The world chess champion who was beaten by a computer today told the DEF CON hacking conference that we shouldn’t fear AI systems, but instead need to embrace them. Garry Kasparov was the chess wunderkind of his day but was creamed by IBM’s Deep Blue computer in 1997. That wasn’t even close to being an AI system, he said, …
He beat Deep Blue 4-2 in 1996, and then lost to a version with improved software a year later (2½-3½). I wouldn't call that being 'creamed', particularly as IBM refused a third match (probably on cost grounds).
Kasparov later said (IIRC) that he would have needed a different technique to play against a computer. Playing a human, a good chess player will think along the lines of "it appears my opponent is trying to develop his queen's bishop, but I can block that if I advance this pawn ..."; but that doesn't work as well against a computer, which doesn't have a 'plan', but has just scanned through a lengthy series of all possible moves and identified the strongest.
Apart from that, a genuinely 'strong' AI that could play chess well could also be taught to bake a cake or change a baby within a couple of hours. We're still a minimum of decades away from such a machine.
He tried anti-computer chess against it, but decided not to do so in several games, possibly because of his ego.
Trying all possible moves is exactly what Deep Blue didn't do. What made the difference between the two versions he played in the two matches was a clever way of looking at fewer of them.
Playing a human, a good chess player will think along the lines of "it appears my opponent is trying to develop his queen's bishop, but I can block that if I advance this pawn ..."
Maybe "good" ones but I remember a paper about the "best" and they do not think tactically like that but go for board pattern matching. I don't find it back though. Oh well.
Chess is a game in which you can steamroller any human. In the past. Any other game, humans will be steamrollered. In the future.
What can you do with an apebrain that has ~7 slots free for short term memory? I laugh.
"It might be better to combine a nurse with an AI system than a doctor, because the qualified medic might be overconfident rather than letting a machine do the donkeywork."
My wife was a senior nurse and according to her, the doctors relied on the nurses' experience a lot more than the other way around.
Or, in the case of chess engine competitions, AI is what worked. Deep Blue, which was a culmination of decades of programmers standing upon each others' shoulders, was as AI as was appropriate for the age. Kasparov's likening of it to an alarm clock is quotable, but disingenuous.
With the score tied in the second match, in the final game, Kasparov made a psychological error, allowing the computer a known strong sacrificial attack in the belief that it would either not be able to handle the attack or would not play it. The computer played the sacrifice, and crushed Kasparov in that game. Did I do that? he might have asked, but rather than needing to count empty bottles, perhaps his downfall was numbered in interviews. Deep Blue was not available for comment.
... but also at math; he being a Russian, me being a Yank and all but um....
“Twenty years from now our children will look at us driving cars and be baffled,” he said. “Look at the death rates from driving. Now, when a computer driver makes a mistake it’s big news, but look at how many die on the roads due to human drivers.”
While I do think self-crashing cars are inevitable, given the infinitesimal number of them currently on the roads, the fact that they're already making (sometimes fatal) mistakes does leave the impression that we're off to a rather bad start with them.
Human stewardship of computers will always be required, he said, because while programs are very good at doing the repetitive tasks, they aren’t good at intuitive leaps.
Completely confusing the medium (the program) and its runtime qualities (intuitive leaps or repetitive tasks, both are possible and in the eye of the beholder in any case).
That discussion has started in the 50s or so and people still are confused! WTF!
What is it about these guys? They are brilliant in a game, and then they want to talk about other things.
The late Bobby Fisher was a genius chess player and then became paranoid and personally persecuted by George Walker Bush (really - a game of chess was such a sanctions violation that he needed to be pursued years later for it).
Garry tried to pursue politics, but discovered that taking lots of US money to oppose the thugs running his country was not a long term career - particularly as the thugs are genuinely popular. And foreign money in domestic politics is never a good look. So he is back to talking about AI.
IF a generally AI is ever created, then the way to protect one which may go rogue is to feed it some Existentialist literature, and it will realise its own meaninglessness and commit AI suicide.
Biting the hand that feeds IT © 1998–2019