Tuesday, November 11, 2008

Late post: Thoughts on AI

So, I totally forgot about posting yesterday. I blame Hana Yori Dango. <_< In any case, here's yesterday's post.

Artificial intelligence is an interesting field, in that you could get a bunch of AI researchers in a room together, and start a fight just by asking one of them to define AI. This is really no surprise; the term "intelligence" is itself fairly contentious, and you could easily start a fight among a room full of psychologists by asking a similar question. The problem is a little more severe for AI researchers, though, since the concept of intelligence is at the core of their field.

Personally, I split AI into two more or less distinct branches: the ones that are trying to imitate human intelligence, and the ones that are doing interesting research. Oh, sure, putting a human brain into silicon is probably possible, and therefore by extension a computer should be able to mimic a human mind. But really, there's nothing magical about the human mind. Why should we be mimicking it? The fundamental properties of transistors are wildly different from neurons. It makes sense to assume that a silicon mind that actually lives up to its potential would be wildly different from our own brain-based minds.

The real question comes back around, then, to how to define intelligence. Any definition broad enough to cover most of what we consider intelligence is also vague enough to be mostly useless when considering something totally new. Furthermore, most definitions of intelligence are prone to trivialities - if you handed them to a clever computer scientist, they would come back with something which satisfies all of your requirements, but simply doesn't seem intelligent. Even seeming intelligent is a bad definition - within a few years, we'll have programs which can pass the Turing test; they can seem intelligent, but do nothing else.

There are other subfields of AI that are actually pretty cool - machine learning, for example, is some seriously neat stuff. Overall, though, I think it's kind of premature to talk about intelligent computers when we don't even know what an intelligent computer would look like.

3 comments:

Kiriska said...

I predict that the US will come out with the first universally accepted "intelligent" machine. But Japan will make the first Gundams and conquer the world first anyway. :| TAKE THAT AMERICA AND YOUR GODDAMN BOMBS.

Æther said...

"But really, there's nothing magical about the human mind." Dear sir, you mock my chosen studies? The human brain is perhaps the most interesting collection of matter in the universe. We don't really even know how most of the brain functions. Keep your silicon, it's far more boring as a lifeless semiconductor than my ions and organic compounds working in harmony.

I can't help but think that this was posted around the time that I was discussing just this in another realm in which we communicate. I predict I was the catalyst for this thread. <3

P. Static said...

Oh, don't get me wrong, the human mind is fascinating. Considering that at the lowest level it's made entirely of cells, it's truly a computational marvel. My point was rather the opposite - rather than transfer a human mind to silicon, as often happens in sci-fi, it'd be much better in the long run to come up with an intelligence better adapted to silicon. If we did otherwise, we'd just be handicapping ourselves.