Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process.
I had to read this statement 3 times before it hit me: What's true in chess is also often true in business. A little background...
I recently wrote a forecasting system for a company that processes 7 million orders per year. Worse, this company was the merger of two other companies, each of which did forecasting differently. One had a very expensive Oracle based "strong comptuer" that calculated almost everything and told the planners exactly what to do. The other just dumped data into Excel files and teams of "strong humans" manipulated them until they intuitively worked out the best plan. Neither team could believe the way the other team worked.
The system I wrote using guidance from both teams turned out to be "weak human + machine + better process" which leveraged the strengths and minimized the weaknesses of the two extremes. But I didn't realize that until I read this article.
Thank you Garry Kasparov. As much as I love chess just for the sake of playing it, it's nice to understand how it's thinking applies to other stuff as well.
Yes! I thought that was one of the key insights of the article too.
A similar example that shows the strength of great organisation is a famous 1999 online chess match known as Kasparov versus the World. In that game Irina Krush organized the world team well enough to take Kasparov to 62 moves and nearly won the game.
To quote him:
It is the greatest game in the history of chess. The sheer number of ideas, the complexity, and the contribution it has made to chess make it the most important game ever played.
Fascinating article.
I was wandering what it means to democracy. Can we use those ideas in collaboration to have a better decision making in real governments?
Linear programming was hoped to have been a strong enough framework to replace the market-based resource allocation system. Unfortunately, no single centrally managed computer could outperform a network of autonomous free participants. Modern government is the realization that there can be no prosperity without free agents, and that propaganda can persuade these free agents to acquiesce to being governed.
Active tournament player here, FIDE 2100. They key point to remember here is that computers are able to avoid the tactical pitfalls that humans cannot. In other words, they can see all the possible outcomes within a given number of moves. The fewer the pieces, the farther ahead they calculate and play perfectly. Indeed there are databases of up to 7 pieces that are used for perfect play.
Humans cannot see as far as a computer, but they don't need to do so. Grandmasters recognize over 10,000 common positions (after the opening phase) and their strategies based on experience and study. This allows them "know" the probable outcome of a given move far beyond that of a computer. The problem arises when the complexity of a position grows to a point that tactics > position. Human players avoid these positions when playing computers. Strong human players aim for these positions when playing a weaker opponent. And computers can be programmed for "anti-human" strategies that create these complications.
Kasparov was aware of all these variables, but underestimated the tactical ability of the machine. And, ultimately failed, because he fell for a well-known trap in the opening during the last game. The computer was programmed with a database that included the trap. Kasparov was too - with his memory - but unlike silicon, simply forgot for a moment and made the fatal move that sealed the match.
What does the mean for business? A superior long-term strategy that anticipates short-term problems trumps a short-term strategy that knows all the immediate outcomes but neglects to consider the end game.
I liked this article. Especially the end bit on poker since teaching my laptop to play poker (NL Holdem) is a bit of hobby of mine. Its learning very well and is already much better than me (I am only above average from a random sample of humans, so relatively weak).
My goal is to have it at least not lose badly vs my professional high stakes friend heads up. It does quite well against weaker players... Its not a rule based player, and my favourite bit is I can actually see it get better per game. It makes mistakes and learns from them. Poker is a very good place to try machine learning algos and also make practical use of game theory since it can't be effectively brute forced. Ive been iteratively bootstrapping newer more complex AI components on top of the previous bases. wrote it in F#.
There are so many people doing this that I can't play Holdem online anymore! I'm not good enough to beat the decent bots today, they have improved massively over the last 5 years.
The dreams of creating an artificial intelligence that would engage in an ancient game symbolic of human thought have been abandoned. Instead, every year we have new chess programs, and new versions of old ones, that are all based on the same basic programming concepts for picking a move by searching through millions of possibilities that were developed in the 1960s and 1970s.... Like so much else in our technology-rich and innovation-poor modern world, chess computing has fallen prey to incrementalism and the demands of the market.
Here Kasparov misses the mark. Should airplanes flap their wings like birds? Should bicycles have feet? What matters in engineering is the quality of the result, not how the result is achieved.
What's more, the demands of market that Kasparov disparages actually help drive innovation, for if someone were to invent a novel chess program that wipes the floor with all the other programs on the market, is there any doubt it would outsell them?
I liked your analogies, but it's difficult to compare the flapping of birds' wings and the movement of our feet to the extreme complexities of the human brain. In the first two cases, humans have invented more effective ways of doing a task than the biological methods. These examples are elementary compared to the human brain, and they are most likely not to be the most effective ways of doing the tasks - they are only more effective then the current methods. Perhaps if we dissected the processes to a more rudimentary level, we could revise the systems and make them even more effective. However, with simple processes like these it is probably not economically worth it as Kasparov said.
As a result, when you apply these analogies to the human intuition and more complex processes, you are confronted with a dilemma. Our society is reaching a limit where our brute-force methods of recreating systems are no longer working. We are looking too much at the face value of these systems, and thus reaching problems with efficiency and further innovation. In my humble opinion, Kasparov is trying to say that the demands of the market have caused this behavior which benefits in short term, but limits us in the long run. Taking risks to work on a longer-term project of recreating systems at a more fundamental level is what keeps innovation going, and it's time we take it to a new level with respect to human intuition.
