identifies three essential dichotomies, three binary switches, And by looking at the cross products of these switches, he creates an eight region, three dimensional map that categorizes every single AI system ever built. I want to walk through these three switches incredibly carefully because this is where the actual engineering mechanics of AI get exposed. The first switch is dichotomy one, asymmetric versus symmetric processing. How do these two mechanisms actually work at the base level? This dichotomy describes the directional flow of processing. Think about how information moves. Asymmetric processing is strictly directional. It moves in a one -way street. Okay, so like a straight line? The classic example in computer science is an if -then rule. If the condition is met, then execute the action. It is a rigid, causal chain. Symmetric processing, on the other hand, is omnidirectional. I love the physical metaphors here. Asymmetric processing is like a Rube Goldberg machine. A marble rolls down a track, hits a specific lever, which strictly triggers a specific pulley, which drops a specific bucket. The action only flows forward step by step. Exactly. But symmetric processing is like a spider web. The spider web is the perfect analogy. In a symmetric neural network, you have thousands of interconnected nodes. If you pluck one string on a spider web, the vibration travels omnidirectionally. It goes everywhere at once. Every node influences every other node simultaneously. The entire web adjusts and resonates until the whole system settles into a state of equilibrium. There is no strict A to B to C pathway. There is just a holistic settling into a pattern. OK, Rube Goldberg versus Spiderweb. Got it. The second switch is dichotomy 2, controlled versus uncontrolled processing. This dichotomy is about the mechanism of deliberation. In a controlled process, a decision explicitly structurally precedes an action. So it thinks before it acts? Yes. The system pauses, evaluates multiple different options, selects the optimal one based on a goal, and then executes the action. It requires deliberate architectural control. An uncontrolled process requires no such pause. It just acts? It executes automatically, on its own impetus. The moment the input is received, it is entirely reactive. To bring this back to the human experience, it's the difference between reflex and deliberation. If I accidentally touch a hot stove, my hand violently pulls back before I even consciously realize what happened. That is an uncontrolled, automatic, biological reaction. Absolutely. But if I am standing in my kitchen deciding what to cook for dinner... I have to pause, evaluate what's in the fridge, consider my dietary goals, and make a deliberate choice before acting. Rosenblum maps that exact same biological dichotomy directly onto AI systems. It does. When a deep neural network recognizes a human face in a photograph, it doesn't pause and decide to see a face. It just sees it. The pixel data enters the input layer and triggers an automatic, uncontrolled cascade of mathematical activations through the web until the output layer fires. It is an incredibly complex reflex. Which brings us to the final switch, dichotomy three, data versus metadata. This dichotomy circles back to our earlier debate about symbols. Does the core processing of the system operate on symbolic data, which means discrete categorical structural elements like a specific word, a logical concept, or a distinct object? Or? Or does the system operate on numeric metadata? which means continuous quantitative values like probabilities, mathematical weights, or activation thresholds. So we have three binary switches governing the physics of AI. The flow of information, asymmetric versus symmetric. The mechanism of action, controlled versus uncontrolled. And the nature of the information itself, data versus metadata. That's the 3D map. By crossing these three switches, you generate an eight region 3D map. When I first encountered this map in the text, I was frankly highly skeptical. Why's that? Can you really take the sprawling 60 -year history of artificial intelligence from the clunky 1980s expert systems diagnosing blood infections, all the way to modern miracles like AlphaZero mastering the game of Go, and reduce all of that staggering complexity into just three binary toggles? I mean, it feels way too neat. It is a profound level of lumping. which naturally invites skepticism. But the power of this map isn't about reducing the field into a simplistic cartoon. It's about discovering the foundational physics that govern how these incredibly diverse systems actually operate. Give me an example. Let's look closely at the example you just mentioned, Alpha Zero. Alpha Zero is the Google DeepMind AI that famously mastered chess, Shogi, and Go entirely from scratch, playing against itself. Which is still mind -blowing. It is. If you map AlphaZero's architecture onto Rosenblum's 3D grid, you see that it beautifully spans across multiple dichotomies. And that structural spanning is the exact mechanical reason it is so devastatingly powerful. Okay, walk me through the mechanics of that. How does AlphaZero physically map onto these