I made this post on Slashdot.org today, and felt like I wanted to mirror it in my weblog. I really think that an “inclusive” approach toward Artificial Intelligence, one that is multi-disciplinary, can be the key toward creating our ultimate goal, which is (basically) a new class of slave. I’ll leave the moral implications of creating our own mechanical slaves out of the discussion for now, but really, the main benefit of creating an artificial intelligence, besides having it do stuff for us, would be to give us some company…
Posting follows:
Once we can define what it is we’re trying to artificialize, maybe we can make more progress in artificializing it.
There’s the rub. I don’t think we can begin to understand how the brain works until we build stuff that approximates its operation enough to create working theories. The Wright Brothers didn’t have a full understanding of flight dynamics prior to building their first aircraft. We’ve come to that knowledge through a lot of trial, error, and testing.
The main problem is that intelligence is such a nebulous thing. The Wrights had one goal: “make it fly”. Now when engineers design, they use that basic goal and expand it with “faster”, “safer”, “more manueverable”, “more fuel-efficient”, “able to carry this weapons loading”, “lower stall speed”, “stall avoidance in low-speed turns”, etc. We need to come up with some basic rules on what capabilities we expect out of an AI, then expand on it.
We have some pieces of that now. Like the little round vacuum-cleaner thing. The goal is for it to vacuum floors. Great, it does that. Now make it navigate stairs. Once it can do that, then make it learn to pick up stuff on the floor, rather than vacuuming around it. Then maybe create an attachment that allows it to load a dishwasher.
You get the idea.
I think the point of view of people that think AI has “failed” is a bit skewed. Yeah, we don’t have any AI that can reason at human level yet. But we have devices that can easily beat the intelligence of roaches. And we’re working on things that can exceed or augment the intelligence of small mammals at this point. We’ll get there, but incremental progress is the only way it’s going to happen, IMHO. And we’ll arrive at the goal from several different directions, probably including analog (mechanical approach, responding to stimuli using non-digital means), emergent behaviors (colonies of processes), neural net, and hybrids of these, each of which can complement the other in creating reliable systems that use different logic depending on the context of the item they are analyzing.
This discussion really, really makes me want to go back to school and get a doctorate specializing in AI. I feel like such a goober noober discussing this stuff in public, but my gut feeling is that competent, ubiquitous AI will be the catalyst toward improving the human condition around the world.