Part 1 | The Architecture of Intuition

What if the uncanny, almost mystical "gift" of a veteran wargamer predicting an enemy's hidden position isn't a supernatural trait at all, but a highly refined biological process? In Part 1 of this episode, Will strips the mythology away from expert intuition, grounding it in decades of Nobel Prize-caliber cognitive science. We explore the history of simulating the "fog of war"—from 19th-century Prussian military exercises to high-fidelity digital algorithms—and map exactly how the human brain internalises constraints to read through the unseen. 

Key Topics Covered

• The Anatomy of the Fog of War: Understanding how uncertainty and command friction have been systematically engineered into simulations, beginning with the 1824 Prussian Kriegsspiel double-blind referee system. 
• Physical Mechanics of Concealment: How commercial tabletop designs replaced resource-heavy umpires with elegant systems like the upright wooden "block wargame" and non-deterministic "chit-pull" activation markers. 
• Digital Formalisation & Logistical Physics: A deep dive into modern digital simulations like Gary Grigsby's War in the East 2, exploring its dynamic mathematically rigorous "Detection Level" algorithm and the brutal historical realities of Operation Barbarossa's supply constraints.
• Deconstructing Expert Intuition: Unpacking the landmark psychological research of Herbert Simon and Adriaan de Groot, proving that "intuition is nothing more and nothing less than recognition" operating across a massive parallel database of internalised patterns. 
• The Recognition-Primed Decision (RPD) Model: Examining Dr. Gary Klein's framework for how real-world professionals make high-stakes, split-second decisions through simple matching, feature matching/story building, and mental simulation. 
• The Kahneman-Klein Boundary Conditions: The fascinating adversarial collaboration that settled when expert intuition can actually be trusted based on two specific environmental factors: high-validity environments and rapid, unambiguous feedback loops. 

Notable Quotes from the Episode

"The fog of war is not random noise. It is structured uncertainty, governed by explicit rules. And structured uncertainty can be deduced by anyone—or anything—that has internalised the structure well enough." 

"What felt like a thunderclap of intuition from the inside was actually a story-building process running at the speed of subconscious computation. The mysticism is a misreading of the experience. The mechanism is entirely legible." 