switches? When AlphaZero looks at a complex chessboard, it first uses deep neural networks to evaluate the state of the board. It looks at the arrangement of pieces and intuits which overall positions are structurally strong. Just a quick glance, basically. Right. That rapid valuation process lives in one very specific region of our 3D map. It is symmetric. It uses the omnidirectional spider web of a neural net. It is uncontrolled. It happens as an automatic reflex upon seeing the board. And it relies on metadata. It calculates numeric probabilities of winning. So it has this incredibly fast, intuitive, uncontrolled gut feeling about the board, much like a human grandmaster looking at a setup and just feeling the tension on the squares. Exactly. But it doesn't just play its first gut feeling immediately. It takes that intuitive evaluation and feeds it into an entirely different algorithmic process called Monte Carlo Tree Search. This is where it explicitly plans out future moves. And how does Monte Carlo Tree Search work mechanically? Imagine the AI standing in a maze. Instead of just running down the path that looks best, it pauses and sends out thousands of invisible ghosts down all the different branching paths. Like scouts. Yeah. These ghosts play out the game, thousands of moves into the future, reporting back on which specific sequence of moves actually leads to a victory. This search process lives in a completely different region of Rosenblum's map. Which switches does it hit? It is asymmetric. It follows directional branching pathways of cause and effect. And it is controlled. It explicitly evaluates these future realities before choosing the optimal path and taking a single action in the present. Oh, wow. It's like Dr. Strange viewing 14 million alternate futures before making a single move. That is a brilliant comparison. Yes. So it leverages the symmetric uncontrolled side for rapid intuition and the asymmetric controlled side for rigorous, deliberate planning. It marries the two beautifully. And this 3D map proves something vital for the entire future of the field. The path forward isn't about one side of the academic war winning. Right. It's not about symbolic logic, permanently defeating neural networks, or deep learning proving that symbols are useless. The future of true full spectrum intelligence requires spanning these dichotomies to get the absolute best of both structural worlds. You need both. desperately need the fast, uncontrolled, symmetric metadata of neural nets to handle messy real -world perception, integrated flawlessly with the controlled, asymmetric symbolic reasoning of traditional architecture to handle logic, planning, and goal -setting. The map reveals that everyone in the field actually needs each other. It transforms a messy, tribal academic war into a structured, navigable landscape. It does. But you know, the creation of a map that elegant naturally leads to a question about the map maker. Who builds a framework like this? This brings us to the final section of the book, which I found incredibly moving. We transition from the science to the scientist himself. Part 3, The Architect's Memoir. Yeah, Rosenblum shifts gears to explore his own deeply personal, triphasic career as an architectural explorer. And it is fascinating that a career dedicated to the highest levels of artificial intelligence began with a piece of plastic from a toy store. The Hasbro Thinkatron. The Thinkatron. He describes this toy from the 1960s. It was advertised as a computer that thinks like a man. It was basically a hollow plastic shell. You fed it these little punch cards with multiple choice trivia questions. A plastic gear would grind inside and it would spit out the right answer. Very high -tech. Mechanically, it was just a simple sorting machine. But to a child, it sparked the illusion of a synthetic mind. It planted the seed for a life entirely dedicated to building the real thing. And that childhood seed grew into what he categorizes as his grand challenge. his tier one architectural exploration. For decades, he was at the absolute forefront of the field. He really was. He worked alongside legendary figures like Alan Newell and John Laird at Carnegie Mellon during the golden era of AI, helping to create the SOAR cognitive architecture. For listeners who might not know, SOAR is one of the most famous, longest lasting cognitive architectures in the history of the field. It was the gold standard for symbolic reasoning. It was a monumental achievement. But Rosenblum's relentless pursuit of that functional elegance we discussed earlier eventually led him to step away from the SOAR project to develop his own entirely new architecture from scratch, which he called Sigma. Sigma. Sigma was a massive, ambitious attempt to unify graphical models, which is the complex math underlying probabilities in modern neural networks, with traditional cognitive architecture. He wanted to build an architecture that inherently spanned those dichotomies we just mapped out. But his career wasn't just about building individual systems. He also talks about his tier two exploration, the common model of cognition. Right. This was his effort to herd the cats of the