Jeff Hawkins (Numenta Founder) – The Thousand Brains Theory of Intelligence (Dec 2019)


Chapters

00:00:00 A Thousand Brains Theory and the Quest for Intelligent Machines
00:05:33 Neocortex: A Model of the World Inside Our Brain
00:08:59 Predictive and Structural Features of the Neocortex
00:12:57 Complexity of the Neocortex
00:17:37 Brain Columns through the Numenta Coffee Cup Experiment
00:21:28 Cortical Columns, Sensory Perception, and Grid Cells: Insights into Brain Function
00:31:28 Key Concepts of Jeff Hawkins' Thousand Brains Theory of Intelligence and Its Implications for AI

Abstract

Bridging the Chasm: Jeff Hawkins’ Thousand Brains Theory and the Future of AI

In a groundbreaking discussion, Jeff Hawkins introduces the “Thousand Brains Theory of Intelligence” as he expounds upon the structure and function of the brain’s neocortex and its implications for the future of artificial intelligence (AI). Drawing upon a wealth of neuroscience research, Hawkins delves into the complexity and adaptability of the neocortex, the primary organ responsible for human intelligence. He spotlights the limitations of current AI technology and offers a new conceptual framework for understanding brain function. The ultimate objective is to bridge the vast divide between the capabilities of human cognition and existing AI systems, driving a revolution in machine intelligence grounded in neuroscience.

The Neocortex: The Engine of Human Intelligence

Central to Hawkins’ thesis is the neocortex, a sheet of cells enveloping the brain. Governed by a hierarchical structure of “cortical columns,” the neocortex is responsible for sensory perception, motor behavior, and abstract thought. Far from being a simple input-output processor, its main function lies in constructing a highly distributed internal model of the world. This intricate model serves various purposes, including planning, goal-oriented behavior, and decision-making. Each cortical column has its own specialized “reference frame,” allowing the brain to learn the relational positions and functions of objects within its environment.

The Limitations of Contemporary AI

In contrast, today’s AI technologies are considerably limited in comparison to the human brain. Hawkins asserts that current AI systems are energy-intensive and lack the ability to learn rapidly. They struggle with the intricacies of flexibility and generalization, owing in part to a lack of a structured world model similar to that of the human neocortex. These limitations underscore the necessity for a different approach in the field of AIone that incorporates the findings from neuroscience to create more efficient and adaptable machine learning models.

Rethinking Brain Functionality

Hawkins finds inspiration in Francis Crick’s 1979 article, which lamented the absence of a conceptual framework for understanding brain function. In an attempt to fill this gap, Hawkins presents a reimagined way of thinking about brain functionality. It is predicated on the neocortex’s abilities to predict sensory experiences and to update its internal model when those predictions fail. Hawkins emphasizes the brain’s predictive nature, allowing for incremental learning rather than the need for mass data exposure, a feature yet to be efficiently replicated in AI.

Neocortical Columns and Sensory Perception

Zooming into the structure of the neocortex, Hawkins introduces the notion of “cortical columns,” roughly a millimeter in diameter, as the repetitive building blocks. With a remarkable similarity in architecture across different functionalities like vision, touch, and language, these columns are immensely complex systems. Their uniformity has led to speculation that if their functions can be decoded, it would significantly advance our understanding of the brain. Furthermore, these columns communicate with each other through “voting” to reach a common understanding of sensed objects. This hierarchical structure is based on the presence of specialized cells like grid cells and place cells, helping to create object-based reference frames.

The Thousand Brains Theory and Implications for AI

Hawkins posits the “Thousand Brains Theory,” suggesting that intelligence is not predicated on a single model but rather on thousands of complementary models housed in different cortical columns. These columns work in a hierarchical structure, voting to reach a consensus, thereby enabling structured learning. Hawkins sees this as the future of AI and believes that a technology mirroring these principles would dominate the 21st century. His company, Numenta, is already in the process of implementing these insights.

Conclusion

The intricacies of the human brain, particularly the functions and structures of the neocortex, offer invaluable lessons for the advancement of AI. Jeff Hawkins and his team at Numenta are on a quest to unlock these neuroscientific mysteries. Their goal is not just to gain a deeper understanding of human cognition but to apply this knowledge to bridge the current gap between human intelligence and machine learning. It’s an endeavor that holds the promise of a new era in AI, fueled by a nuanced understanding of the human brain.


Notes by: professor_practice