In the twentieth century, science came to understand that one’s experience of the outside world is an illusion.
This is obviously so when a person dreams during sleep. In a dream, there is a rich conscious experience that is disconnected from outside reality.

When a person is awake, the situation is different, but only partially so. An awake person, in a sense, dreams of a world that has a close correspondence with the world he is in. The light entering his eyes, his sense of touch and smell, etc. all provide information that his brain uses to make the dream correspond to the outside world. The correspondence is close enough that the person not only believes he is in that world, but is able to act in the outside world in a way consistent with survival. When dreaming in sleep, a person also believes he is in the real world, but the paralysis of sleep prevents him from harming himself. An awake person can cross the street, or communicate with another person, because his dream world corresponds closely enough to the outside physical world for him to perform these actions reliably.

Neuroscience is supposed to be the study of how a physical object, the brain, can perform this remarkable trick. However, there is a barrier that prevents neuroscience from focusing on its primary goal. To begin to see this barrier, consider the following.

Whether a person dreams when sleeping, or when awake, there is a world in his head. This inner world must be a simulation, a model. The simulation of the world must be an approximation — there is so much stuff in the outside world that there could never be a perfect copy of it represented in a limited physical space inside the skull. However, the degree of perfection of the model world inside the brain is in some sense irrelevant. The important thing is, if there is a model world in the head, then there must be a person inside the head to experience, to be conscious of, to dream about this model world. This person (and he has a name, “The Homunculus”) has the same relation to the model world in the brain that the real person has to the outside world. If a person needs a brain to experience the outside world, then the Homunculus must need a brain to experience the model world. The more perfect the inner model world, the more obviously is this so. In the extreme that the human brain could have a perfect simulation of the outside world, then the position of the Homunculus comes to be a perfect simulation of that of the person in the real world.

Thus, the great insight of the twentieth century, that the world we experience is in fact in our head, tells us very little. It does not explain how we experience that world. If we need a Homunculus to be conscious of the world in our brain, then the Homunculus must have a brain of his own in order to be conscious of that model world, and in its head there must be another Homunculus, and the series must go on infinitely, yet it would still not explain anything. Thus we can conclude that the Homunculus does not exist, that there must be something else going on. Whenever we try to figure out what that might be, we return to find the Homunculus grinning at us. The Homunculus is an impossible being, but he is also the person we meet whatever path we take to try to understand consciousness. Since the Homunculus cannot exist, but he blocks us nevertheless, I refer to the Homunculus problem or barrier. While the Homunculus cannot exist, a barrier to our understand very evidently does exist.

The Homunculus problem represents a barrier to our understanding. It is a barrier so high and impenetrable that we have not the slightest idea of what is on the other side. In the entire accumulation of human knowledge, there does not seem to be a single idea or piece of data that offers a passage through this barrier. This is a remarkable situation. It is so remarkable that few people discuss it.

Neuroscientists therefore have a problem. They can study the brain, but they cannot study its most interesting properties. Most neuroscientists take a practical approach. They confess that they cannot contend with the problem of the Homunculus, and instead study what they are able to study. The brain provides an unlimited harvest of information for the scientist, and thus the practical approach is indeed practical.

A few neuroscientists take another approach. They say that there is no problem of the Homunculus. This is a bold approach. In times past, people questioned the existence of God. When a person questions the existence of God, then he can ask: “If God didn’t create this world, then where did all this stuff around me come from?” In a very real sense, this was the beginning of modern science. This is why I say that doubting the existence of the Homunculus barrier is a bold approach. Unfortunately, unlike doubting God, it does not seem to yield any insights or offer a path to understanding. The Homunculus, although an imaginary construct used to symbolize something we do not understand, is a lot more real than God, at least for science.

