Sunday, February 7, 2016

More Deep AI masquerading as intelligence

   The computer beating a Go master for the first time, has caused the world's science writing media to burp up the usual exaggerated encomiums on the significant advances in AI. As the Chinese say "Mayo...mayo...mayo", because, there is nothing about a good Go playing program that requires intelligence, it is a computational algorithm for finding optimizations. You cannot get any more mechanical or deterministic.
   Of course it takes a great deal of intelligence on the part of the people writing the Go playing program and picking the optimized strategies the computer can follow. Even if this programming happens iteratively through a (misnamed) "learning" program. It also takes a lot of computing power.
   Why is it that the intellectual community persists in confusing smart programming and vast computational power as "intelligent"?  To put this into a 1970'ish perspective, imagine headlines about the new TI-1000 hand held calculator, demonstrating "deep" AI. I think I can answer this question: it is because intelligence is not defined and, in humans, is measured using tests that we imagine applying to machines.
   I believe none of that has anything to do with actual intelligence. I don't have a definition either but then I don't claim to be "deep" [except I do: see "Best Models".]. I am guessing that there is real pressure to monetize investments in "Big Data", "Analytics", and all that other "deep" stuff. The fact that these technologies are being dumped into open source gives us a clue as to how that is working out. The good news is that these terms are being neutralized from over use and hype, leaving a new generation of words available for a future in which actual progress is made - coming from psychologists and philosophers and not from computer scientists who are not students of human behavior.

No comments:

Post a Comment