scientific community and find actual consensus. And buried in this section is a scientific nugget that absolutely blew my mind. I know exactly what you're going to say. The common model, based on intense study of human neurobiology and various AI models, relies on a cognitive cycle that operates at roughly 50 milliseconds. The roughly 50 millisecond cycle. This is a profound biological reality. In human brains, cognitive processing is not a smooth, continuous flow of information. It is a discrete, rhythmic loop. It's like a heartbeat for thoughts. The cycle of perceiving sensory input processing it cognitively and initiating a motor action bottoms out at a biological speed limit of about 50 milliseconds. The common model adopts this rhythm as a fundamental requirement. It suggests that human -like intelligence is inherently cyclical. It's the biological clock speed of human thought. Exactly. And the engineering implication is massive. If you want to build a truly human -like synthetic mind, you likely cannot just build a continuous feedforward system. You have to engineer that rhythmic, cyclical, biological architecture into the silicon. His ambition just keeps scaling up. In what he calls Tier 4, the relational model, he actually takes an entire decade away from building cognitive architectures altogether. Just steps away entirely. He steps back from the lab and writes a book attempting to establish computing as the fourth great scientific domain, arguing that it deserves to stand right alongside the physical sciences, the life sciences, and the social sciences. He was zooming out as far as humanly possible, trying to define where the concept of computing even sits within the entire history of human knowledge. But the part of this memoir that really stopped me, the part that grounds all of this dense meta -scientific theory, was the personal trials. Oh, absolutely. Deep in the middle of this academic text, Rosenblum reveals his brutal battle with AML, acute myeloid leukemia. And this diagnosis didn't just happen... at any time. It happened right during the height of the COVID -19 pandemic. The Jackson position is incredibly stark. He describes the absolute terrifying isolation of undergoing cancer treatment during a global lockdown. The hospitals were closed to visitors. No one could come see him. He endured grueling bone marrow transplants completely alone. He faced the incredibly grim survival statistics associated with AML. And he makes it through. There's this visceral deeply human moment where he describes finally being wheeled out of the hospital, officially discharged, and the hospital staff lining the halls to give him a standing ovation. It's a beautiful moment. To read about that moment in a book dedicated to the abstract mathematics of intelligence, it completely recontextualizes the work. It grounds the entire intellectual pursuit in human fragility. It does. But it also throws into sharp relief an incredible paradox that he writes about quite openly. Here is a man who was stationed at the absolute epicenter of AI history, Carnegie Mellon in its prime. Yep. Decades at the University of Southern California's Information Sciences Institute, he worked with the titans of the field. And yet, despite being at the center of the universe, he describes feeling like a perpetual outsider throughout his career. How does someone with that pedigree feel like an outsider? Well, it ties directly back to everything we've unpacked today. Remember his methodology, lumping versus splitting. Right. You have to look at the economic and academic incentive structures of modern science. The entire system aggressively rewards splitters. Researchers are incentivized to carve out tiny, highly specialized niches, publish 100 iterative papers on one minor algorithmic tweak, and ruthlessly chase whatever the latest white -hot, venture capital -funded trend is. Which right now is pure deep learning. Exactly. If you are a lumper in that environment, you naturally stand apart. Because you are actively trying to unify the field rather than carve out a profitable slice of it. Exactly. When you prioritize structural functional elegance over whatever is currently winning benchmark tests, you are going to feel alienated. You are the architect trying to figure out how to build the entire watch while everyone around you is getting famous and wealthy for slightly optimizing the shape of one single gear. And his brush with mortality, surviving AML, seemed to act as a profound catalyst for this memoir. He describes returning to his work with a new perspective, something he beautifully calls the freedom of the superannuated. The privilege of the superannuated. Having survived a highly lethal cancer and entering the late stages of a long career, he realized he was finally free. Free from the rat race. He no longer had to care about academic trends or the grueling cycle of securing grant funding or what the deep learning purists thought of his unified theories. he had the ultimate intellectual freedom to just sit down and explore the deep, philosophical, unifying theories of mind that truly mattered to him. It's incredibly freeing. This book is the result of a scientist who realized he had