A Thousand Brains - A New Theory of Intelligence

Jeff Hawkins  ·  2021

2025-05-01  · 

 ·  top-100  ·  neuroscience   artificial-intelligence   human-intelligence

A Thousand Brains - A New Theory of Intelligence cover
Cover of A Thousand Brains - A New Theory of Intelligence

What if we’ve been thinking about the brain—and intelligence itself—all wrong? In A Thousand Brains, neuroscientist and tech entrepreneur Jeff Hawkins presents a new theory of intelligence rooted in how the brain actually works. It’s not just about neurons firing or layers of perception, but about how the brain models the world using reference frames—and what that means for building intelligent machines.

Here are some of the key ideas that reshaped my thinking.

The Neocortex: The Engine of Intelligence

The neocortex, sometimes called the “new brain,” is what sets mammals apart. It’s responsible for vision, language, music, and more. It comprises about 70% of the human brain, and yet, interestingly, it doesn’t connect directly to the body. To control movement or respond to stimuli, it has to communicate through older brain regions.

Despite handling vastly different functions—from recognizing faces to understanding speech—the structure of the neocortex is surprisingly uniform. What makes one region different from another is simply the type of data it receives. Vision, hearing, language—they’re all processed using the same kind of circuitry.

Modeling the World

A core function of the neocortex is maintaining a world model: a constantly evolving internal simulation of the world around us. This model helps us make predictions about what we’ll see, feel, or hear next. Most of the time, we don’t even notice it—until something surprises us. Then, the model gets updated.

But how does the brain actually represent this model?

Reference Frames: The Brain’s Secret Weapon

Hawkins introduces the idea that the brain uses reference frames—similar to coordinate systems—to represent knowledge. When you touch a cup, for instance, your brain needs to know not just what a cup is, but where your fingers are in relation to it. The neocortex builds these reference frames for objects and concepts alike, allowing us to predict what we’ll experience based on our position in space—or thought.

This approach isn’t limited to physical objects. Abstract ideas, like democracy or justice, are also stored in reference frames, though likely in higher-dimensional mental spaces. We move through these conceptual landscapes much like we move through physical space, which is why thinking often feels like navigating from one idea to the next.

Grid Cells and Location-Based Intelligence

Inspired by discoveries in neuroscience, Hawkins explains that the brain uses grid cells—a kind of internal GPS—to figure out where we are in a mental map. The brain essentially overlays grids on previously experienced environments or ideas, which helps determine where we are and what to expect next.

This location-based intelligence is powerful not just for spatial navigation, but also for learning and reasoning. It’s how we know what to expect when we walk around a familiar room—or reason through a complex argument.

A Critique of Traditional Theories

Hawkins argues against the long-held hierarchical theory of the brain, where perception flows upward in a series of increasingly abstract layers (like V1, V2, etc. in vision). He points out that this model doesn’t hold up in practice: our eyes are constantly moving, and senses like touch or sound unfold over time. Instead of strict hierarchies, he proposes a more dynamic, distributed model where multiple cortical columns create their own models and vote on the outcome.

What Machines Need to Be Truly Intelligent

So, what does all of this mean for artificial intelligence? According to Hawkins, we’ve been building AI systems that lack the core ingredients of real intelligence. Here’s what machines need:

  • Embodiment: True intelligence comes from interaction with the world. An AI must move, sense, and manipulate things to learn like we do.

  • A neocortex equivalent: To build robust, flexible intelligence, machines need something like the neocortex—capable of building reference frames and learning across multiple domains.

  • An “old brain” equivalent: Intelligence alone has no goals. For an AI to be useful or aligned, it needs motivational systems like those in the human brain that drive behavior.

Intelligence vs. Speed

Even if we one day create AI that operates a million times faster than the human brain, that doesn’t mean it will outpace us in everything. Physical constraints still apply. Building hardware, collecting data, and making scientific discoveries all take time. Speed alone doesn’t guarantee progress.

Final Thoughts

A Thousand Brains is more than just a theory of human intelligence—it’s a blueprint for how to build machines that can think more like we do. By grounding his ideas in neuroscience, Hawkins bridges the gap between biology and AI, challenging our assumptions along the way.

If you’re interested in how the brain really works—or where AI is headed next—this book is a must-read.