Amsterdam – What if the human brain does not build one single model of the world, but many models at the same time?

That is one of the central questions behind A Thousand Brains: A New Theory of Intelligence, the 2021 book by computer pioneer and neuroscientist Jeff Hawkins. The book presents Hawkins’ Thousand Brains Theory of Intelligence, a theoretical framework that attempts to explain how the neocortex might learn, represent and recognize objects and concepts.

The idea is particularly interesting for people following artificial intelligence. Hawkins argues that understanding how biological intelligence works could provide clues for developing different kinds of machine intelligence. However, it is important to distinguish between what neuroscience has established and what the Thousand Brains Theory proposes.

The theory is not a universally accepted explanation of how the human brain works. The underlying research is a proposed theoretical framework, and several of its predictions remain subjects for further experimental investigation.

So what exactly is the Thousand Brains Theory, and why does it matter to the future of AI?

Table of Contents

• What Is A Thousand Brains About?
• Why the Neocortex Matters
• What Does “Thousand Brains” Mean?
• Reference Frames and How We Represent the World
• The Role of Grid Cells
• Why Movement Matters
• Learning, Memory and Intelligence
• What Does This Have to Do With AI?
• What the Theory Does Not Prove
• Why This Idea Is Interesting for AI and Expats
• Key Takeaways
• Dutch Learning Corner
• Community Discussion

What Is A Thousand Brains About?

A Thousand Brains: A New Theory of Intelligence was published in 2021 by Basic Books. Jeff Hawkins, who previously co-founded Palm and Handspring and later co-founded the neuroscience research company Numenta, uses the book to present his ideas about the biological basis of intelligence and their possible implications for artificial intelligence.

The book is divided broadly into three areas: the brain, machine intelligence and the future of human intelligence.

Rather than presenting a conventional introduction to neuroscience, Hawkins uses the book to argue for a particular way of thinking about the neocortex. His central proposal is that intelligence may depend on many similar cortical structures learning models of the world in parallel.

This distinction is important. Hawkins is not simply summarizing an established scientific consensus. He is presenting a theory that attempts to explain several observations about the neocortex within one framework.

The research behind the idea was also published in the peer-reviewed journal Frontiers in Neural Circuits. The 2018/2019 research paper proposed a framework based partly on grid-cell-like mechanisms and introduced the hypothesis that different parts of the neocortex could learn complete models of objects.

Why the Neocortex Matters

The neocortex is the outer layer of the mammalian brain and is associated with many higher-level functions, including perception, language and aspects of thought.

Hawkins’ theory focuses heavily on the neocortex because he argues that many apparently different forms of intelligence may be produced by similar underlying computational principles.

In his framework, seeing an object, recognizing an object through touch, understanding a concept or learning how something behaves may involve related mechanisms for representing information and relationships.

This is one reason the theory is relevant to AI research. If apparently different intelligent abilities can be explained by a common computational principle, researchers might theoretically be able to build machines using a similarly general architecture.

That is a proposal rather than a demonstrated fact. Neuroscience has not reached a universally accepted single theory explaining exactly how the neocortex produces intelligence.

What Does “Thousand Brains” Mean?

The name can sound as though Hawkins is claiming that humans literally have a thousand separate brains. That is not what the theory means.

The term refers to the idea that many cortical regions could independently learn models of objects and concepts.

According to the Thousand Brains Theory, instead of one central representation of an object being constructed somewhere at the top of a processing hierarchy, multiple cortical columns may learn models of the same object in parallel.

Imagine holding a coffee cup.

Your brain receives information through several sensory channels. You can see the cup, touch its surface, feel its weight and move your hand around it. Different sensory experiences provide different information about the same object.

Hawkins’ framework proposes that different cortical regions can learn models from these experiences and that these models can work together to produce a consistent perception.

Numenta describes the theory as proposing that every part of the neocortex can learn complete models of objects and concepts. The important word here is proposes.

