Saturday, October 09, 2004
When will computer hardware match the human brain?
I ran across Hans Moravec's paper about when computer hardware may match the processing power of the human brain. It is written in a nicely understandable format and well worth a look.
It predicts that, by 2020, personal computers costing about $1000 should have the processing power equivalent of the human brain.
He goes on to argue that computing power is the primary limitation being experienced by AI researchers today.
I have no doubt that more computing power leads to better AI, but I am deeply skeptical that this is the primary limiting function. My guess is that the methodology for creating human-like thinking is the primary bottleneck at this point, and merely the fact of having a computer of equivalent power will by no means guarantee human-like thinking (and, for the record, I don't believe that Moravec is asserting this particular relation).
What is worth thinking more deeply about is the time beyond when, assuming that Moravec's calculations are valid, a $1000 personal computer has the power of the human brain.
As I stated, I think more than computing power is required, but I would say that vast amounts of computing power allow us to do things terribly less efficiently than they might have evolved and still get to the same solution. Said another way, once we have computing power several orders of magnitude greater than humans, we can use programs that may simulate brain functions very inefficiently, but due to the much greater power at their command, still approximate human thought.
Another interesting way to think about this is that a few years after the computing power of the brain is readily available for consumption by Windows 2018, the same computing power will be available in a much smaller package. Therefore, if Moore's Law continues apace (and, though I have my reservations, there seems to be plenty of possibilities to keep it going for 20 more years) we will have surpassed evolution's ability to manufacture a general purpose thinking device efficiently. Certainly there are a number of reasons why evolution isn't particularly concerned with more power and smaller space, but it's an interesting thing to think of nonetheless.
Extended further, at some point, an area smaller than that of one of the keys I'm typing on now will contain more intelligence than all of existing humanity. It is when we start thinking about these kinds of extrapolations that my doubts on the continuation of Moore's Law grow very deep indeed.
Two additional things of interest. First, there are a couple of neat graphs included in Moravec's paper. I have added them to my Flickr account, and will link them in here as soon as I get a chance.
Second is that the Comments on the paper are an interesting read in their own right. I was surprised to find Robin Hanson commenting here only because I'd just been reading some of his work about betting on science as I had come around to a similar conclusion relating to the future (more later).
When will computer hardware match the human brain? by Hans Moravec