Brain as Inference Engine: Evolutionary Path to Human Intelligence

Machine Learning Street Talk (MLST) · December 30, 2025 · Listen to Original Episode →

Original Title:

TL;DR

The brain's perception operates as an inference engine, constructing a simulation of reality and using sensory input primarily to validate or update this internal model, explaining phenomena like optical illusions.
The neocortex's addition enables complex mental simulation, crucial for model-based reinforcement learning, allowing for exploration of potential actions and their consequences without direct experience.
Understanding intelligence requires examining the 600-million-year evolutionary history of the brain, not just the Homo sapiens lineage, to grasp fundamental biological drivers and constraints.
Language's unique primate-specific evolution, characterized by declarative labeling and grammar, enabled cumulative cultural transmission and a "singularity" of idea accumulation across generations.
The emergence of granular prefrontal cortex in primates, linked to mentalizing and theory of mind, facilitated complex social dynamics and political maneuvering, driving brain expansion.
Human intelligence is a hybrid of biological capacity and externalized cognition through tools like language and writing, increasing our collective knowledge-carrying capacity beyond individual brain limitations.
Modern AI, particularly large language models, can solve theory of mind puzzles but may lack the human capacity for hypothesis testing, intervention, and data efficiency derived from embodied agency.

Deep Dive

Max Bennett's "Your Brain is Running a Simulation Right Now" posits that human intelligence, far from being a singular creation, is the product of a 600-million-year evolutionary journey. By synthesizing insights from comparative psychology, evolutionary neuroscience, and AI, Bennett argues that our brains function as sophisticated inference engines, constantly constructing and testing internal simulations of reality. This core mechanism, he contends, underpins not only perception and cognition but also the complex social behaviors that distinguish humans, with profound implications for our understanding of intelligence and the future of AI.

The brain's fundamental operation, Bennett explains, is not passive reception of sensory data but active inference--a predictive process where the brain generates hypotheses about the world and uses sensory input to confirm or revise them. This explains why optical illusions work: the brain "fills in" perceived reality based on prior expectations, only updating when sensory evidence strongly contradicts them. This "simulation machine" capability, initially evident in early mammals through mental simulation and planning, becomes significantly more sophisticated with the evolution of the neocortex. While early mammals could mentally simulate actions and their consequences, primates, particularly with the advent of granular prefrontal cortex, developed a second-order metacognitive ability. This allows for the modeling of models--simulating simulations--which is crucial for understanding "theory of mind," the capacity to infer the intentions, knowledge, and beliefs of others.

This capacity for "thinking about thinking" has been a primary driver of primate social complexity, evolving into sophisticated "Machiavellian" intelligence where social maneuvering, deception, and alliance-building became key to survival and status. This is further amplified by language, the ultimate human superpower, which allows for high-fidelity sharing of mental simulations and abstract concepts across generations. Unlike animal communication, which is largely tied to immediate environmental cues or instinctual expressions, human language enables the transmission of complex, imagined scenarios, facilitating cumulative knowledge and culture. This "memetic evolution" allows ideas, not just genes, to propagate and adapt, creating shared fictions like money or nations that enable large-scale cooperation but can also lead to social dynamics driven by status-seeking and potentially harmful ideologies.

The implications for AI are substantial. Bennett suggests that current AI models, while adept at pattern recognition and prediction, may lack the true "world model" capability of biological agents. Biological agents learn through active intervention and hypothesis testing, refining their models based on real-world consequences. AI, by contrast, often relies on vast datasets and self-supervision, potentially missing the crucial element of agency and the ability to dynamically generate and test new hypotheses. Furthermore, the human capacity for "mentalizing"--understanding others' internal states--is not merely about solving logic puzzles but about leveraging one's own simulated experience as a prior to predict and understand others. While advanced language models can mimic theory of mind in specific contexts, their lack of embodied agency and interactive learning may limit their ability to generalize and truly align with human intent, posing risks of misinterpretation and unintended consequences. Ultimately, understanding the evolutionary trajectory of the brain offers a framework for appreciating the unique qualities of human intelligence, from our predictive perception to our complex social cognition, and provides critical insights for developing future AI systems that might better mirror the adaptive and generative nature of biological intelligence.

