The Thousand Brains Theory of Intelligence and the Future of AI

When Charles Darwin worked out the theory of evolution in the nineteenth century, he didn’t have all the details in which the theory would ultimately depend. After all, On the Origin of Species was published in 1859—a full 49 years before the term genetics was introduced and 94 years before the discovery of the double-helix structure of DNA.

But while Darwin didn’t know the details of genetic inheritance, he was able to deduce, through countless observations, that all plant and animal life—including all of its apparent diversity—is a manifestation of a single mechanism or algorithm: evolution by natural selection. He would leave the details of the mechanism to be filled in by future scientists.

Major scientific breakthroughs often follow a similar pattern. The details of planetary orbits, for example, are highly complex, but the discovery of heliocentrism provided a simplified framework in which future scientists could work out the complex mathematics. 

In the contemporary world, our understanding of intelligence is in need of a similar breakthrough. While we know a lot of details about the workings of the brain—just as Darwin knew a lot of details about different forms of life before he developed the theory of evolution—we are in need of a single mechanism or algorithm that can explain the diverse functions of the human neocortex, the seat of human intelligence in the brain.

Jeff Hawkins, in his latest book A Thousand Brains, proposes just such a theory. The Thousand Brains Theory of Intelligence, as Hawkins calls it, has the potential to do for our understanding of intelligence what the theory of evolution did for our understanding of biology, and what heliocentrism did for our understanding of planetary orbits. 

Let’s see what the theory entails, and why it might not only change our understanding of the brain but also fundamentally alter the approach of artificial intelligence (AI) research. 

First, consider that the neocortex is, to the naked eye, an undifferentiated mass of tissue. We know that certain regions perform certain functions (vision, hearing, higher-order thinking), but this isn’t obvious by inspecting the organ—it simply all looks the same. 

Zooming in to the level of the neuron doesn’t offer much help. All parts of the neocortex contain neurons arranged in what are called “cortical columns,” or vertical strips about the size of a strand of spaghetti. The neocortex is essentially composed of about 150,000 of these cortical columns that process information vertically and communicate across columns horizontally. These cortical columns all have the same component parts (with minor differences), but produce very different functions. 

If you think about it, the problem of explaining intelligence appears to be analogous to the problem encountered by Darwin. Faced with the apparent diversity of life, Darwin proposed a single mechanism responsible for explaining all of the variety (natural selection). In the same way, is it possible that the diversity of intelligence can also be explained by a single mechanism? The homogeneity of nervous tissue and cortical columns seems to suggest that the answer is yes. 

The first scientist to propose that this common mechanism must exist was the neurophysiologist Vernon Mountcastle in the late 1970s. As Hawkins writes:

“Mountcastle proposed that the reason the [neocortex] regions look similar is that they are all doing the same thing. What makes them different is not their intrinsic function but what they are connected to. If you connect a cortical region to eyes, you get vision; if you connect the same cortical region to ears, you get hearing; and if you connect regions to other regions, you get higher thought, such as language.”

All cortical regions, Mountcastle hypothesized, are doing the same thing; the problem is, Mountcastle didn’t know what that same thing was. What Hawkins is proposing is that not only was Mountcastle right, but the latest research is starting to piece together the specifics of this underlying algorithm of intelligence. 

The key to understanding the diversity of intelligence, it turns out, is to think of the brain not as a single unit that processes information in an input-output manner similar to the way computers are currently built, but to think of the brain as being composed of thousands of smaller brains (cortical columns) that each develop independent predictive models of the world and communicate that information across columns in a way that creates a unified experience of perception.   

Each one of your 150,000 cortical columns builds models of the world using what Hawkins calls reference frames. When you look out your window at passing cars, for example, the reason you see the cars as occupying a location that is at a distance from you—and not in your eyes where visual perception is actually taking place—is because your cortical columns contain both place cells (to decipher what it is you’re looking at) in addition to grid cells (to decipher the location of objects in the environment relative to your body). Your brain attaches similar reference frames to all objects in the world and even to abstract concepts like democracy and mathematics, according to Hawkins.

The act of thinking is the process of moving through reference frames, and the reason memory-enhancement techniques like the “memory palace” work so well is because knowledge is stored in the brain spatially. Using the memory palace technique, you would create an imaginary location, such as your house, and proceed to “store” important information in each room. As you move through the house in your imagnization, you are able to better recall the information than if you tried to simply memorize the list, and that is because the technique is taking advantage of the way your brain naturally stores information. 

