If you've followed the debate that's been going on in the blogosphere, you know that Ray Kurtzweil, computer scientist, inventor,"futurist" and self-proclaimed prophet of "the singularity", recently proposed that it would be possible to reverse engineer a human brain - basically simulate it in a computer - within the next 20 years. Developmental biologist and blogger PZ Myers was not impressed and wrote a harsh and critical evaluation of Kurtzweil's arguments. So did other people: How to build a brain wrong, What Ray Kurzweil and geoengineering have in common, Reverse-Engineering the Human Brain? Really?, Baseless speculation variety hour.
Kurtzweil rebutted and Myers was quick to counter his claims once again.
It all makes for some very interesting reading, especially PZ Myers' lucid thrashing of Kurtzweil's ideas about the brain. His main contention is Kurtzweil's statement that "the design of the brain is in the genome", which completely ignores the importance of ontogeny, the developmental history of an organism within its own lifetime, in the formation of the brain's architecture. That "information" is most certainly not in the genome, instead it arises in the interaction between the developing brain and its environment. Myers offers some direct examples of the kind of complexity that arises even when you look at individual genes/proteins that are involved in the development of the brain.
You can guess by now which position I hold. As someone who studies genomes at a neuroscience department, I couldn't possibly do anything else but agree completely with PZ Myers. Just saying something like that - to "reverse-engineer a brain", shows that you have no idea what it is you want to simulate. "Reverse-engineering" a brain is a far more complicated proposition than Kurtzweil's optimistic time-scale lets on.
This whole story made me think of a section from one of my favorite textbooks; "Developmental Biology" by Scott F. Gilbert. Tellingly, it's the opening paragraph of the very first introductory chapter.
Between fertilization and birth, the developing organism is known as an embryo. The concept of an embryo is a staggering one, and forming an embryo is the hardest thing you will ever do. To become an embryo, you had to build yourself from a single cell. You had to respire before you had lungs, digest before you had a gut, build bones when you were pulpy, and form orderly arrays of neurons before you knew how to think. One of the critical differences between you and a machine is that a machine is never required to function until after it is built. Every animal has to function as it builds itself.
The concept presented here so skillfully is central to understanding how brains and nervous systems work. Read this part again: "...a machine is never required to function until after it is built. Every animal has to function as it builds itself." To "reverse-engineer" a brain in the true sense of the term would require a machine that is not only able to produce the wanted outcome - a cognisant brain - it would have to build itself while it matured, experienced and learned how to build itself. Not only to produce the wanted outcome, the behaviors we want as an output from the system, but in order to build the neural networks in the first place. The two processes are inseparable.
Even when Kurtzweil acknowledges the role of ontogeny and clarifies his calculation of the amount of "information" required to build a brain, he commits the same error.
It is true that the brain gains a great deal of information by interacting with its environment – it is an adaptive learning system. But we should not confuse the information that is learned with the innate design of the brain. The question we are trying to address is: what is the complexity of this system (that we call the brain) that makes it capable of self-organizing and learning from its environment? The original source of that design is the genome (plus a small amount of information from the epigenetic machinery), so we can gain an estimate of the amount of information in this way.
I think that the question he poses is a very sensible one and it clarifies his point a great deal, but "we should not confuse the information that is learned with the innate design of the brain" is the understatement of the century. He just doesn't get it.
I'm not involved in technology at all, and I have to admit that my understanding of engineering and computer science is very basic bordering on nonexistent. But I have not seen that Kurtzweil or any other AI researcher has addressed the practicality of developing AI by creating a reverse engineered computer that learns as it builds itself, like the human brain, without us interfering with the process due to our lack of understanding of how it works. To me it seems like a very impractical and backwards proposition, to "get the ball rolling" so to speak and end up with precisely the outcome we want - artificial emergent qualities such as cognition. I'm willing to accept that it's because of my limited knowledge though. At the very least it's a far more difficult proposition than Kurtzweil lets on.
There's a philosophical hurdle as well. If we succeed, how will we know it?
Swedish blog tags: Vetenskap, Teknologi, Neurovetenskap, Hjärnan, Artificiell intelligens,
Technorati tags: Science, Technology, Neuroscience, Brain, AI, Ray Kurtzweil