Transcript

SPEAKER_01 0:00

There is a moment in competitive wargaming that veterans describe almost reverently. You are deep into a simulation of Operation Barbarossa, the Eastern Front, summer 1941. The map stretches across the table like a small continent. Your German forces are visible. Your opponent's Soviet reserves are not. They are hidden, somewhere behind a curtain of structured uncertainty that the hobby calls the fog of war. And then the player across from you, 40 years of this in his bones, does something that makes the room go quiet. He does not deliberate out loud. He does not consult any notes. He reaches across the hexagonal grid, places a finger near a rail junction southeast of Kharkov, and says, Third Guards Tank Army, staged here. They're going to swing south and cut your supply corridor in three turns. Everyone waits. The hidden markers are revealed. He is right. Not directionally right, not roughly right, precisely, uncomfortably, embarrassingly right. People in that room will call it instinct. They will call it experience. They will call it, and I have heard this word used with genuine reverence, a gift. Science has a different word for it, and that word, when you understand it fully, opens up one of the most provocative questions in all of artificial intelligence research. Because here is what I want you to sit with before we go any further. What if that gift, that uncanny, apparently inexplicable ability to see through the fog, is not a human gift at all? What if it is just a process? A process that a machine can run too? What does that change about expertise, about AI, about how we should make the most important decisions in the world? We are about to find out. This is the show where we follow ideas wherever the evidence leads, across disciplines, across assumptions, across the comfortable stories we tell ourselves about what makes us uniquely human. Today's episode lives at a crossroads I find genuinely electrifying: military history, cognitive science, artificial intelligence. Three fields that rarely appear in the same room, but when you line them up correctly, reveal something remarkable. A unified theory of expertise that applies equally to biological brains and silicon processors. Here is the promise I'm making to you right now. By the time this episode ends, you will have three things you didn't have before. A clear, mechanistic understanding of what expert intuition actually is, stripped of mythology, grounded in decades of Nobel Prize caliber research, a genuine understanding of how modern AI replicates it, including systems that currently outperform humans at specific deductive tasks, and a precise map of where that leaves us, which kinds of judgment we should trust AI to handle, and which kinds of judgment we should never let it replace. This is not a story about robots versus humans, it is a story about what knowing actually is, and it starts with a fog of war. Let's begin with the environment itself. Because to understand why that wargamer can do what he does, you have to understand the extraordinary cognitive pressure the environment creates, and that means understanding the history of fog of war simulation. The phrase fog of war was coined by the Prussian military theorist Karl von Klausowitz to describe the irreducible uncertainty facing every real battlefield commander. You never have complete information. Reports arrive late, get corrupted in transit, contradict each other. Your mental picture of the battlefield is always a guess, a probabilistic sketch of a reality that is moving and changing even as you observe it. The first person to systematically engineer that experience into a game was Georg von Reiswitz, whose formalized Kriegsspiel, German for war game, appeared in 1824. His key invention was the double-blind referee system. Two opposing commanders sat at entirely separate maps. Neither could observe the masterboard. They issued written orders to a neutral umpire who resolved the action on a central map and returned only what a real commander on that terrain would have been able to observe. The information that came back was fragmentary, delayed, sometimes wrong, exactly like actual war. This was not entertainment. The Prussian military made Kriegsspiel mandatory officer training, and the doctrinal agility it produced in Prussian commanders, the capacity to make confident decisions under genuine uncertainty is credited by historians as a genuine strategic advantage. The double-blind referee system is still the most authentic method we have for simulating command friction. But it is expensive. It requires dedicated space, multiple maps, and a skilled umpire. As wargaming commercialized through the 20th century, designers sought more elegant physical solutions. The block war game, pioneered by Columbia Games with their 1972 release, Quebec 1759, was a breakthrough in simplicity. Combat units became wooden blocks standing upright on the table. Your face of the block, the side showing unit identity, movement rating, and strength, faced only you. Your opponent stared at a wall of blank wood. That is it. No referee required. The uncertainty is baked into the physical object itself. Think about what that demands from a player. You are looking at your opponent's formations and you cannot read a single one. A cluster of eight blocks massing on the northern rail line. Is that a screening force of exhausted militia clinging to a position or three full-strength armored corps pre-staged for a breakthrough assault? The only way to know is to engage, and by then it may be too late to respond. So you reason, you examine everything else, the spacing between units, the timing of their movements, the historical doctrine of the force they represent. You construct a theory of what those blank faces are hiding. Alongside block games, designers developed the chit-pull activation system to address a different dimension of battlefield uncertainty, not who is where, but when they will act. In a standard turn-based game, each player moves all their units in a clean, predictable sequence. Nothing like real warfare. In a chit pull game, activation markers for different formations are drawn randomly from a cup. You might activate two of your core in succession, then your opponent gets a turn, then one of your units sits idle for two draws. The sequence is genuinely unpredictable. Plans made for ideal activation sequences shatter constantly. Contingency thinking becomes not just useful, but mandatory. And in the digital realm, designers have been able to implement fog of war with mathematical precision that tabletop mechanics can only gesture toward. Gary Grigsby's War in the East II, widely regarded as the benchmark simulation for the Eastern Front, uses a detection level algorithm that assigns every unit on the map a score from 0 to 10. Detection level 10 means your opponent has perfect intelligence, unit designation, exact combat value, fatigue, supply status, all of it. Detection level 0 means the unit does not appear on the map at all, it is invisible. And the score between those extremes is not static, it fluctuates continuously based on terrain concealment, how close your own supplied forces are, and whether you have invested in air reconnaissance missions. A unit sitting in open step is far easier to detect than one sheltering in marshland three hexes behind the front. Send your reconnaissance aircraft over an area and the detection level ticks upward, but only so far and only under specific conditions. The fog is not random noise, it is structured uncertainty, governed by explicit rules. And this brings us to the historically grounded dimension that gives the whole cognitive puzzle its texture. Operation Barbarossa, the German invasion of June 1941, was defined at every level by German logistical failure. Hitler had refused