This is the central hypothesis of the Thousand Brains Theory, not an established description that all neuroscientists agree completely explains how the human brain operates.

Reference Frames and How We Represent the World

One of the most important ideas in Hawkins’ framework is the concept of reference frames.

A reference frame can be thought of as a system for representing the relationships between different elements of something.

Consider a simple object such as a coffee mug. Knowing that a handle exists is useful, but knowing where the handle is relative to the rest of the mug provides additional information.

As you move your hand around the mug, your brain receives changing sensory information. Hawkins’ theory proposes that the neocortex could use location-related representations to build a structured model of the object.

This approach is important because intelligence is not only about identifying things. It is also about understanding relationships.

A person does not merely recognize a chair as “chair.” They can understand where the seat is, where the back is, how the chair might move and how their body can interact with it.

The Thousand Brains framework attempts to explain how representations of objects and their relationships could emerge from this type of sensorimotor interaction.

However, extending this framework from physical objects to abstract concepts such as mathematics, language and complex reasoning involves additional theoretical assumptions. These areas should therefore not be presented as experimentally solved by the theory.

The Role of Grid Cells

Grid cells are real biological neurons that have been studied extensively in neuroscience, particularly in relation to spatial representation and navigation.

They were originally identified in the entorhinal cortex and are associated with representing an organism’s location within an environment.

Hawkins and his colleagues built part of their theoretical framework around the idea that mechanisms similar to grid-cell representations could also play a role in the neocortex.

Their 2018 research paper proposed that grid-cell-like mechanisms could provide a location-based framework for representing objects and concepts.

This is one of the most interesting and also one of the most important distinctions to understand:

Grid cells themselves are an established neuroscience finding. The broader claim that similar mechanisms operate throughout the neocortex as proposed by the Thousand Brains framework is a theoretical hypothesis.

In other words, the existence of grid cells does not by itself prove the entire Thousand Brains Theory.

Why Movement Matters

Another important part of Hawkins’ framework is movement.

Imagine recognizing a familiar object while your eyes are closed. You may be able to identify it by moving your fingers across its surface.

The information you receive changes as your hand moves. Your brain can combine those different observations to build a more complete representation.

The Thousand Brains framework proposes that this type of sensorimotor learning is fundamental to how cortical representations are formed.

This perspective is particularly relevant to robotics and AI.

Many AI systems learn from enormous quantities of static digital information. A physical organism, however, can interact with its environment. It can move, touch, observe consequences and update its internal model.

Hawkins’ research therefore suggests that movement and interaction may be important ingredients in building more general forms of machine intelligence.

That does not mean current AI systems cannot learn without physical movement. Nor does it establish that robots must use the exact mechanisms proposed by Hawkins. It is better understood as a research direction inspired by a theory of biological intelligence.

Learning, Memory and Intelligence

One of the broader ideas in A Thousand Brains is that intelligence cannot be separated easily from learning and memory.

To recognize something, an intelligent system needs some form of stored knowledge about what it has previously encountered.

For humans, learning changes the internal representations used to interpret future experiences. A familiar object can be recognized more quickly because the brain has already built knowledge about it.

Hawkins’ framework places significant emphasis on the idea that the neocortex continuously learns models of the world.

This creates an interesting contrast with a simplistic view of intelligence as merely calculating answers.

From this perspective, intelligence involves building internal models, using those models to interpret sensory information and applying previous knowledge to new situations.

It is important, however, not to reduce the scientific relationship between memory and intelligence to the statement that “intelligence is simply memory.” Human cognition involves many interacting biological systems and processes, and the Thousand Brains Theory is only one theoretical attempt to explain part of this picture.

What Does This Have to Do With AI?

This is where the book becomes especially relevant to today’s technology discussions.

Hawkins argues that studying the brain could provide useful principles for designing more general machine intelligence.

The idea is not necessarily to copy the brain neuron by neuron. Instead, researchers could try to identify computational principles that appear important for biological intelligence and implement those principles in machines.

