Issue link: https://cp.revolio.com/i/5203
40 winter 2009 Artificial Intelligence & Reverse Engineering the Brain emergent property of what goes on in the neocortex, which is where we do our thinking. And you could build a neocortex. In fact, they are being built in the Blue Brain project, and Numenta also has some neocortex models. In terms of hierarchies and number of units in the human brain, these projects are much smaller. But they certainly do interesting things. There are no tubules in there, there's no quantum computing, and there doesn't seem to be a need for it. Another theory is the idea of purposeful complexity. If you achieve a certain level of complexity, then that is conscious. I actually like that theory the most. I wrote about that extensively in The Singularity is Near. There have been attempts to measure complexity. You have Claude Shannon's information theory, which basically involves the smallest algorithm that can generate a string of information. But that doesn't deal with random information. Random information is not compressible, and would represent a lot of Shannon information, but it's not really purposeful complexity. So you have to factor out randomness. Then you get the concept of arbitrary lists of information. like, say, the New York telephone book is not random. It's only compressible to a limited extent, but it's not a high level of complexity. It's largely an arbitrary list. I describe a more meaningful concept of Purposeful Complexity in the book. I propose that there are ways of measuring purposeful complexity. In this theory, humans are more conscious than cats, but cats are conscious, but not quite as much because they're not quite as complex. A worm is conscious, but much less so. The sun is conscious. It actually has a fair amount of structure and complexity, but probably less than a cat, so... So: how do you go about proving something like that? RK: Well, that's the problem. My thesis is that there's really no way to measure consciousness. There's no "Consciousness Detector" that you could imagine where the green light comes on and you can so, "OK, this one's conscious!" Or, "No, this one isn't conscious." Even among humans, there's no clear consensus as to who's conscious and who is not. We're discovering now that people who are considered minimally conscious, or even in a vegetative state, actually have quite a bit going on in their neo-cortex and we've been able to communicate with some of them using either real-time brain scanning or other methods. Today, nobody worries too much about causing pain and suffering to their computer software. But we will get to a point where the emotional reactions of virtual beings will be convincing, unlike the characters in the computer games today. And that will become a real issue. That's the whole thesis of my movie, The Singularity is Near. But it really comes down to the fact that it's not a scientific issue, which is to say there's still a role for philosophy. Some scientists say, "Well, it's not a scientific issue, therefore it's not a real issue. Therefore consciousness is just an illusion and we should not waste time on it." But we shouldn't be too quick to throw it overboard because our whole moral system and ethical system is based on consciousness. If I cause suffering to some other conscious human, that's considered immoral and probably a crime. On the other hand, if I destroy some property, it's probably OK if it's my property. If it's your property, it's probably not okay. But that's not because I'm causing pain and suffering to the property. I'm causing pain and suffering to the owner of the property. And there's recognition that animals are probably conscious and that animal cruelty is not okay. But it's okay to cause pain and suffering to the avatar in your computer, at least today. That may not be the case 20 years from now. RK: I'm working on a book called How the Mind Works and How to Build One. It's mostly about the brain, but the reason I call it the mind rather than the brain is to bring in these issues of consciousness. A brain becomes a mind as it melds with its own body, and in fact, our sort of circle of identity goes beyond our body. We certainly interact with our environment. It's not a clear distinction between who we are and what we are not. My concept of the value of reverse engineering the human brain is not that you just simulate a brain in a sort of mechanistic way, without trying to understand what is going on. David Chalmers says he doesn't think this is a fruitful direction. And I would agree that just simulating a brain... mindlessly, so to speak... that's not going to get you far enough. The purpose of reverse engineering the human brain is to understand the basic principles of intelligence. Once you have a simulation working, you can start modifying things. Certain things may not matter. Some things may be very critical. So you learn what's important. You learn the basic principles by which the human brain handles hierarchies and variance, properties of patterns, high-level features and so on. And it appears that the neocortex has this very uniform structure. If we learn those principles, we can then engineer them and amplify them and focus on them. That's engineering. Now, a big evolutionary innovation with homo sapiens is that we have a bigger forehead so that we could fit a larger cortex. But it's still quite limited. And it's running on the substrate that transmits information from one part of the brain to another at a few hundred feet per second, which is a million times slower than electronics. The intra-neural connections compute at about 100 or 200 calculation per second, which is somewhere between 1,000,000 to 10,000,000 times slower than electronics. So if we can get past the substrates, we don't have to settle for a billion of these recognizers. We could have a trillion of them, or a thousand trillion. We could have many more hierarchal levels. We can design it to solve more difficult problems. So that's really the purpose of reverse engineering the human brain. But there are other benefits. We'll get more insight into ourselves. We'll have better means for fixing the brain. Right now, we have vague psychiatric models as to what's going on in the brain of someone with bipolar disease or schizophrenia. We'll not only understand human function, we'll understand human dysfunction. We'll have better means of fixing those problems. And moreover we'll "fix" the limitation that we all have in thinking in a very small, slow, fairly-messily organized substrate. Of course, we have to be careful. What might seem like just a messy arbitrary complexity that evolution put in may in fact be part of some real principle that we don't understand at first. I'm not saying this is simple. But on the other hand, I had this debate with John Horgan, who wrote a critical article about my views in IEEE Spectrum. Horgan says that we would need trillions of lines of code to emulate the brain and that's far beyond the complexity that we've shown we can handle in our software. The most sophisticated software programs are only tens of millions of lines of code. But that's complete nonsense. Because, first of all, there's no way the brain