The Thousand Brains Theory of Intelligence, then, says that your brain is composed of 150,000 cortical columns that build models of the world using reference frames and that thought is a form of movement through information and ideas stored in relation to these spatial frames. Rather than simply processing information, your brain is creating thousands of independent models of the world, using all available senses, and integrating them together into a cohesive whole.  

If Hawkins is right, this new way to think about intelligence will open the door to a new approach to the creation of artificial intelligence, which currently lacks true intelligence in the sense of the mental flexibility we see in humans. Future AI research, according to Hawkins, will need to replace neural networks with reference frames if we are to ever truly achieve general AI.

So is Hawkins right? As he admits, there is still much we do not know about the brain, and further research will be needed to confirm the Thousand Brains Theory. Nevertheless, Hawkins is confident that, while the details of the theory will inevitably be altered as new information is discovered, the general outline of the theory will hold up to scrutiny as the only explanation that can fully explain intelligence. 

In the second part of the book, Hawkins tackles the future of AI and the prospect of machines attaining consciousness. Hawkins convincingly demonstrates that current AI is not really intelligent at all, and that the prospect of creating truly intelligent machines will depend on our ability to replicate cortical columns (which we can’t currently do). Hawkins also dismisses the exaggerated fears of AI as an existential threat, noting that motivations must be programmed into the machines and therefore there is no reason to assume inherent malevolent intent. Like Steven Pinker and others, Hawkins is confident that robust safety protocols—which are core components of the engineering profession—will prevent AI from getting out of control.  

It is when Hawkins turns to the topic of consciousness that he is least persuasive; in fact, I don’t think it’s an exaggeration to say that the chapter on consciousness is embarrassingly simplistic and philosophically naive. He presents no counterarguments to his position, discusses none of the philosophical literature relevant to the topic, dismisses the hard problem of consciousness outright and without argument, and confidently proclaims “I have no doubts that machines that work on the same principles as the brain will be conscious.”

It’s interesting to witness Hawkins display the appropriate humility when discussing brain science—a topic he is thoroughly familiar with—yet when the topic turns to something he clearly knows very little about, consciousness, he becomes dogmatic and self-assured. This is a common phenomenon you see in contemporary scientists living in a reductionist age. If science can’t currently (or even conceivably) solve the problem, it’s better to just pretend it doesn’t exist. 

Hawkins’ very own theory demonstrates the hard problem of consciousness. Remember, Hawkins is claiming that the brain is composed of thousands of cortical columns, which are all essentially doing the same thing. Each column is composed of neurons and electro-chemical activity. If this is the case, how is it that some columns create the perception of color—the actual experience of seeing red, for example—while some columns create the perception of sound, if every column is doing the same thing physically? Hawkins offers no explanations for how the same swirl of electricity in every cortical column could possibly produce different forms of conscious perception, or even how nerve impulses and chemicals traveling across synapses can result in actual three-dimensional experience. He wants the reader to simply pretend the problem doesn’t exist and to take his word for it that there’s nothing to explain. He even at one point shifts the burden of proof onto the skeptical reader. Hawkins writes:

“If you believe that consciousness cannot be explained by scientific investigation and the known laws of physics, then you might argue that I have shown that storing and recalling the states of a brain is necessary, but I have not proven that is sufficient. If you take this view, then the burden is on you to show why it is not sufficient.”

We currently have no satisfactory scientific explanation of consciousness. Further, we have not, to date, been able to create consciousness in machines. Therefore, if Hawkins is making the case that we can create consciousness in machines, you would think the burden of proof would be on him to explain how consciousness arises in the brain (winning him a Nobel prize in the process) and how exactly this could be replicated in machines. But he can’t, so he just assumes that artificial consciousness is possible and then tells his skeptics to prove him wrong. Of course, if it’s not possible to create consciousness in machines, this will be very difficult to prove, other than continuing to not be able to do it. 

Hawkins’ shallow treatment of the philosophy is actually unfortunate, as it tarnishes what would otherwise have been a fascinating and thought-provoking book. It seems that Hawkins might be on the right track to explaining intelligence and providing the groundwork for more effective AI, but the reader should not mistake the Thousand Brains Theory for an all-encompassing theory of consciousness and the brain, because it is not.  


A Thousand Brains: A New Theory of Intelligence is available on Amazon.com