In the first two cases, humans have invented more effective ways of doing a task than the biological methods.
Kasparov acknowledges that computers are superior to unassisted humans at playing chess, so how is the third case (computer chess) different?
[B]rute-force methods of recreating systems are no longer working.
On the contrary, brute-force algorithms are working better and better as a result of the trend currently known as Moore's law. The faster the hardware, the less intelligent the software running on it has to be to do the same job.
The fact the human + weak machine far outplays a strong machine should show you that how the result is achieved matters a lot.
The ability of a human to think better than a machine puts them in a completely different classes.
They do not accomplish the same result - they think and calculate different things. In certain tests they may accomplish the same result, but not in all tests.
If they did then a strong machine plus a weak machine should do better than a strong machine alone - but it doesn't.
This is a great article. I especially liked the idea that
> humans today are starting to play more like computers
ie, the advent of computers that are prejudice-free pushed human players into a whole different way of playing; from just the previous sentence:
> a move isn't good or bad because it looks that way or because it hasn’t been done that way before. It’s simply good if it works and bad if it doesn't.
What I don't like as much is the constant complaint that we're not trying hard enough to make computers think like humans, that Deep Blue and all its ilk dumbly calculate millions of moves without really "thinking".
This reminds me of the efforts that are made to have robots look like humans (walk like us for example).
What's the point? A Roomba doesn't have legs and yet it's better than me at vacuum cleaning (meaning, we both suck, but I suck more).
We wouldn't want pliers to be shaped like a hand; computers are pliers for the brain: they extend it, but they don't need to be shaped like it.
It's encouraging to see that Kasparov framed his famous tournament as "programmer vs. machine" rather than "man vs. machine." In hindsight, it's easy to chalk Deep Blue's victory up to computational inevitability, but there was some seriously cool engineering going on behind the scenes. Deep Blue was a purpose-built chess machine: a 30-node RS/6000 supercomputer controlling 480 custom chess chips. It was the culmination of nearly a decade of research at IBM.
The Kasparov v. World match was a democratic process in the modern understanding in that people were offered equivalent choices (almost equally strong), selected by the experts, with little public understanding of the deeper ramifications of the decisions.
In a democracy, people are asked to make decisions in fields that really need expert knowledge to discern, but fortunately (or some say, 'unfortunately), these choices are pre-selected by a group of experts. These people are both voters (where the experts are "public opinion leaders) or legislators (where the experts are lobby firms).
Yeah, I remember this result. I saw Kasparov before his match with Deep Blue (my dad knows him). I wish there was a rematch against him and computer. And it's true, process is very important -- not just in chess but elsewhere. And so is making use of human creativity!
I don't think that's the way to look at computer chess. When you play against a computer you have to equate that with playing against a really smart collective (the programmers of the computer) that have access to as much stored information about chess as they could get their hands on with perfect recall that has the limited ability to see in to the future on all your possible moves, countermoves and so on until you hit the limits of computational complexity allowed by the time constraints of the game.
That's a fairly uneven match, and saying you 'lost' in that situation is as meaningful as saying a human 'lost' a run over 100 meters against a car.
What it does give you is a measurement of the complexity of the tasks that a bunch of dedicated programmers (all of which are probably collectively not as good at chess as your average grand master but still better than most normal players) and a sufficiently fast computer can accomplish, and I think that it is a tribute to our brains that it takes such computational violence with elements of libraries of openings and brute forcing until a 'stable' state has been reached to match our humble 3 pounds of wetware running at a power consumption that would make most netbooks look pretty silly.
That 'pointless game' is not glorified by society but is enjoyed by those that play it because it is not a pointless game but actually helps you to understand stuff and develop your skills at strategic and analytical thinking.
Chess has never made anybody feel 'stupid', no more than that any hammer ever killed someone.
People make other people feel stupid, mostly the people with a little more skill than you. People do stuff like that all the time, it's petty behaviour but you can't blame the game of chess for that.
The harmful part here is in part the person that allows the judgment of others to have such devastating effect on them, it takes two to play the 'your stupid' game, and if you let your skills at chess and others' opinions of you weigh that heavily then I would suggest you either lighten up or you get better at chess ;)
The thing I personally don't like about chess (and which is my main reason for no longer playing it competitively) is the people that memorize book openings and use that as a substitute for original play. I don't want to play some canned opening to the mid game, I want to enjoy a game with the person on the others side of the board.
My standard solution to that is to make a really bad but totally off the books opening. This puts me at an instant disadvantage but it certainly makes the game more interesting ;)
edit: regarding the 'stupid', here is a quote from the article, arguably from the worlds very best chess player straight to you:
"Excelling at chess has long been considered a symbol of more general intelligence. That is an incorrect assumption in my view, as pleasant as it might be."
