Surely the whole Deep Blue thing is just a false dawn for AI. Beating a human player through brute force isn't _that_ impressive. If anything its obvious.
What would be more impressive is a capped AI that can only look so many moves ahead and uses principles (such as say Nimzovich's opinionated twaddle) to win.
In quite possibly my only point of agreement with Noam Chomsky, I think our discussions of AI are impoverished because they turn on a distinction of language not of fact. Machines that can "think" are AI, machines that can do impressive calculations but not "think" are just impressive calculators. This is not a statement about AI, this is a statement about the English word "think". It sheds about as much light as asking whether an aircraft carrier can swim. (I am informed that aircraft carriers do indeed swim... in Russian.)
A lot of the other "Oh, that's just a cheap trick" non-AI things are a) much more useful to humans than the sci-fi notions people think of when they think AI and b) actually exist in the status quo. For example, if I told you that there exists a program which can make evidence-based judgments of a human's character to tell you whether they are likely to be trustworthy or not in their future business dealings, you would probably say "Egads! That is clearly AI! Holy cow, that is lightyears ahead of passing the Turing Test, and we're nowhere near passing the Turing Test!"
But if I told you "The program is called a FICO score" then, pfft, that isn't AI, that's just clever math and good selection of data sources.
This is one of the reasons why academic AI is, ahem, pretty dead as a field: after you actually get something working we drum you out of the field. Telling the topic of a document by reading it and understanding what it is about? AI! Leveraging anchor text on the link graph to do the work for you? Boooooooring. Granted, billion dollar business boring but still boooooooooring.
Indeed. And from the other end work in neurology and applied computer science continues to push out the boundaries of things that count as mere "cheap tricks". 40 years ago, no one would have looked at the face recognition and red-eye reduction features on our $100 cameras and said it was "obvious". It looks a lot like "thinking" if you don't understand it. Likewise the operation of much of the brain is better understood than it was in the 60's, and we're less likely to ascribe magic terms like "thinking" to the brain; even things like consciousness are starting to look like a trick (albeit not a cheap one).
What if someone told you that between 2 and 3 seconds into Deep Blue's calculation it ruled out all the moves predicted by Nimzovich's opinionated twaddle?
How is this fundamentally different from a human player contemplating the move and then 2 seconds later recalling Nimzovich, etc.?
The circumstances of that match are still very very shady. Deep Blue prepared for the match using full knowledge of every game Kasparov ever played. Kasparov was not granted access to any game that Deep Blue ever played.
During the match the team kept on reprogramming Deep Blue, to better combat Kasparov's style.
Still, these days computer chess programs are probably better than most grand masters. I completely agree that that doesn't say much about AI though.
I would not call that shady. All parties agreed on those conditions before the match, and of course Deep Blue was reprogrammed between games. Kasparov changed his approach based on Deep Blue's play, as well.
When I graduated from an ultra-liberal California university in the mid-70s, computing technology was consistently reviled as soulless, cold, and dangerous.
It is interesting to have watched this attitude transmute over the years into a belief, typified in this piece, that such technology will so revolutionize the world as to work some great deliverance in the human condition.
Back in the day, those who criticized the "Machine" would glorify the artisans and revel in the idea of human creativity in its pure form (unaided by advanced tools of any kind). The attitude was almost anti-science and was in any event highly "humanistic."
Now creativity and technology appear to be closely coupled in the eyes of young people in academic settings, to the point where information technology, at least, is seen as having almost redemptive qualities.
Not saying any of this is profound - just wondering if others have observed this changed perception as well or have thoughts on how it might have come about. Is it because technology became "personal" as opposed to "corporate"?
By the way, don't play Fritz as a chess adversary - it is like playing a demon from hell!
Not only that, but people's relationships with each other have been changed by machines.
It used to be that the FBI keeping a file open on somebody famous was considered scandal-worthy. Now FaceBook keeps much more personal information on all of us and people don't bat an eye about it. Younger people today seem to have completely abandoned the idea that they might have a private life and that privacy might have great value to themselves and others.
The noble savage and artisan view of humanity has changed into a view that we're all mostly just interchangeable nodes on the world network.
We have become the very machines that we once so hated.
It used to be that the FBI keeping a file open on somebody famous was considered scandal-worthy. Now FaceBook keeps much more personal information on all of us and people don't bat an eye about it.
That's a really good observation. People who are below a certain age don't have much perspective on
a) how mundane most information in FBI files always has been,
and
b) how much more pervasively personal information about most people is now shared by computerized networks than it was shared by small town gossip.
Is it not that in (a) the FBI have the file that is the issue. Just as now we'd be concerned if the FBI was keeping particularly close tabs on our FB details (I mean closer than the standard trawl-net approach).
Once I had some (considerable, 3 months) spare time and decided to fulfill my ambition of writing a chess program. This one would be special - it would work to a strategy, rather than brute force.
3 months later, and having written it first in Java, then C++, then C, I realised that it could never compete with the brute-force millions-of-nodes-per-second crunching machines.
This actually made me become disillusioned with chess as a whole, although the emergence of Rybka has restimulated my interest somewhat.
The idea that strong chess programs mean that AI is on the march is, sorry, laughable.
Chess is not where the action's at in AI anyways, so you're right, that idea is indeed laughable. These days, poker is hot new(ish) thing for AI research, with it's stochastic gameplay and psychological twist (bluffing and calling bluffs), and wide range of strategies. See, for example: http://poker.cs.ualberta.ca/
Don't the likes of Kasparov use strategy to beat the number crunchers though? How about a series of parallel strategies rather than simply crunching all moves and then assessing the best of the strategies. That is imitating a number of lesser players pitted against a better player - could the world #2 and #3 beat the world #1 (if not then probably no mileage¹ in this technique).
Sure, but does it matter? Being the best human chess player on the planet is still impressive, even if millions of machines can beat you. Not even only ‘still’ impressive – just as impressive as it ever was. It’s probably (theoretically) possible to build machines that are better at anything a human might attempt.
Probably not for long, though. At the rate they're getting better, computers will probably eclipse humans within my lifetime.
What I find interesting is that even though I get crushed by both, it's very easy for me to tell the difference between a 1d human and a 1d computer. I sometimes wonder if it would be informative to try to build a weak Go AI that gets mistaken for a weak human player.
I am surprised Kenneth Rogoff equates chess AI with "real AI."
"Real AI" is a long way off; my definition: ability for systems to learn from their environment (physical or Internet), change their internal models, and generally evolve their own abilities without requiring external help.
Off topic: I have never met grandmaster Rogoff, but I watched my friend Carl Wagner play him in a telephone San Diego vs. Boston match in 1975. My company SAIC hosted the San Diego end of the match, and I was very surprised to see my friend Carl playing number 1 board: Carl had been clobbering me at chess during lunch time for months and I was just about ready to give up the game since I had a long loosing streak, and I am a poor loser. So, Carl was an international master, and never mentioned it :-)
I was also quite surprised to see Ken Rogoff's comment on AI. He is very smart and I think very highly of his work, but I think his comment on AI is far off the mark.
What would be more impressive is a capped AI that can only look so many moves ahead and uses principles (such as say Nimzovich's opinionated twaddle) to win.