I was thinking about AI in the shower this morning, about what computers can and can’t do, about what computers are like as far as “life” is concerned. What follows is probably obvious, probably something that one of the AI gurus like Ray Kurzweil or David Weinberger scribbled on a scrap of paper and then threw away as too painfully obvious, but it helped me see AI in a somewhat different light.
Somewhere between the shampoo rinse and drying off, I sensed that computers are much more like trees (or other plants) than like animals. A tree is, in a reductive sense, an algorithm for living and reproducing — but it’s power is limited by immobility (outside Middle-Earth). Botanical life accomplishes marvelous and intricate feats, but its capacities are limited by its need to stay rooted (or in the case of the rootless plant forms — epiphytic bromeliads — with which we became acquainted when we lived in Florida, umm, “niched” maybe).
A computer likewise operates under the constraint of needing roots — a cord (to sustain the analogy) or battery pack (which offers short-term mobility, but as any road warrior knows, the battery begins pooping out when there’s no outlet in sight). But a battery pack is much more like the constraint of a bromeliad plant than like the freedom of a grazing animal. A robot, on present models, is still much closer to being a Venus Fly-Trap than a centipede.
All this doesn’t mean that technologists won’t be able to devise a clever way of simulating animal self-sustenance; I expect they will, surprisingly soon. But I’m more interested (for the moment) in the energy-sustenance point than I am in the ballyhooed (cue the scary music) Singularity. David W has convinced me not to be unduly impressed by comparisons of data storage devices with human memory and deliberation.
Posted by AKMA at August 9, 2004 09:02 AM | TrackBackDavid, another David (Dubin), and I were talking about this shortly before we all left Montreal, admittedly in the context of attempts to get computers to understand natural language.
My husband's point, which Dubin agreed with (and so do I, not that I matter), was that a great deal of language understanding rests in understanding of the world, a world no computer as yet can get out and participate in. It's pointless to expect a purely linguistic system to recapitulate that knowledge; ergo, systems that try to do AI based on pure Chomskybottism or (worse) random collections of "facts" are doomed to fail.
Would like to say more, but in pain, sorry.
Posted by: Dorothea Salo at August 9, 2004 09:35 AMFollowing up the previous comment about understanding the world, of participating it, is the very aspect that some (but not all) AI researchers claim will lead to the development of some form of AI. For them the development of embodied AI, such as humanoid robots, will be where the field will develop most. In a sense, personhood (or at least machine intelligence) becomes linked to embodiment and incarnation.
For example: http://www.ai.mit.edu/projects/humanoid-robotics-group/
Posted by: Stephen at August 9, 2004 09:11 PMI think many AI researchers fell into the same trap as many middle managers: they confused the name for a thing with being the thing. Hmmm, that's a little too dense, even for me. Maybe just better to say that management, and AI researchers, want to be able to completely characterize the the making of thought solely through mathematics. Okay, still not there. But I'll leave it at that because I do have other work to do, but before leaving here's another thought-making comment from another non-computer guy (English professor, I believe) for:
"Computer languages provide the possibility of an exact and precisely limited correspondence not only between what is said and what is meant, but also between what is meant and what is so in the strictly defined system about which, and only about which, statements can be made."- Richard Mitchell, "Underground Grammarian" vol 8:1 Posted by: John Miller at August 10, 2004 07:43 AM
All is vanity. But the Richard Mitchell remark quoted by John Miller is remarkably precise and topical.
And consider the sponge, "niched" as it were.
And the bumper sticker: My Australian Shepherd is Smarter Than Your Honor Student
Ultimately Deep Blue will win every time. Does that make it intelligent?
Oh... we want to remember that Vernor and Joan are heavy into Science Fiction and Fantasy. Proclamations regarding the end of the human era are far more appropriate to an environmental assessment of viral replication than they are to machine state incrementalism as re. processing power, stoprage, and retrieval. zo much for ze zingularity
Posted by: Frank Paynter at August 10, 2004 10:04 AMRichard Mitchell is a writer for the ages. If I'm writing something that seems, to me, to be particularly eloquent or insightful it usually turns out I am just paraphrasing something written by Mr. Mitchell. For those piqued, I would recommend reading The Underground Grammarian. When you get through that, well Richard also wrote a few books. When you get through that, well, maybe you'll want to try on Happy Tutor's recipe for world domination. Or not. Whatever floats your boat, I say.
And, by the way, of course it doesn't make Deep Blue intelligent, it just means Big Blue is damn good at chess.