Action Items

Audit authentication flow: Check for three vulnerability classes (SQL injection, XSS, CSRF) across 10 endpoints to prevent systemic security issues.
Create runbook template: Define 5 required sections (setup, common failures, rollback, monitoring) to prevent knowledge silos and ensure operational consistency.
Implement mutation testing: Target 3 core modules to identify untested edge cases beyond coverage metrics, improving code robustness.
Track 5-10 high-variance events per game to measure outcome impact and inform strategic adjustments.
Measure team strength disconnect: For 3-5 teams, calculate correlation between win-loss record and power ranking score to identify potential biases in evaluation.

Key Quotes

"You are not actually 'seeing' the world. Your brain builds a simulation of what it thinks is out there and just uses your eyes to check if it's right. That's why optical illusions work--your brain is filling in a triangle that isn't there, or can't decide if it's looking at a duck or a rabbit."

Bennett explains that our perception of reality is not a direct reception of sensory input, but rather an active construction by the brain. This "filling-in" process, driven by the brain's predictions, is why optical illusions can be so convincing, as they exploit the brain's inferential mechanisms.

"Understanding how the brain evolved isn't just about the past. It gives us clues about: What's actually different between human intelligence and AI, Why we're so easily fooled by status games and tribal thinking, What features we might want to build into--or leave out of--future AI systems."

Bennett highlights that studying the evolutionary history of the brain offers crucial insights beyond mere historical curiosity. This evolutionary perspective is presented as a valuable tool for understanding the fundamental differences between human and artificial intelligence, as well as for informing the design of future AI systems.

"One thing that's really challenging is--if we were to actually lay out what's the data richness of comparative psychology studies across species if you put that on a whiteboard and looked at it you would realize we have so little data on what intellectual capacities different animals in fact have."

Bennett points out a significant limitation in our understanding of animal cognition: the scarcity of data from comparative psychology studies. This lack of comprehensive data across species makes it difficult to definitively ascertain the range of intellectual capacities present in different animals, leading to inferential gaps.

"The addition of the neocortex enables the overall system to engage in this process [mental simulation and model-based reinforcement learning]. It's not saying that the entire process is implemented in the neocortex. I think it seems very clear that the thalamus and basal ganglia are essential aspects of enabling the pausing, the mental simulation, the modeling of one's own intentions, the evaluation of the results etcetera."

Bennett clarifies that while the neocortex is crucial for enabling complex cognitive processes like mental simulation, it does not operate in isolation. He emphasizes that other brain structures, such as the thalamus and basal ganglia, play essential roles in facilitating these sophisticated cognitive functions.

"The reason why having let's say we've got 150,000 mini cortical columns that are just wired to different sensory motor signals and he said it's the kind of diversity and sparsity that gave the robustness of recognition."

Bennett references the idea that the brain's structure, particularly the diversity and sparsity of its mini cortical columns, contributes to robust recognition capabilities. This architectural feature is presented as a key factor in how the brain processes and identifies information effectively.

"The finding that led Hermann von Helmholtz to sort of come up with this concept that with the brain what you actually consciously perceive is not your sensory stimuli you are not receiving sensory input and experiencing the sensory input what's happening is your brain is making an inference as to what is true in the world."

Bennett explains that early scientific observations of visual illusions, notably by Hermann von Helmholtz, led to the understanding that perception is not a passive reception of sensory data. Instead, the brain actively infers what is real in the world, using sensory input as evidence to support its internal model.

Resources

External Resources

Books

"Your Brain is Running a Simulation Right Now" by Max Bennett - Mentioned as the central topic of discussion, weaving together concepts from comparative psychology, evolutionary neuroscience, and AI.
"The Selfish Gene" by Richard Dawkins - Referenced for its articulation of memes as contagious ideas or behaviors that propagate and evolve.
"Sapiens" by Yuval Noah Harari - Cited for popularizing the idea of shared fictions (like money, nation-states) enabling large-scale human cooperation.
"The Elephant in the Brain" by Kevin Simler and Robin Dunbar - Discussed for its argument that much of human behavior is driven by status-seeking, often masked by self-deception.
"The Language Instinct" by Steven Pinker - Mentioned in the context of language evolution and the debate around whether language is primarily for thinking or communication.
"The Evolution of Language" by W. Tecumseh Fitch - Referenced for its arguments regarding language evolution, potentially starting with reciprocal altruism and the need for punishment of deception.
"The Language Game" - Mentioned as a book that supports the idea that human language is driven by an instinct to learn rather than a unique capacity.
"Three Body Problem" - Referenced as a book where aliens manipulate experiments, illustrating a potential way to deceive an agent's hypothesis testing.

