Jeff Hawkins: The Thousand Brains Theory of Intelligence

The Human Brain and Artificial Intelligence

Jeff Hawkins provides a fascinating exploration of his life's work, detailing the quest to reverse engineer the neocortex to understand human intelligence and develop truly intelligent machines. He argues that studying the brain is not separate from creating AI; rather, it is the most effective path forward.

The Neocortex and Intelligence

• The neocortex is the seat of high-level cognitive functions, making up approximately 75% of the human brain's volume.
• Hawkins posits a common cortical algorithm: every part of the neocortex functions on essentially identical principles, regardless of whether it processes vision, touch, or abstract mathematical concepts.
• Understanding this core circuit is the key to mastering both human cognition and machine intelligence.

The Thousand Brains Theory

"There isn't one model of a cup. There are thousands of models of this cup. There are thousands of models of your cell phone... It's a distributed modeling system."

• The brain uses reference frames—similar to spatial maps—to organize and understand concepts. Even abstract ideas are mapped to these spatial-like structures within the cortex.
• Intelligence is a distributed voting system. Thousands of cortical columns independently form models based on sensory input and movement, then vote to reach a consensus on what is being perceived.

Challenges to Conventional AI

• Current deep learning methods rely on "point neurons" and backpropagation, which Hawkins argues are biologically inaccurate and fundamentally limited.
• Sparse connectivity and synaptogenesis (the growth of new synapses) are highlighted as essential components that current AI architectures lack.
• Establishing sparse representations makes systems significantly more robust, addressing common issues like adversarial examples.

The Legacy of Humanity

Hawkins presents a bold perspective on the future: humanity's true legacy is knowledge, not biology. By creating intelligent machines that embody our understanding of the universe, we can preserve that knowledge far beyond the biological lifespan of the human species, potentially even carrying it to the stars.