The History of My Brain

The brain is a remarkable computational device (not material? bah. get your faith out of my office.) A rough estimate of the computational power needed to simulate the brain’s neural power is around a billion MIPs (millions of instructions per second.) When I was growing up, I was fascinated by the possibilities of artificial intelligence. In fact, this forms the basis of my three ways to do theoretical physics:

1. Do it yourself. This requires lots of work and the ego to believe you can actually figure things out. The benefit is that you can perhaps contribute now. Unfortunately, I don’t believe that this is a long term solution to doing theoretical physics (see 2 and 3)
2. Build a computer to do it for you. Or build a computer to speed up certain functions in your brain. How long before we can build computers with vastly greater ability than our mushy brains? Are our brains really operating at some optimal limit for computation? For number crunching tasks, most definitely not. This solution seems to me the much more realistic long term way to contribute to theoretical physics. Of course, you may not get the credit for any of this. The computer that did all the work will surely the lions share of the credit. Can a computer win the Nobel prize?
3. Join the Search for Extraterrestrial Intelligence (SETI). The theory here is that if we discover intelligent alliens they will more than likely be more advanced that we are. Thus they will have used 1 and 2 to have learned vastly more about theoretical physics than we currently know. So if you join SETI and SETI succeeds, you will just be able to ask the aliens about advanced theoretical physics. Of course, this assumes that the aliens are nice and not bug-like (because as every fan of sci-fi knows, bug-like aliens are all about killing fleshy humans.)

But back to the topic at hand: simulating the human brain. One problem with an analysis that just looks at the raw computational speed of the human brain is that it disregards the computational complexity inherit in the structure of the brain itself. In particular, the brain is the product of 500 million plus years of evolution (I want to start counting from at least some ancestor which had some sort of nervous system). So the question I’ve always found fascinating is how large this overhead of so many years of evolution costs in producing the structure of the brain? Do we need to simulate all 500 million plus years of evolution, yielding a computational power far beyond even our best projections of the computational power of the brain? This is particularly troublesome when you think about the power of evolution: our ancestors have been roaming around the landscape of possible brains for a long long time. A counter to this is the idea that we don’t need to simulate all of the evolution, but only the developemental biology which leads to a full grown human. This later point of view seems less computationally challenging, but still daunting. And I’ll bet that this is the computational hangup for simulating human intelligence: not the raw power of the brain, the but the power to reproduce the developement of the brain. I wonder if anyone has done any rough calculations on the computatonal power needed for this latter task.
As I said, when I was growing up, I was fascinated by the possibility of advanced computer intelligences. I can still remember the frustration of not being able to produce intelligence on my TRS-80 Color Computer II. So I left those dreams behind me, but I promised myself that if I ever thought computers were powerful enough to really start producing intelligence I would return to the field. I guess you shouldn’t be surprised, then, if in thirty years you find me hacking away in some dark basement beside a machine named “Newton.”