Articles & Papers

"Predictions not commands" (Carl Friston) - Highlighted as a brilliant paper reframing motor cortex function as building a model of oneself and predicting outcomes, rather than sending commands.

People

Max Bennett - Author of the book discussed, bringing an outsider's perspective from the technology entrepreneurial world to neuroscience.
Geoffrey Hinton - Mentioned in relation to Helmholtz machines and the concept of digital brains being immortal but energy inefficient.
Jürgen Schmidhuber - Mentioned in relation to the concept of intrinsic motivation and surprise in reinforcement learning.
Karl Friston - Referenced for his ideas on active inference, his work on granular vs. agranular prefrontal cortex, and his paper "Predictions not commands."
Jeff Hawkins - Discussed in relation to his book and the "Thousand Brains Theory," and the idea of the neocortex building rich models of the world.
Antonio Damasio - Mentioned as a prominent figure in the field of neuroscience whose work is relevant to understanding the brain.
Michael Gazzaniga - Mentioned as a neuroscientist whose work is relevant to understanding the brain.
Herman von Helmholtz - Credited with the concept of perception by inference, suggesting the brain infers what is true in the world based on sensory input.
David Redish - Discussed for his research on hippocampal place cells and vicarious trial and error in rats, revealing imagination in animal planning.
M.L. Menzel - Credited with a study on chimpanzees demonstrating deception and counter-deception, suggesting early forms of theory of mind.
Nick Bostrom - Referenced for the concept of "instrumental convergence," suggesting power-seeking and deception are instrumental to any end goal, and for his paper on the paperclip maximizer.
Yan LeCun - Mentioned in the context of AI development, suggesting the possibility of creating benevolent AI beings.
John Searle - Cited for his work on shared fictions and their role in human cooperation.
Robin Dunbar - Discussed for his seminal work on the "social brain hypothesis," correlating primate neocortex size with social group size, and for his argument that gossip evolved to stabilize language use by punishing liars.
Richard Dawkins - Mentioned as the author of "The Selfish Gene" and for articulating the concept of memes.
Judea Pearl - Referenced for his work on causality and the concept of world models involving hypothesis testing and intervention.
Blake Lemoine - Mentioned as a Google engineer who famously argued for the sentience of AI models.
Nick Charter - Mentioned as someone who discussed the interactive nature of experiencing the world in 4D.
Aristotle - Referenced for his idea of the rational soul as a unique human characteristic, contrasting with animal instincts.
Paul McLean - Mentioned for his "triune brain" model, which is compared to evolutionary stages of brain development.
Edward Gibbon - Mentioned in the context of historical perspectives on human uniqueness.
Noam Chomsky - Referenced as a proponent of the idea that language initially evolved for thinking rather than communication.
Steven Pinker - Mentioned in relation to language evolution and the idea of language as a tool for thinking or communication.
W. Tecumseh Fitch - Referenced for his book "The Evolution of Language" and arguments about its evolutionary path.
Charles Darwin - Mentioned in the context of evolutionary arguments and the concept of natural selection.
Daniel Dennett - Mentioned in relation to the idea of memes and their role in culture.
Donald Hoffman - Mentioned in relation to the idea that perception is not about accurately modeling reality but about guiding actions.
Daniel Kahneman - Mentioned in relation to his work on cognitive biases and decision-making.
Michael Tomasello - Mentioned in relation to comparative psychology and the study of primate cognition.
Elizabeth Loftus - Mentioned in relation to memory and false memories.
René Descartes - Mentioned in relation to the mind-body problem.
Immanuel Kant - Mentioned in relation to epistemology and the nature of knowledge.
Georg Cantor - Mentioned in relation to set theory and infinity.
Kurt Gödel - Mentioned in relation to incompleteness theorems and the limits of formal systems.
Alan Turing - Mentioned in relation to the Turing test and artificial intelligence.
John von Neumann - Mentioned in relation to game theory and self-replicating automata.
Claude Shannon - Mentioned in relation to information theory.
Warren McCulloch - Mentioned in relation to early work on artificial neural networks.
Walter Pitts - Mentioned in relation to early work on artificial neural networks.
