A Thousand Brains_ A New Theory of Intelligence

The cortical columns, in their world-modeling activities, work semiautonomously. What “we” perceive is a kind of democratic consensus from among them.

We human mammals are the victims of a recurrent dispute: a tussle between the old reptilian brain, which unconsciously runs the survival machine, and the mammalian neocortex sitting in a kind of driver’s seat atop it. This new mammalian brain—the cerebral cortex—thinks. It is the seat of consciousness. It is aware of past, present, and future, and it sends instructions to the old brain, which executes them.

The old brain, schooled by natural selection over millions of years when sugar was scarce and valuable for survival, says, “Cake. Want cake. Mmmm cake. Gimme.” The new brain, schooled by books and doctors over mere tens of years when sugar was over-plentiful, says, “No, no. Not cake. Mustn’t. Please don’t eat that cake.”

But if you ask neuroscientists, almost all of them would admit that we are still in the dark. We have learned a tremendous amount of knowledge and facts about the brain, but we have little understanding of how the whole thing works.

The neocortex occupies about 70 percent of the volume of a human brain and it is responsible for everything we associate with intelligence, from our senses of vision, touch, and hearing, to language in all its forms, to abstract thinking such as mathematics and philosophy.

We deduced that the neocortex stores everything we know, all our knowledge, using something called reference frames. I will explain this more fully later, but for now, consider a paper map as an analogy. A map is a type of model: a map of a town is a model of the town, and the grid lines, such as lines of latitude and longitude, are a type of reference frame.

We realized that the brain’s model of the world is built using maplike reference frames. Not one reference frame, but hundreds of thousands of them. Indeed, we now understand that most of the cells in your neocortex are dedicated to creating and manipulating reference frames, which the brain uses to plan and think.

How do our varied sensory inputs get united into a singular experience? What is happening when we think? How can two people reach different beliefs from the same observations? And why do we have a sense of self?

scientific papers are not well suited for explaining large-scale theories, especially in a way that a nonspecialist can understand.

False beliefs can be difficult to eliminate, and how false beliefs combined with our more primitive emotions are a threat to our long-term survival.

humans are defined by our genes, and the purpose of life is to replicate them. But we are now emerging from our purely biological past. We have become an intelligent species. We are the first species on Earth to know the size and age of the universe. We are the first species to know how the Earth evolved and how we came to be. We are the first species to develop tools that allow us to explore the universe and learn its secrets. From this point of view, humans are defined by our intelligence and our knowledge, not by our genes.

The neocortex can temporarily control breathing, as when you consciously decide to hold your breath. But if the brain stem detects that your body needs more oxygen, it will ignore the neocortex and take back control.

Neurons have treelike appendages called axons and dendrites that allow them to send information to each other.

Darwin proposed that the diversity of life is due to one basic algorithm. Mountcastle proposed that the diversity of intelligence is also due to one basic algorithm.

He said that the fundamental unit of the neocortex, the unit of intelligence, was a “cortical column.” Looking at the surface of the neocortex, a cortical column occupies about one square millimeter. It extends through the entire 2.5 mm thickness, giving it a volume of 2.5 cubic millimeters. By this definition, there are roughly 150,000 cortical columns stacked side by side in a human neocortex. You can imagine a cortical column is like a little piece of thin spaghetti. A human neocortex is like 150,000 short pieces of spaghetti stacked vertically next to each other.

Humans can do many things for which there was no evolutionary pressure. For example, our brains did not evolve to program computers or make ice cream—both are recent inventions. The fact that we can do these things tells us that the brain relies on a general-purpose method of learning. To me, this last argument is the most compelling. Being able to learn practically anything requires the brain to work on a universal principle.

In the end, our quest to understand the brain, our quest to understand intelligence, boils down to figuring out what a cortical column does and how it does it.

To make predictions, the brain has to learn what is normal—that is, what should be expected based on past experience. My previous book, On Intelligence, explored this idea of learning and prediction. In the book, I used the phrase “the memory prediction framework” to describe the overall idea, and I wrote about the implications of thinking about the brain this way. I argued that by studying how the neocortex makes predictions, we would be able to unravel how the neocortex works.

Neurons look like trees. They have branch-like extensions of the cell membrane, called axons and dendrites. The dendrite branches are clustered near the cell and collect the inputs. The axon is the output.

the brain has two types of neurons: neurons that fire when the brain is actually seeing something, and neurons that fire when the brain is predicting it will see something. To avoid hallucinating, the brain needs to keep its predictions separate from reality.

This observation means there must be neurons in the neocortex that represent the location of my finger in a reference frame that is attached to the cup. The movement-related signal we had been searching for, the signal we needed to predict the next input, was “location on the object.”

Reference frames were the missing ingredient, the key to unraveling the mystery of the neocortex and to understanding intelligence.

Recall that a reference frame is like the grid of a map. For example, on a paper map you might locate something using labeled rows and columns, such as row D and column 7. The rows and columns of a map are a reference frame for the area represented by the map. If an animal has a reference frame for its world, then as it explores it can note what it found at each location. When the animal wants to get someplace, such as a shelter, it can use the reference frame to figure out how to get there from its current location. Having a reference frame for your world is useful for survival.

The details of how place cells and grid cells work are complicated and still not completely understood, but you can think of them as creating a map of the environment occupied by the rat. Grid cells are like the rows and columns of a paper map, but overlaid on the animal’s environment. They allow the animal to know where it is, to predict where it will be when it moves, and to plan movements. For example, if I am at location B4 on a map and want to get to location D6, I can use the map’s grid to know that I have to go two squares to the right and two squares down.

Every cortical column learns models of objects. The columns do this using the same basic method that the old brain uses to learn models of environments.

Cortical grid cells in what columns attach reference frames to objects. Cortical grid cells in where columns attach reference frames to your body.

