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

[Carson Dyle]

It just has to be able to search far enough ahead that its increased awareness of the possible outcomes trumps the human player's better judgement of individual board positions.

And the further ahead that it can search, the bigger the advantage. Which is why well-written software will beat anything using only pure force. You want to stop exploring a line as soon as possible, so that you can examine as many possible lines in a given amount of time. But by the same token, you don't want to cut off the search prematurely.

 

[smackababy]

The computer obviously doesn't have to search all possible chess moves, just those that are possible from a given starting board state. Even that is too many moves to search to the end of the tree for every possibility. But then it doesn't have to. It just has to be able to search far enough ahead that its increased awareness of the possible outcomes trumps the human player's better judgement of individual board positions. I think that point was reached years ago.

This. Search 10 or so moves ahead from that point, and assigning a value to each isn't so taxing.


That is where Go AI's have failed. The assigning of value for every move is extremely difficult.

 

[GagHalfrunt]

LOL.

There are MANY MORE combinations, like orders of magnitude more, in chess than atoms in known universe.

To compute all of them is just a brute force calculation. Human brain/strategy does not work like that. Especially while doing chess.

There are several computational problems that computers can't touch unless we get quantum ones, and this is not given.

Traveling salesman, have you heard of it?

Meteorology, fluid dynamics problems will kill any known computer quickly.

I find your answer pitiful.

Speaking of pitiful....

The computer doesn't brute force every possible line of play. It examines all possible moves from a given position a couple of moves deep, discards 95% of them as being useless and then drills down through the lines that have a higher chance to succeed. It's exactly what a human does except the computer can cover FAR more possible lines and check them to a much greater depth. The point where computers and humans can play even is long long long gone, it was more than 15 years ago when the best programs could outplay the best humans. Now the programs get more sophisticated as to how they determine which lines to pursue and they have a shitload more power to check those lines millions of branches deep. A computer doesn't need to be able so solve the earth's weather to beat a puny human brain at chess.

 

[norseamd]

http://en.wikipedia.org/wiki/Chess#Mathematics_and_computers


You tell me.

my opinion is that the hardware of today is capable of driving at least somewhat advanced ai

you just need the right software code to run the hardware

 

[Gibsons]

LOL.

There are MANY MORE combinations, like orders of magnitude more, in chess than atoms in known universe.

To compute all of them is just a brute force calculation. Human brain/strategy does not work like that. Especially while doing chess.

There are several computational problems that computers can't touch unless we get quantum ones, and this is not given.

Traveling salesman, have you heard of it?

Meteorology, fluid dynamics problems will kill any known computer quickly.

I find your answer pitiful.

One estimate is 10^123, which is just too big for me to even comprehend well.

http://en.wikipedia.org/wiki/Shannon_number

 

[Shlong]

I think I remember reading Hikaru Nakamura (#1 US player, top 10 in the world) say that the computer programs on his computer could beat him regularly and could do the same against Magnus Carlsen.

 

[Noo]

let's say the cpu somehow gets fried during a match. Would that be considered a forfeit?

Same as a human dropping dead with a heart attack during a match?

 

[Rakehellion]

let's say the cpu somehow gets fried during a match. Would that be considered a forfeit?

You can easily put a program into another computer.

 

[BladeVenom]

Yet they still haven't made a good AI for the Total War games.

 

[Hayabusa Rider]

One estimate is 10^123, which is just too big for me to even comprehend well.

http://en.wikipedia.org/wiki/Shannon_number

It's an EXPTIME-complete problem. Only supernatural beings could get it.

 

[Gibsons]

Yet they still haven't made a good AI for the Total War games.

Computer strategy games can leap past chess in depth/complexity very quickly. Take chess: add a tech tree, production and economy,, make the board 100x bigger, randomly generated map/tiles, with 5 different tile types instead of two.

That's four (or five) things that each alone make it exponentially more complex than chess. But it's a simple game by modern standards. So, I think a good AI for a really complex game is very hard to make.*

But yeah, I gripe about poor AIs too. :\

*I'm just talking hypothetical, not Total War specifically, thought it might apply, I don't know. I've barely scratched the surface of that franchise.

 

[Ruptga]

Computer strategy games can leap past chess in depth/complexity very quickly. Take chess: add a tech tree, production and economy,, make the board 100x bigger, randomly generated map/tiles, with 5 different tile types instead of two.

