Jeff Hawkins on The Thousand Brains Theory of Intelligence, Prediction, and the Evolutionary Origin of Consciousness

Jeff Hawkins introduces his "Thousand Brains Theory of Intelligence," proposing that the neocortex is not a single, monolithic intelligence but rather comprises approximately 150,000 independent, complete modeling systems, or "columns." Each column learns a model of the world, and these "little brains" communicate through a voting mechanism to achieve a unified perception. Intelligence, in this framework, is defined as the ability to learn and build an internal model of the world, representing its structure, objects, and their relationships.

A crucial distinction is made between the individual activity of these cortical columns and our conscious experience. We are only aware of the stable, consensual "voting outcome" of these thousands of brains, with the vast majority (95-98%) of underlying neural activity remaining unconscious. The learning process is emphasized as fundamentally reliant on movement and interaction, where sensory inputs are constantly compared against predictions generated by the internal model. Prediction is not merely an outcome but an inherent property and a primary mechanism for the model to test and correct itself, driving learning and adaptation.

While the podcast primarily delves into theoretical neuroscience, the emphasis on learning through movement and interaction offers an implicit insight into effective learning strategies. Actively engaging with an environment, manipulating objects, and exploring spaces are presented as fundamental to building robust internal models, whether learning a new house or a new app. The discussion on prediction as a core mechanism for model correction suggests that embracing surprise and errors is essential for refining one's understanding of the world.

The theory offers a compelling evolutionary origin story for intelligence, tracing it back to the need for intelligent movement in early organisms and the development of spatial mapping systems (place cells, grid cells). This specialized mapping mechanism was then generalized and replicated to form the flexible, universal learning algorithm of the neocortex. The conversation also touches upon the profound implications for artificial intelligence, suggesting that current deep learning approaches, while powerful, might benefit from incorporating principles of distributed, complete modeling systems and reference frames. Lex Friedman's opening thought experiment about preserving human knowledge for alien civilizations further broadens the scope, questioning what truly constitutes essential human experience and knowledge, ultimately pointing to consciousness and subjective experience as potentially unique.

"if human civilization were to destroy itself all of knowledge all our creations will go with us he proposes that we should think about how to save that knowledge in a way that long outlives us"

"the only unique thing we may have is consciousness itself and the actual subjective experience of suffering of happiness of hatred of love if we can record these experiences in the highest resolution directly from the human brain such that aliens will be able to replay them that is what we should store and send as a message not wikipedia but the extremes of conscious experiences the most important of which of course is love"

"there are synapses in my brain that have formed that reflect my knowledge of you and the model i have of you in the world"

"the neocortex it's a sheet of neural tissue it's about 75% of your brain it runs on this very repetitive algorithm it's a very repetitive circuit"

"the thousand brains theory the the big idea there if i had to summarize into one big idea is that we think of the the brain the neocortex is learning this model of the world but what we learned is actually there's tens of thousands of independent modeling systems going on"

"each what we call a column in the cortex is about 150,000 of them is a complete modeling system so it's a collective intelligence in your head in some sense"

"we are only consciously able to perceive the voting we're not able to perceive anything that goes on under the hood"

"intelligence is the ability to learn a model of the world so to build internal to your head a model that represents the structure of everything you know"