How are humans able to play better than computers? [chess]

[Avichal]

I don't know about the current situation but earlier humans were able to beat computers at chess. The first major loss was I guess Kasparov against Deep Blue.
But still I think humans still beat computers today.

How is this possible? Computers can think far further than we humans can. They can calculate many steps ahead whereas we cannot. Then what gives us the advantage over computers? How are we able to beat them?

 

[Jim Kata]

The answer to your question is no we can't. Computers can crush humans at chess, the best players included. The answer to why is simple, chess is only 8x8 64 squares. A computer can calculate most of the possibilities on a board this small. However, consider Go. The Go board is 19x19 and computers get destroyed by humans on it for the simple reason that a 19x19 board has a lot more possibilities than an 8x8 board does. However, computers on 9x9 Go boards are about even with the best human players, once again for the simple matter of computability. To your question on how humans win? Take Go again for example, computers can only calculate moves ahead, while humans understand harder to compute concepts like good shape, proper moves of intuition. In that, humans are doing complicated calculations on feeling that are not reading ahead per se, but is move recognition from experience. That is my opinion of it at least.

 

[Ryan Delaney]

As i have heard although this may not be reliable is that most chess systems have strength in short term strategy and piece placement while humans are better at long term strategies also humans that have played chess for long periods of time may be able to recall past positions for a hard to come by checkmate

 

[phinds]

Eventually, computers will be fast enough to crush humans at ANY game that is amenable to simplistic mechanical analysis.

Humans can't begin to do such analysis on games as complex as chess and go but compensate by having skills the computers don't (as has been pointed out above). Only recently have computers become to be able to overcome the human's skills and this ability will only increase, while human skills at those games are likely to not increase at all or at least not remotely as quickly as computer "skills" will.

 

[fluidistic]

A computer can calculate most of the possibilities on a board this small. However, consider Go. The Go board is 19x19 and computers get destroyed by humans on it for the simple reason that a 19x19 board has a lot more possibilities than an 8x8 board does.

This is not true at all. First of all, nowadays chess programs do not calculate any possible move at chess, else they would not see more than 3.5 chess moves (about 7 plies) after a minute of thinking time on home computers. Any strong program use many algorithms to get rid of many "useless" lines. I can easily get a depth of 20 using Stockfish (free chess engine in the top 5 of chess engines according to many rankings) under an old computer (2006). Deep Blue calculated all lines but could not see further than more or less 6 chess moves (average of 12 plies) in average despite running on a monster of a hardware.

Second, bots at go do not "get destroyed" that easily. The best bot constantly crushed top professionals with 4 handicap stones (including Takemiya Masaki as well as the last winner of a very famous championship; the games were not blitz but not as long as real tournament games either. Like 30 to 45 minutes for each player with additional byo yomi of 1 minute). The bot is ranked over 6 dan on KGS which is basically stronger than more than 97% of all go players on that website.

To your question on how humans win? Take Go again for example, computers can only calculate moves ahead, while humans understand harder to compute concepts like good shape, proper moves of intuition. In that, humans are doing complicated calculations on feeling that are not reading ahead per se, but is move recognition from experience. That is my opinion of it at least.

Once again absolutely false. In go, all bots who reached dan level use the Monte Carlo algorithm (and of course many other algorithms). Top bots calculate a lot of possible games mostly randomly generated (okay with some restrictions), up to the very last move. Then they do statisitcs and choose the move that leads to a higher win percentage. In that sense, a bot do not "calculate moves ahead" but goes up to the very last move several thousands times for each move.
To come back with chess, keep in mind that it's been a long time -several decades- people were interested in creating chess programs and that computer speed of calculation increased a lot since then. Computer programs went significantly stronger each time a new method was invented, examples: alpha beta pruning and null move heuristic.
By the way humans cannot win anymore a serious match vs a top bot such as Houdini, Stockfish or Critter. It's funny to analyze grand masters games and spot their blunders within 0.1 second. They are very, very far from top bots level running on nowaday home computers.

 

[PhizKid]

I thought those Go bots are amateur level and not pro level?

 

[fluidistic]

I thought those Go bots are amateur level and not pro level?

Right, 6 dan level is not pro level but strong amateur level. They beat pro with 4 stones handicap though. I'm guessing they should now try 3 stones (they have not tried yet). I'm sure the bots can win some games with 3 handicap since they win most of the games with 4.

Edit: Results of games between human (pros) and programs in go: http://www.computer-go.info/h-c/index.html.

 

[Jim Kata]

Right, 6 dan level is not pro level but strong amateur level. They beat pro with 4 stones handicap though. I'm guessing they should now try 3 stones (they have not tried yet). I'm sure the bots can win some games with 3 handicap since they win most of the games with 4.

Edit: Results of games between human (pros) and programs in go: http://www.computer-go.info/h-c/index.html.

I'm sure you realize a three stone handicap is a huge deal at the dan level. The ranks are not linear they get progressively harder to improve. So even if the best bot can play at a 6d kgs level that is still nowhere near Lee Sedol or Gu Li. I think computers won't reach that level for another 10 years.

 

[fluidistic]

I'm sure you realize a three stone handicap is a huge deal at the dan level. The ranks are not linear they get progressively harder to improve. So even if the best bot can play at a 6d kgs level that is still nowhere near Lee Sedol or Gu Li. I think computers won't reach that level for another 10 years.

Yes I do. What you said here is right, there's still a huge gap to reach any pro 9p in an even game.
But it's still strong for most humans. For example 6 dan is the top level in Argentina for humans. Enough to be the champion of many countries.