absolutely nothing left to lose and an entire lifetime of knowledge left to synthesize. So we have covered an immense amount of ground today. Let's bring this journey full circle for you, the listener. We started by looking at the murky X -ray of current AI. recognizing that we are building minds without a blueprint. Right. We zoomed all the way in to the microscopic definition of a single symbol, redefining it as a fluid hybrid placeholder. We defined the massive goal of full -spectrum intelligence, recognizing that today's chatbots, while possessing incredible immediate intelligence, are architecturally mindless because they lack cumulative learning and architectural surprise. And then we zoomed out to a three -dimensional map that plotted the entire history of AI across three underlying physical dichotomies. And we anchored all of that theory in the deeply human story of an architectural explorer surviving cancer to write a unified theory of his life's work. The ultimate goal here is that the listener is now armed with a robust mental framework. The next time a tech CEO steps onto a stage and announces a new AI model that is supposedly approaching human intelligence, You don't have to just accept the hype. You have Rosenblum's toolkit. You can actively ask the structural questions. Is this new system demonstrating true functional elegance? Or is it just another display of massive computing exuberance? Does it possess the architecture for cumulative lifelong learning? Or is it just performing incredibly fast, immediate parlor tricks? Does it successfully span the dichotomies of intuition and deliberation? Or is it stuck in one region of the map? You essentially have the x -ray machine now. You can look past the impressive conversational output and see the structural architecture, or the lack thereof, beneath it. The structural perspective changes everything. But before we let you go, there is one final highly provocative thought that Rosenblum leaves us with, almost as a brief philosophical detour at the very end of his exploration. And it has to do with the concept of free will. Oh, this is fascinating. He directly addresses a very popular argument made by a prominent Stanford neurobiologist, Robert Sapolsky. Right. Sapolsky recently published a comprehensive book arguing strongly, based on biology and physics, against the existence of free will. Sapolsky's argument is that humans are ultimately biological machines. Every action we take is entirely predetermined by a complex chain of neurochemistry, genetics, trauma, and environmental stimuli. So it's all mapped out already. Therefore, the concept that we possess independent free will is a complete neurological illusion. It's a very common mathematically sound, if somewhat bleak, scientific perspective. We are all just complex dominoes falling in a chain reaction that started at the Big Bang. But Rosenblum tackles this argument not just as a biologist looking at chemistry, but from the perspective of a cognitive architect who has spent 50 years trying to engineer a mind from scratch. and he introduces a brilliant structural twist to the debate. What's the twist? He essentially argues that from an architectural standpoint, it absolutely does not matter if the universe is strictly deterministic. Wait, it doesn't matter if free will is real or not? Functionally, mechanically, no. Because if you want a mind to operate at level three, full spectrum intelligence, if you want a system to function, adapt to unexpected surprises, evaluate complex moral choices, and initiate deliberate controlled action across a lifetime. That architecture must possess the internal belief that its goals matter and that it has the active agency to achieve them. So the system literally has to believe it possesses free will simply in order for the cognitive software to successfully run? That is the architectural requirement. Whether you are a human being trying to navigate the complexities of society, or a highly advanced synthetic AI trying to achieve general intelligence. The illusion of free will is not just a psychological quirk or a comforting lie we tell ourselves. It is a strict structural necessity. Without the internal representation of agency, without the system believing that its choices actively alter the future, the architecture cannot evaluate options and initiate controlled action. Wow. The underlying engine requires the belief in choices to function, completely regardless of whether the laws of physics say those choices were predetermined 14 billion years ago. That is a staggering thought to end on. The illusion of free will is an unavoidable structural requirement for the existence of intelligence. The architecture demands it. That is something for you to mull over the next time you make a conscious choice today. Whether your decision to listen to this deep dive was entirely predetermined by a chain of cosmic dominoes, or whether you truly independently authored the decision, your cognitive architecture required you to fully believe in that choice either way, just so you could get out of bed this morning and hit play. Keep exploring the incredible architecture of your own mind, and we'll see you on the next deep dive.