I suspect that we’ll see AI more or less when it’s inevitable.
Which is to say…
The factors we’re talking about here would be made moot by an N-times increase in computational speed, for some N.
Looking 20 years in the future, following Moore, I’ve read that supercomputers should have 10,000 times the “computing power” of the brain. Forgetting for a moment that I have no idea how this is defined…
I think differences in definition, too, will be absorbed by a sufficient increase in computational power.
So, to me, 20 to 30 years seems like the timeline. I suspect we will first see AI simply as a result of a self-improving system run on a sufficiently powerful computer.
Wait long enough, and you can always argue Ad Brutus Vis Moore – appeal to brute force using Moore’s law.
Maybe someone develops a system for finding faces for cameras, and it’s particularly good.
Ad Brutus Vis Moore – nice…
I really think the main problem right now is a lack of horsepower.
And perhaps architecture, but that, too, is heading in the right direction.
You don’t think that ultimately (in hindsight, and possibly not understandable by humans) there’ll sit some fairly “trivial” (meaning elegant) idea underneath recursive self-improvement?
Even if not invented originally as such by humans, at least at some point reduced to that by the AI itself?
Not to say that that one particular idea would be the “only” way to do intelligence, just that it would be one particular idea.
Sure… I think that after the initial creation of AI, we should see refinement of the concept.
The same way that evolution - if you zoom out far enough - is actually pretty damn simple. And ”just another way” of creating intelligence.
A rather inefficient one I suspect, but arguably so far the only one that has been shown to work… :-)
Right. That said, I don’t think we’ll ever look back and thing, ”If only we had the right idea in 2009.”
It’d be like saying that if you just had the right body, you could win a drag race with a 5 HP lawn mower engine.
Sure, the science has been refined, but you still need a 1000 HP engine to win.
Jup, just like the science of neurons is refined, but you still need a gazillion linked in an icky wet network before they’re useful.
Exactly… :-)
And, even at that, I’m sometimes surprised how limited human potential is.
…and, truth be told, heartened.
Just like a need an atom clock on a sattelite and a large hadron collider to be a physicist - which isn’t the case….
I’m trying to figure out why you’re less excited by this topic than I am.
You don’t think you could possibly contribute to this idea? Or it doesn’t interest you enough to pursue it?
Well, I’m totally stoked by the idea, but I feel like right now our biggest limiting factor is horsepower, and intuitively, I think the rest is non-critical.
Which is to say…
I must admit, AI has been a long passion of mine. Unfortunately, after seeing too many if-then-else forests in game development, I had my personal ”AI winter”.
Right.
Video-games are not AI, they’re magic. Having been a fairly serious hobby magician to the point of making money at weddings and birthday parties, I can really make that analogy strongly. It’s all smoke and mirrors.
For me, I think, the problem is that I’ve done a bunch of stuff with genetic systems, neural networks, and so on, and have always been slightly disappointed with the possibilities.
But dude, have you seen NVIDIA Fermi and the like?
Well, I don’t have to tell you; you’ve had lunch with the Acceleware guys.
Okay, I admit, CUDA seems really interesting to me.
I haven’t researched it enough, but was thinking the other night that it should be high on my list.
I am very excited by the present trend of massive parallelism.
The thing is, my gut tells me that some amalgamation of “dumb” ideas let loose on Fermi^N (i.e., corporate level desktop super computing five years from now) will help thaw the AI winter frost.
And the minute it happens, that almagamation will (in retrospect) turn out to be closer to 10,000 lines of code, rather than 1,000,000.
10,000 LOC is doable for mortals, the out-of-reach nature of retrospect being a significant obstacle of course.
Agreed.
So why not you?
See, I have a good excuse…
I don’t want to destroy the world.
Ah… good reason.