Mouse brain simulated on computer.
More accurately, half a mouse brain was simulated for 10 seconds, at one-tenth the processing speed of a biological mouse brain—or for one second, in artifical-mouse-brain time.
Despite repeated analysis, the researchers’ data unambiguously portray the artifical mouse’s cortical activity for that one second as being something along the lines of “Oh no, not again!” The team cannot explain this bizarre result, but suspect that petunias may have something to do with it.
Back when taking survey courses in AI and the like, it always seemed to me that modeling biological neural nets themselves, rather than one thing those nets happen to do (eg, logical operations on propositions), was clearly the “right” way to approach the problem.
(Other approaches have been used, but none have yielded anything helpful, even when this research has been carried out by mice, themselves.)
I have not heard much about neural nets in recent years, so I’m glad to see people are still working on them.
I also, in my unbounded expertise on the subject, have felt that contemporary hardware just isn’t up to the task. While software can exhibit non-deterministic behavior, computer hardware is linear and deterministic. Using that to try to model something not so constrained—consciousness—is of course going to be problematic. It’s kind of nifty that we’ve got the horsepower now to try to sort-of simulate even half a mouse brain, for one mouse-brain second. I think—expert that I am—that quantum computing looks like a good bet for delivering the goods in some future. If not, I suppose there is always consolation to be had.
Anyway, I’ve become a bit skeptical of top-down attempts to create real AI. The whole “born from the sea of information” thing always struck me as more poetic, certainly, if not simply more plausible. Crap, that wasn’t a Hitchhiker’s Guide reference. OK, time to end the post.