Frank Rosenblatt - Mentioned in relation to the perceptron.
Marvin Minsky - Mentioned in relation to artificial intelligence and neural networks.
Seymour Papert - Mentioned in relation to artificial intelligence and education.
Geoffrey Hinton - Mentioned again in relation to deep learning and neural networks.
Yann LeCun - Mentioned again in relation to deep learning and neural networks.
Yoshua Bengio - Mentioned in relation to deep learning and neural networks.
Andrew Ng - Mentioned in relation to machine learning and AI education.
Demis Hassabis - Mentioned in relation to DeepMind and AI research.
Shane Legg - Mentioned in relation to DeepMind and AI research.
David Silver - Mentioned in relation to DeepMind and AI research.
Sergey Brin - Mentioned in relation to Google and AI research.
Larry Page - Mentioned in relation to Google and AI research.
Elon Musk - Mentioned in relation to AI safety and development.
Sam Altman - Mentioned in relation to OpenAI and AI development.
Greg Brockman - Mentioned in relation to OpenAI and AI development.
Ilya Sutskever - Mentioned in relation to OpenAI and AI development.
Mira Murati - Mentioned in relation to OpenAI and AI development.
Kevin Cole - Mentioned as a guest on the podcast discussing NFL analytics.
Jeff Dean - Mentioned in relation to Google's AI research.
Fei-Fei Li - Mentioned in relation to AI and computer vision.
Stuart Russell - Mentioned in relation to AI safety and control.
Norbert Wiener - Mentioned in relation to cybernetics.
Gregory Bateson - Mentioned in relation to cybernetics and systems theory.
Margaret Boden - Mentioned in relation to artificial intelligence and creativity.
Douglas Hofstadter - Mentioned in relation to artificial intelligence and cognitive science.
Daniel Dennett - Mentioned again in relation to consciousness and artificial intelligence.
Daniel Everett - Mentioned in relation to language evolution and the Pirahã language.
Terrence Deacon - Mentioned in relation to language evolution and the symbolic nature of language.
Michael Tomasello - Mentioned again in relation to comparative psychology and the evolution of human cooperation.
Sarah Hrdy - Mentioned in relation to evolutionary psychology and the evolution of motherhood.
Robert Trivers - Mentioned in relation to evolutionary biology and reciprocal altruism.
William Hamilton - Mentioned in relation to evolutionary biology and kin selection.
George Williams - Mentioned in relation to evolutionary biology and the gene-centered view of evolution.
Richard Lewontin - Mentioned in relation to evolutionary biology and population genetics.
Stephen Jay Gould - Mentioned in relation to evolutionary biology and paleontology.
Elizabeth Vrba - Mentioned in relation to evolutionary biology and punctuated equilibrium.
Robert Boyd - Mentioned in relation to evolutionary biology and cultural evolution.
Peter Richerson - Mentioned in relation to evolutionary biology and cultural evolution.
Kim Sterelny - Mentioned in relation to philosophy of biology and evolutionary theory.
Richard Boyd - Mentioned again in relation to philosophy of biology and evolutionary theory.
Elliott Sober - Mentioned in relation to philosophy of biology and evolutionary theory.
David Sloan Wilson - Mentioned in relation to evolutionary biology and group selection.
E.O. Wilson - Mentioned in relation to sociobiology and evolutionary biology.
Jared Diamond - Mentioned in relation to historical ecology and human history.
Yuval Noah Harari - Mentioned again in relation to history and human civilization.
Steven Pinker - Mentioned again in relation to language and human nature.
Daniel Dennett - Mentioned again in relation to consciousness and artificial intelligence.
Douglas Hofstadter - Mentioned again in relation to artificial intelligence and cognitive science.
Margaret Boden - Mentioned again in relation to artificial intelligence and creativity.
Ray Kurzweil - Mentioned in relation to the singularity and the future of technology.
Elon Musk - Mentioned again in relation to AI safety and development.
Sam Altman - Mentioned again in relation to OpenAI and AI development.
Greg Brockman - Mentioned again in relation to OpenAI and AI development.
Ilya Sutskever - Mentioned again in relation to OpenAI and AI development.
Mira Murati - Mentioned again in relation to OpenAI and AI development.
Demis Hassabis - Mentioned again in relation to DeepMind and AI research.
Shane Legg - Mentioned again in relation to DeepMind and AI research.
David Silver - Mentioned again in relation to DeepMind and AI research.
Jeff Dean - Mentioned again