to fully mobilize the German war economy before the invasion began. The Wehrmacht confiscated civilian vehicles from occupied territories across Europe, but those vehicles were built for paved roads and they dissolved on the Soviet Union's largely unpaved network. Rain turned the roads to axle-deep mud. Rail conversion was painfully slow, Soviet lines ran on a different gauge than German stock. Partisan attacks targeted supply convoys with devastating regularity. For the wargamer, all of this is not just historical color, it is encoded in the simulation's mechanics. Soviet reserve armies have a limited number of strategic movement points available by rail. Rail junctions have explicit handling capacities. Certain terrain types are impassable to motorized units in mud season. These constraints define where a Soviet reserve army can physically be positioned. They are not an open field of possibility, they are a constrained solution space, and the veteran player who has internalized every one of those constraints for 40 years can look at a map, run the constraints in his head, and systematically eliminate the places a reserve army cannot be, until only a handful of viable locations remain. His tap on the map near Harkov is not a guess, it is the output of a constraint satisfaction process he runs in seconds, because he has run it thousands of times before. That is the first major insight. The fog of war is not random, it is structured, and structured uncertainty can be debuced by anyone or anything that has internalized the structure well enough. Now let's open that black box. The wargamer taps the map. People say intuition, but what is actually happening inside that brain? I want to start not with wargaming, but with chess, because chess is where the science of expert intuition was first rigorously cracked open. In the late 1940s, Dutch psychologist Adrian de Groot ran a series of experiments that should have changed how we talk about expertise forever, though the lesson took decades to fully absorb. De Groot showed chess positions to players at different skill levels and asked them to think aloud as they analyzed. The conventional assumption was that Grandmasters would demonstrate deeper, more extensive calculation, more moves ahead, more branches explored. That is not what he found. Grandmasters were not calculating dramatically more than strong intermediate players. What distinguished them was something different. They almost always identified the best move within their first few candidates considered. Their attention went immediately to the right area of the board. They did not need to search exhaustively because they were not searching at all in the way we usually picture it, they were recognizing. Herbert Simon, Nobel laureate, one of the foundational architects of both cognitive psychology and artificial intelligence as disciplines, extended this work with William Chase in the 1970s. They estimated that a chess grandmaster has internalized roughly 100,000 distinct board patterns over the course of their career. When a grandmaster looks at a position, their brain is not performing a sequential logical analysis. It is conducting an instantaneous parallel search through that enormous library. A pattern matches, a response fires. The experience from the inside is a flash of clarity, a sudden sense of, of course, it's there. What that flash actually represents is a retrieval event from a vast biological database. Simon drew the conclusion that applies far beyond chess. He wrote, intuition is nothing more and nothing less than recognition. That sentence is either a profound insight or a slight disappointment, depending on your relationship with the idea of human uniqueness. I think it is profoundly liberating because it means intuition is not a gift you either have or don't, it is a skill built through a specific process, and understanding that process tells you exactly how to cultivate it. Gary Klein took Simon's insight into the most extreme naturalistic environments he could find: fire commanders entering burning buildings, intensive care nurses reading patient monitors, naval officers tracking radar contacts in hostile waters, people making decisions in seconds under irreversible conditions with lives directly on the line. And Klein found exactly what Simon predicted. These experts were not deliberating. They were not listing options and weighing them against criteria, they were recognizing situations and acting on the recognized response, sometimes refining but rarely starting from analytical scratch. Klein formalized this as the recognition primed decision model, the RPD model, and it operates through three mechanisms I want you to hold clearly because they are going to map directly onto AI architecture in a few minutes. First, simple match. The situation presents a pattern the expert has encountered before in substantially the same form. The recognition is immediate, the response fires automatically, no deliberation, the library has the answer. Second, feature matching and story building. The situation is partially obscured, as in fog of war, or slightly novel. No single pattern fires cleanly, so the expert scans for individual cues, discrete, visible signals that carry meaning, and rather than analyzing them separately, the expert synthesizes them into a narrative, a story that explains the hidden present by linking the visible fragments into a coherent account. This story is not fiction, it is inference, structured, evidence-based inference running at speed. Third, mental simulation. Before committing to action based on the story, the expert subjects it to a forward projection test. They mentally run the proposed action forward in time. What happens next? What is the adversary's likely response? Where does the plan break down? If the simulation runs clean, they act. If something snags, if the story produces an implausible future state somewhere, they revise the story and run it again. Apply this directly to our war gamer at the Barbarossa map. He is not standing there in mystic communion with the game. He is running all three mechanisms, layered and fast. He reads the frontline spacing. There is a subtle gap in zone of control coverage that a screening force would not normally permit, 1st Q. He calculates in his head the rail capacity of the junctions southeast of Kharkov against the movement point cost of staging a guard's tank army, 2nd Q. He recalls that his current opponent, in three previous games this season, has used aggressive northern screening to mask a southern reserve, 3rd Q, from opponent modeling. The story assembles itself. Reserve staged behind the rail junction, southern axis, counter-offensive within three turns. He runs the simulation. Does a guard's tank counterattack through that corridor produce the supply disruption that would make strategic sense? Yes, the story holds. He taps the map. What felt like a thunderclap of intuition from the inside was actually a storybuilding process running at the speed of subconscious computation. The mysticism is a misreading of the experience. The mechanism is entirely legible. But here is the critical question: is this process trustworthy? Because we have all encountered people who are confidently, consistently, professionally wrong, who have decades of experience and intuitions that lead them straight into error. Experience clearly does not guarantee accuracy. So when does it? This is where two of the most important researchers in the field met in productive disagreement, and what they discovered when they compared notes is the most practically useful finding in this entire story. Daniel Kahneman and Gary Klein represent genuinely different schools of thought. Kahneman's career, which also earned him a Nobel Prize, is essentially a systematic inventory of how human intuition fails.