A well-known trick for remembering a list of items, known as the method of loci or sometimes the memory palace, is to imagine placing the items you want to remember at different locations in your house. To recall the list of items, you imagine walking through your house, which brings back the memory of each item one at a time.

However, if you are not trained in mathematics, then equations and other mathematical notations will appear as meaningless scribbles. You may even recognize an equation as one you have seen before, but without a reference frame, you will have no idea how to manipulate it to solve a problem. You can be lost in math space, in the same way you can be lost in the woods without a map.

This is one of the reasons why the binding problem is considered a mystery, but we have proposed an answer: columns vote. Your perception is the consensus the columns reach by voting.

To stop seizures, doctors will sometimes cut the connections between the left and right sides of the neocortex. After surgery, these patients act as if they have two brains. Experiments clearly show that the two sides of the brain have different thoughts and reach different conclusions. Column voting can explain why. The connections between the left and right neocortex are used for voting. When they are cut, there is no longer a way for the two sides to vote, so they reach independent conclusions.

The most important component of how brains learn continuously is the neuron. When a neuron learns a new pattern, it forms new synapses on one dendrite branch. The new synapses don’t affect previously learned ones on other branches. Thus, learning something new doesn’t force the neuron to forget or modify something it learned earlier. The artificial neurons used in today’s AI systems don’t have this ability. This is one reason they can’t learn continuously.

The unit of processing in the neocortex is the cortical column. Each column is a complete sensory-motor system—that is, it gets inputs and it can generate behaviors. With every movement, a column predicts what its next input will be. Prediction is how a column tests and updates its model.

Tags: #favorite

Once AI researchers understand the essential role of movement and reference frames for creating AGI, the separation between artificial intelligence and robotics will disappear completely.

In essence, this is the same criticism I am making, that AI needs reference frames. Hinton has proposed a solution to this problem that he calls “capsules.” Capsules promise dramatic improvements in neural networks, but so far they have not caught on in mainstream applications of AI. Whether capsules succeed or whether future AI relies on grid-cell-like mechanisms as I have proposed remains to be seen. Either way, intelligence requires reference frames.

“Qualia” is the name for how sensory inputs are perceived, how they feel. Qualia are puzzling. Given that all sensations are created by identical spikes, why does seeing feel different than touching?

Do we have a moral obligation to not turn off a conscious machine? Would that be equivalent to murder? No. I would have no concerns about unplugging a conscious machine. First, consider that we humans turn off every night when we go to sleep. We turn on again when we wake. That, in my mind, is no different than unplugging a conscious machine and plugging it in again later.

Our fear of death is created by the older parts of our brain. If we detect a life-threatening situation, then the old brain creates the sensation of fear and we start acting in more reflexive ways. When we lose someone close to us, we mourn and feel sad. Fears and emotions are created by neurons in the old brain when they release hormones and other chemicals into the body. The neocortex may help the old brain decide when to release these chemicals, but without the old brain we would not sense fear or sadness.

Older parts of the human brain control the basic functions of life. They create our emotions, our desires to survive and procreate, and our innate behaviors.

The emotions that direct our behaviors are determined by the old brain. If one human’s old brain is aggressive, then it will use the model in the neocortex to better execute aggressive behavior. If another person’s old brain is benevolent, then it will use the model in the neocortex to better achieve its benevolent goals.

Finally, an intelligent machine must have goals and motivations. Human goals and motivations are complex. Some are driven by our genes, such as the desire for sex, food, and shelter.

To endow a machine with goals and motivations requires that we design specific mechanisms for goals and motivations and then embed them into the embodiment of the machine. The goals could be fixed, like our genetically determined desire to eat, or they could be learned, like our societally determined goals for how to live a good life.

Neurons take at least five milliseconds to do anything useful. Transistors made of silicon can operate almost a million times faster. Thus, a neocortex made of silicon could potentially think and learn a million times faster than a human. It is hard to imagine what such a dramatic improvement in speed of thought would lead to. But before we let our imaginations run wild, I need to point out that just because part of an intelligent machine can operate a million times faster than a biological brain doesn’t mean the entire intelligent machine can run a million times faster, or that knowledge can be acquired at that speed.

Intelligent machines do not have the same constraints related to wiring. For example, in the software models of the neocortex that my team creates, we can instantly establish connections between any two sets of neurons. Unlike the physical wiring in the brain, software allows all possible connections to be formed. This flexibility in connectivity could be one of the greatest advantages of machine intelligence over biological intelligence

Another way that machine intelligence will differ from human intelligence is the ability to clone intelligent machines. Every human has to learn a model of the world from scratch. We start life knowing almost nothing and spend several decades learning. We go to school to learn, we read books to learn, and of course we learn via our personal experiences. Intelligent machines will also have to learn a model of the world. However, unlike humans, at any time we can make a copy of an intelligent machine, cloning it.

Concerns about the existential risks of AI, on the other hand, are qualitatively different. It is one thing for bad people to use intelligent machines to do bad things; it is something else if the intelligent machines are themselves bad actors and decide on their own to wipe out humanity.

Although we can make intelligent machines that run a million times faster than a biological brain, they cannot acquire new knowledge a million times faster.

The intelligence explosion adherents sometimes talk about “superhuman intelligence,” which is when machines surpass human performance in every way and on every task. Think about what that implies. A superhuman intelligent machine could expertly fly every type of airplane, operate every type of machine, and write software in every programming language. It would speak every language, know the history of every culture in the world, and understand the architecture in every city. The list of things that humans can do collectively is so large that no machine can surpass human performance in every field.

Intelligence is created through thousands of small models of the world, where each model uses reference frames to store knowledge and create behaviors.

Endowing machines with the ability to learn a model of the world and thus acquire knowledge and skills.