The Thousand Brains framework suggests several areas that could be relevant to AI research:

• Multiple models: Different parts of a system could maintain complementary models of the same object or concept.
• Sensorimotor learning: Intelligent systems could learn through interaction with their environment.
• Reference frames: Systems could represent relationships and locations rather than relying only on pattern matching.
• Continuous learning: Intelligence could involve continually updating internal models as new experiences arrive.
• General-purpose representations: Similar computational principles could potentially be applied to different types of knowledge.

These ideas are particularly interesting at a time when AI research is exploring increasingly capable systems that combine language, vision, reasoning, memory and interaction.

But the theory should not be presented as a proven blueprint for AGI. There is currently no established evidence that implementing the Thousand Brains framework would automatically produce artificial general intelligence.

What the Theory Does Not Prove

This distinction is essential if we want to discuss Hawkins’ ideas responsibly.

The Thousand Brains Theory is a proposed framework for understanding intelligence. Even Numenta’s own research materials describe it in terms of hypotheses and proposed mechanisms and state that experimental work is needed to validate aspects of the framework.

Several conclusions should therefore be avoided:

• It has not been established that the entire human neocortex operates exactly according to the Thousand Brains Theory.
• The existence of grid cells does not prove that the same mechanism operates throughout the neocortex in the way proposed.
• The theory does not prove that current AI systems are unintelligent.
• The theory does not prove that future AGI systems must reproduce the human brain.
• The framework does not provide a complete scientific explanation of human consciousness.

These limitations do not make the theory irrelevant. They simply define what we can reasonably claim about it.

A scientific theory can be interesting precisely because it generates testable ideas. Hawkins and his colleagues have proposed mechanisms that can, at least in principle, be investigated experimentally.

Why This Idea Is Interesting for AI and Expats

For people living in the Netherlands and working in technology, AI is becoming increasingly difficult to ignore.

AI is changing how software is developed, how companies market products, how customer service operates and how people search for information.

Understanding different theories of intelligence can therefore be useful even if you are not a neuroscientist.

The Thousand Brains Theory provides one way to think about an important question:

Is intelligence mainly about processing more information, or does it depend on building structured models of the world and learning through interaction?

There is no simple scientific answer that settles this question today.

But that is precisely what makes the topic interesting. The future of AI may not be determined by one single theory or architecture. Researchers are exploring multiple approaches, including large neural networks, reinforcement learning, robotics, neuroscience-inspired computing and other forms of machine learning.

Hawkins’ work represents one of the neuroscience-inspired approaches in that broader landscape.

Key Takeaways

• The book: A Thousand Brains: A New Theory of Intelligence was published by Jeff Hawkins in 2021.
• The central idea: Hawkins proposes that different parts of the neocortex may learn models of objects and concepts in parallel.
• Reference frames: They are a central concept in the theory for representing relationships and locations.
• Grid cells: These are established neuroscience findings, while their proposed role throughout the neocortex is part of Hawkins’ theoretical framework.
• Movement: The theory places significant emphasis on sensorimotor interaction and learning through movement.
• AI: Hawkins argues that neuroscience may provide useful principles for future machine intelligence.
• Important limitation: The Thousand Brains Theory remains a proposed theoretical framework, not a universally accepted explanation of how the human brain works.

Dutch Learning Corner

Could Brain Science Change AI?

Jeff Hawkins’ Thousand Brains Theory raises an interesting question: could understanding how biological intelligence works help researchers build better AI?

What do you think? Could neuroscience-inspired approaches become more important as AI systems become more capable?

Sources

Primary sources: Jeff Hawkins / Numenta research materials;
A Framework for Intelligence and Cortical Function Based on Grid Cells in the Neocortex, published in Frontiers in Neural Circuits;
Numenta’s Thousand Brains Theory research materials;
Basic Books publication information for A Thousand Brains.

This article distinguishes established neuroscience findings from hypotheses proposed by the Thousand Brains Theory. It is an educational overview and should not be interpreted as a statement that the theory represents a scientific consensus.