You are missing some or all of the point here. Computers are not 'better' at chess than humans, they are simply faster and have access to more and more perfect forms of memory.
Humans learn a few simple rules and combine that with a motivation to win and suddenly 'know how to play chess' with some ability after a few hours, days or weeks. To become a master at the game takes a lot more study but in general people can learn how to play chess with some ease.
You are the one who is taking it seriously! Games are meant to be enjoyed, not to be taken too serious and even the big 'names' from chess play the game as a way to pit their wits against others. The way they arrive at their solutions is totally unrelated to the way computers arrive at theirs.
Imagine if you showed up to a chess contest armed with a veritable library of chess books, a history of all the grand master games ever played, their openings and evaluation of the mid games that came out of those openings and a near infinite number of assistants that are willing to play your current board position through large number of variations to see the possible outcomes of potential moves.
I think the key to continued enjoyment in spite of the fact that computers are 'better' at this is that they indeed arrive by a way that between humans would be considered cheating.
Be impressed by how much resistance our humble brains can put up in the face of such an onslaught and realize that we are the ones that are 'better' at it because no computer that ever got 'taught' chess was any good at all. It takes the combined elements of vast pre-programmed storage and brute force to make it to the higher classes in chess and that has nothing to do with 'playing chess'.
I agree with most of your point, but have to quibble with the assertion that computers are not "better" at chess. Deep Blue was able to beat arguably the greatest chess player ever. Computer chess programs today are far stronger than Deep Blue, and could almost certainly defeat any human player. The way they decide moves is different that humans, but it is also objectively better at achieving the goal (of winning).
At a power budget that literally dwarfs the power expended by the human. At the current state of technology it is an uneven match by a significant margin.
Computers are 'better' not when they reach the goal via the same path as humans do but when all other factors being equal (size, power consumption, access to pre-programmed data and so on) they beat the best human player. And I don't think we're there yet, deep blue was not a universal computer by any stretch of the imagination.
I don't believe the power budget dwarfs a human's, at least not in a measurable way. Resources online suggest that the human brain probably uses about 20 watts. Small laptops don't use much more than that, I don't think, and a small laptop can probably beat any human at chess (since Kramnik lost to a 2006 version of Fritz running on a Core 2 Duo with a crippled tablebase.)
That's an impressive feat, what I remember of the 'deep blue' match was that it was a special purpose rig with thousands of FPGAs wired up with chess specific circuitry.
Iirc the Kramnik match was not exactly a show of strength for Kramnik, also, he is not the current ruling world champion in chess (though he was when he lost that match, but that's why we keep on playing tournaments, the current champion is this guy: http://en.wikipedia.org/wiki/Viswanathan_Anand).
That said, Fritz is imo a very impressive piece of software, much more so than deep blue ever was.
edit: I just looked at the wikipedia entry, the 'crippling' was with respect to the end-game tables, typically that's not a really big deal unless you have the bad luck to end in an end game where you need those tables, in that case such a game usually ends in a draw. In none of the games that were played that seems to have made much difference, and Kramnik messing up was what sealed the match.
About the tablebases, I disagree; having a more extensive tablebase means that you can short-circuit calculations earlier in the game (because you immediately know the evaluation if your calculation results in a tablebase position.) So it helps even if you don't actually end up playing into a six-or-less-piece endgame. It's obviously hard to say, but it's certainly conceivable that it hurt Fritz's play in a few of the games which wound up being drawn. I didn't mention it, but another condition of the match greatly favoring Kramnik was that the computer's opening book was completely open to him as well -- he could view all of Fritz's move weightings and statistics during the openings (an attempt at "fairness" given the computer's theoretically more extensive knowledge.)
If you're not aware, Fritz is also no longer at the top of the heap in computer chess. http://www.rybkachess.com/
What computers 'can do' is limited by our imagination and our ability to express ourselves. As Kasparov points out in the article as soon as IBM had the publicity they wanted (to win from the best human chess player at any cost) they scrapped the project.
But the contest was meaningless the way it was posed anyway. Personally I think that the best way to deal with the situation would have been to 'handicap' the computer to use the same amount of resources that the human has access to, so a given power budget and no access to pre-programmed libraries. That would make the contest much more interesting, and more importantly would drive forward our thinking about solving this sort of problem in an intelligent way instead of using the sledge-hammer of brute force.
I had to read this statement 3 times before it hit me: What's true in chess is also often true in business. A little background...
I recently wrote a forecasting system for a company that processes 7 million orders per year. Worse, this company was the merger of two other companies, each of which did forecasting differently. One had a very expensive Oracle based "strong comptuer" that calculated almost everything and told the planners exactly what to do. The other just dumped data into Excel files and teams of "strong humans" manipulated them until they intuitively worked out the best plan. Neither team could believe the way the other team worked.
The system I wrote using guidance from both teams turned out to be "weak human + machine + better process" which leveraged the strengths and minimized the weaknesses of the two extremes. But I didn't realize that until I read this article.
Thank you Garry Kasparov. As much as I love chess just for the sake of playing it, it's nice to understand how it's thinking applies to other stuff as well.