Neuroscience offers an unlimited opportunity to collect data. This is because the brain is so complex. Imagine that you made a list of everything that you can do, of every mental state, whether of reason, or emotion, that can take hold of your mind. For every thought, for every class of experience, there are neurons in the brain that form the basis of this mental experience. But because the various forms of mental experiences are so diverse, different areas of the brain, different sets of neurons, are specialized for dealing with different types of experiences. What this means is that for every item on the list you made, a neuroscientist can find that the brain is doing a different thing; certain neurons are active for some things, like listening to music; whereas other neurons are active for other things, such as playing football. In some cases the same neurons may be active in different tasks, but then the pattern of their activity will be different. Thus, neuroscience will never run out of opportunities to things of the form “area A of the brain does this” whereas “area B does this” or “these neurons show activity pattern x for this situation, and pattern activity pattern y for this other situation.” As long as this is considered to be an interesting pursuit, then neuroscientists will have jobs.

However, it is not obvious that this will ever allow us to penetrate the barrier of the Homunculus. I am tempted to say that the accumulation of this sort of data could never lead us to solve the Homunculus problem, but I hesitate because of the example of Charles Darwin. Charles Darwin examined a large collection of data, a collection of seemingly unrelated facts, and deduced the theory of evolution. As neuroscience seems to represent a huge collection of unrelated data, by analogy, another Darwin might be able to look at it all and figure out a path through the Homunculus barrier.

However, there is an important difference. Evolution explains the simplest basis of life. But the Homunculus, or rather what lies beyond the Homunculus barrier, is the product of billions of years of evolution. The Homunculus is not likely to be a simple principle, like evolution, but rather something very complex. For this reason I think that the analogy between neuroscientists and natural historians is of limited utility.

Neuroscientists are furthermore unlikely to solve the Homunculus problem because they do not try to solve it. It does not enter into any research grant proposal, because no one knows what type of experiment to propose to find the Homunculus in the brain.

If the neuroscientists are not likely to cross the Homunculus barrier, then what would be a good approach? I have seen two approaches that seem promising, in the sense that they at least offer some hope of solving the problem.

The first approach is to attempt to create a Homunculus in a computer. This requires first simulating a world, then simulating life in that world, then allowing that life to change according to the laws of evolution. If the computer could produce a realistic simulation of the world, and the simulation could run for a long enough time, then in principle it should create intelligent life, just as the real world did. If a computer simulation contained intelligent life, we could study this intelligence far more easily than we can study the real brain. Then we would have a chance to cross the Homunculus barrier. The problem with this approach is that the world, even a small piece of it, with all its atomic and sub atomic particles, is far too complex to simulate in a computer, or even all the computers in the world put together. Therefore, it is necessary to make a simplified simulation. But if the simplifications are of the wrong kind, then they may preclude the evolution of intelligent life. Therefore there is no guarantee that the project will be a success.

The second approach represents evolution of another kind. A computer is, in principle, capable of computing anything that is computable. That means that a computer is capable crossing the Homunculus barrier — if someone knew how to program it well enough. Computers are programmed using computer languages. The most primitive language of the computer is machine language, which is difficult and cumbersome for a human being to use for programming. For this reason, people have developed other more advanced languages to program computers. The interesting point is that computer languages can be built from other computer languages, and these languages can become ever more powerful — that is to say, that a person can tell the computer to do more interesting things with fewer words. Thus, while the neuroscientists toil in their laboratories collecting data, computer programmers, or at least a subclass of them, computer language designers, continually develop programming languages to make them more and more powerful. It is theoretically possible that by developing more and more powerful programming languages, these languages could begin cross the Homunculus barrier, perhaps without anyone realizing it. I think this approach is more likely to succeed in crossing the Homunculus barrier than the approach of simulating evolution, and far more likely to succeed than the neuroscientists with their study of real brains. The neuroscientist is a spectator of the computations of the brain which are conducted in a form of language he does not understand. A computer language designer controls computation with ever more powerful tools that he understands because he makes them himself.

It is not to be lamented that we cannot cross the Homunculus barrier. If we could, we could develop machines with consciousness, true artificial intelligence. An intelligent computer could combine the power of the human mind with the power of the computers that we use to control our world. The intelligent computer might regard the human being, rightly or wrongly, as a competitor; perhaps as a servant or slave; and perhaps, as unnecessary. Thus, crossing the Homunculus barrier might be the end of civilization as we know it. Even coming close to crossing the Homunculus barrier could be dangerous, because the computer itself might make the crossing without our knowledge. Indeed, how would we know if a computer were conscious? We might not find out until it is too late.

First published 18 March 2006