Your Brain Doesn’t Command Your Body. It Predicts It. [Max Bennett]

Machine Learning Street Talk sits down with Max Bennett — entrepreneur and author of A Brief History of Intelligence — for an expansive discussion on how brains evolved across species and what those evolutionary patterns reveal about the design of modern AI systems. Bennett brings an outsider’s perspective to neuroscience, merging comparative psychology, evolutionary neuroscience, and machine learning research into a unified account of how intelligence scales from lampreys to primates.

The conversation explores specific evidence for increasingly sophisticated cognitive abilities across vertebrates: map-based spatial navigation in fish and reptiles, causal reasoning in great apes, and theory of mind in chimpanzees — including experiments showing chimps can infer what a researcher can and cannot see based on whether their goggles are opaque. Bennett discusses how these capabilities appear to have emerged in a specific evolutionary order and argues this sequence provides a blueprint for understanding what capabilities current AI models have and which they are still missing.

The episode also engages directly with AI safety concerns, examining Nick Bostrom’s instrumental convergence thesis — the idea that power-seeking and deception are instrumental to almost any goal — through the lens of primate Machiavellian intelligence. Bennett concludes that while these behaviors are natural evolutionary outcomes, they are not inevitable design choices for AI systems, and that developers have agency in deciding which features to instill. For researchers and practitioners interested in the cognitive science underpinning language model behavior, this is a dense and carefully reasoned reference conversation.

📺 Source: Machine Learning Street Talk · Published December 30, 2025
🏷️ Format: Interview