SPEAKER_00 17:54

The heuristics and biases literature he helped build is a catalog of the ways our pattern recognition machinery produces confident nonsense, availability errors, anchoring, overconfidence, the planning fallacy. For Kahneman, intuition is frequently a liability dressed up as an asset. Klein, coming from his naturalistic decision-making research with firefighters and military commanders, reached almost the opposite conclusion. Expert intuition in the right conditions is extraordinarily reliable. It is not a liability. It is the mechanism that lets experienced professionals make better decisions in two seconds than novices make in 20 minutes. Rather than argue past each other in separate papers, they did something unusual and genuinely admirable. They ran what scientists call an adversarial collaboration, a structured attempt to find the boundary conditions that would reconcile their opposing bodies of evidence. They published jointly in 2009. The paper's title is itself the finding Conditions for Intuitive Expertise, a failure to disagree. What they concluded is this expert intuition is valid and trustworthy when and only when two environmental conditions are satisfied. Condition one, the domain must be a high validity environment. It must have genuine, stable regularities, consistent causal structures, rules that hold, patterns that repeat in predictable ways. Chess is a high validity environment. Emergency medicine has enough regularity to support genuine expertise. Firefighting in building structures has valid patterns. What are low validity environments? Individual stock selection, where the causal relationships are swamped by noise and genuine randomness.

SPEAKER_01 19:51

Long-range geopolitical forecasting in genuinely unprecedented situations where the rules of the system are themselves shifting. In low validity environments, experience does not produce calibrated expertise. It produces overconfident noise. The pattern library fills with false patterns that feel just as compelling as real ones. You get people who have spent 30 years confidently predicting market movements and who are empirically no more accurate than someone with no experience at all. Condition 2. The expert must have adequate feedback loops. Prolonged exposure to the domain is necessary but not sufficient. The exposure must come with rapid, clear, unambiguous feedback that connects specific judgments to specific outcomes. If you make a prediction and discover whether it was right or wrong only years later, or only in indirect, filtered ways, your brain cannot update its pattern library effectively. You learn, but you learn slowly and noisily. True expertise requires feedback that is fast, honest, and directly connected to the original judgment. Wargaming satisfies both conditions so completely it is almost as if the game was designed as a proof of the Kahneman Klein framework. The rules are rigid, mathematical, exhaustively specified, hex grids, combat odds tables, supply chain mechanics, historical orders of battle. The causal relationships are stable and learnable. This is unambiguously a high validity environment. And at the conclusion of every single game, the fog lifts completely. Every hidden unit is revealed in its actual position. The wargamer sees not just whether they won or lost, but precisely where they were right and where they were wrong, where their deductions tracked the hidden reality and where they did not. Perfect feedback. Honest, immediate, direct. Over thousands of games across 40 years, the pattern library grows, the deductions sharpen, and the intuition becomes genuinely trustworthy because the environment has ruthlessly calibrated it against ground truth. What that veteran player has built, in the language of modern machine learning, is a probability estimator trained over thousands of supervised iterations with clean feedback. His biological neural network has done exactly what any well-trained artificial neural network does. He has learned the distribution of hidden states from extensive labeled experience, and that convergence between what biological expertise does and what machine learning does is not a metaphor, it is the actual mechanism, which is why AI can do it too. That is where part one ends. We have established the architecture, what the fog of war actually is, how expert intuition works at the mechanistic level, and why, in the right environment, with the right feedback, a biological brain and a trained neural network are doing the same fundamental thing. In part two, we go further, we look at how AI formally models hidden state problems through POMDPs and belief states. We examine AlphaStar, D Fog GAN, and the Stanford Hoover Wargame experiment. We look at what DARPA is building at the frontier, and we arrive at the final question: not whether AI can navigate the fog, but what that capability should cost us. The map is almost complete. One half remains unlit.