Do you think: Deep Blue (2014) vs. Human player

[Bubbleawsome]

This thread got advanced so fast. It's like watching people talk about the inner workings of CPUs. I'm lost. Carry on.

 

[Triumph]

Chess is a game of calculations. Right up a computer's alley. How about this instead, the CIA wants software that detects sarcasm in Twitter posts. Good luck with that.

http://www.huffingtonpost.co.uk/2014/06/04/secret-service-sarcasm-tw_n_5443528.html

 

[Gibsons]

Does skill level of a human player correlate with IQ? Since a computer doesn't have intelligence other than artificial, wouldn't that debunk the Skills = IQ?

Correlates yes, but not perfectly. There might be some chess savants with low IQ (anyone know?). Plus there are other things, practice being one. Then there's 'style' which might be a product of personality.

 

[smackababy]

As a programmer and AI hobbyist my first reaction is that those are all very hard problems to solve. Most sports games probably don't take it any further than generating random numbers and permuting them with attributes for player skill, environmental conditions, etc. Problem with that is the list above: you'll always notice the times that the PRNG spits out stupid numbers.

Actually modeling why a player attempts a certain action as a means to an end, and the results of that action, is way non-trivial.

The large problem with sports AI is the players all have different levels of everything. In a football game, for example, certain players will be able to recognize the lay type (run or pass) quicker than others. How do you quantify that? They give players an awareness stat and the difficulty in AI is how much they deviate from that (with the harder difficulties scaling it up, possibly). And then, you have to track that behavior among 11 players and make their actions accompany it.

 

[Markbnj]

The large problem with sports AI is the players all have different levels of everything. In a football game, for example, certain players will be able to recognize the lay type (run or pass) quicker than others. How do you quantify that? They give players an awareness stat and the difficulty in AI is how much they deviate from that (with the harder difficulties scaling it up, possibly). And then, you have to track that behavior among 11 players and make their actions accompany it.

The dividing line is the need for real-time response. Given enough samples and effort you can derive a model of just about any system, but the question is how long does it take to evaluate it and get the information you need for the next move? For a game like chess that limit is the amount of time that it is reasonable to wait for the other player to make a move. In a sports game it is real time, which is the most difficult constraint (well, the most difficult until you consider accelerated time).

So you have dual constraints in that the game developer won't be able to invest in creating that awesome detailed model, and the game engine won't be able to run it in real time.

 

[Blackjack200]

Getting back to chess...

The initial move by each player presents 324 potential board positions, and the second move produces roughly 105,000. As you can imagine, trying to evaluate every potential board position, even for a VERY fast computer, can quickly become impractical. Yes, you can prune, and these progams certainly do, but the early game still presents such a problem that the strongest programs rely on databases of thousands of grandmaster games to make sure that their openings are sound, and that they can get to a point where the number of potential board positions is far more manageable.

There are other considerations too, like the fact that the top human players focus on defeating other humans, not computers. There are techniques for playing strong computer opponents, but grandmasters don't waste time refining and practicing them because it is a side show compared to pvp.

In short, I think it's very likely that if you forced a computer to evaluate board positions from the very beginning (i.e. took away the database of grandmaster games), and if you had top grandmasters working to beat computers as hard as they work to beat other players, you would see players beating even the top rated programs.

 

[Markbnj]

In short, I think it's very likely that if you forced a computer to evaluate board positions from the very beginning (i.e. took away the database of grandmaster games), and if you had top grandmasters working to beat computers as hard as they work to beat other players, you would see players beating even the top rated programs.

But it would be unfair to remove the database of grandmaster games because the human opponents have that as well. Not as complete as a static db, certainly, but then they have been able to absorb play principals from a lifetime of studying those games in ways that the computer program can't even encode. It may be they are in an even better position than the computer with its comparatively context free db of moves.

 

[Blackjack200]

But it would be unfair to remove the database of grandmaster games because the human opponents have that as well. Not as complete as a static db, certainly, but then they have been able to absorb play principals from a lifetime of studying those games in ways that the computer program can't even encode. It may be they are in an even better position than the computer with its comparatively context free db of moves.

Right, but what we're trying to figure out is if the principals and intuition of the grandmaster exceed the intelligence of the algorithm.

Grandmasters are more comfortable on certain opening lines, and understand what tends to happen on those lines with accurate play from both sides, but they generally don't memorize sequences of moves. It is similar to the way they play middle games, where various patterns of attack can be constructed and anticipated.

 

[WHAMPOM]

With today's computing power, do you think a human would have a chance competing against a computer in chess?

If you can exploit a flaw in its programming. On concentration the computer repeated a wrong answer given by another contestant, a human would take this data (Oh fuck!) and give an alternate (in the form of a question) answer. It was not given input from the other players.

Chess, stalemate in every game until the human makes a mistake. Human wins after groupies hack the machine. You must have a chess program on your computer, how often do you win?