That's four (or five) things that each alone make it exponentially more complex than chess. But it's a simple game by modern standards. So, I think a good AI for a really complex game is very hard to make.*

But yeah, I gripe about poor AIs too. :\

*I'm just talking hypothetical, not Total War specifically, thought it might apply, I don't know. I've barely scratched the surface of that franchise.

That's all true, but I suspect that [foil hat] there isn't a lot of demand for game AI because companies see AI as a third wheel on a bike. They want humans to play with humans because that increases their products' values through more highly integrated game communities and larger playerbases since even loners will recruit and get together over a good game. [/foil hat]

 

[Gibsons]

That's all true, but I suspect that [foil hat] there isn't a lot of demand for game AI because companies see AI as a third wheel on a bike. They want humans to play with humans because that increases their products' values through more highly integrated game communities and larger playerbases since even loners will recruit and get together over a good game. [/foil hat]

Maybe true now, I agree the motivation to make a good single player only AI isn't that great these days. But bad AIs have existed since the much, much simpler single player only 4x games, like Warlords II level stuff. Yes, partly due to limited cpu/memory back then as well... still.

 

[NikolaeVarius]

Computer strategy games can leap past chess in depth/complexity very quickly. Take chess: add a tech tree, production and economy,, make the board 100x bigger, randomly generated map/tiles, with 5 different tile types instead of two.

That's four (or five) things that each alone make it exponentially more complex than chess. But it's a simple game by modern standards. So, I think a good AI for a really complex game is very hard to make.*

But yeah, I gripe about poor AIs too. :\

*I'm just talking hypothetical, not Total War specifically, thought it might apply, I don't know. I've barely scratched the surface of that franchise.

Eh, good game AI is hard to separate from bad AI.

The simplest of AI routines can seem brilliant while the most complex and advanced AI just makes you think the computer is cheating.

The problem is making fighting against the AI fun, and having a super smart AI usually doesn't make it fun because you always have that feeling the computer is cheating.

 

[darkxshade]

A computer has yet to beat me... everytime it comes close, I turn it off. Now when they can do the same to us, that's when things get interesting... and scary.

 

[Apex]

with Go, computers are still having problems beating amateurs let alone masters.

http://www.wired.com/2014/05/the-world-of-computer-go/

A computer stands no chance against Grand Master Yoda, apparently.

 

[Markbnj]

That's all true, but I suspect that [foil hat] there isn't a lot of demand for game AI because companies see AI as a third wheel on a bike. They want humans to play with humans because that increases their products' values through more highly integrated game communities and larger playerbases since even loners will recruit and get together over a good game. [/foil hat]

There might be some truth to that as well, but the reality is it is a much, much harder problem to solve. The core of most game strategy engines is to run a min-max evaluation of game position and choose the next move based on what good outcomes it leads to. Imagine how much harder it is to do that for a strategy game than a relatively simple game like chess, both in terms of evaluating the quality of the current position, and determining what moves are possible from that position.

 

[Ruptga]

Yes, I did say that's all true. It's hard to make good AI and there's not a lot of incentive to building them, so it doesn't happen very often.

How to half-ass a game AI:
1. hardcode its opening moves, make them somewhat sensible but unimaginative. Use one such template for each unique faction, or just throw up your hands and have all factions open with the same template.
2. give the AI a handful of choices to make about its broad strategy going forward, it should choose according to either the faction it controls or by RNG roll, and all strategies should be largely hardcoded with a little room for adapting to circumstance.
3. have the AI initially choose its low-level tactics by RNG, then vary them by responding somewhat (+/-50%) to variable thresholds such as resource production, discovering a resource to exploit, or enemy unit composition or density.
4. add a difficulty slider; tie the slider to AI health, damage, resource production, morale, etc. Overall behavior should be unchanged.

 

[Slew Foot]

Yes, I did say that's all true. It's hard to make good AI and there's not a lot of incentive to building them, so it doesn't happen very often.

How to half-ass a game AI:
1. hardcode its opening moves, make them somewhat sensible but unimaginative. Use one such template for each unique faction, or just throw up your hands and have all factions open with the same template.
2. give the AI a handful of choices to make about its broad strategy going forward, it should choose according to either the faction it controls or by RNG roll, and all strategies should be largely hardcoded with a little room for adapting to circumstance.
3. have the AI initially choose its low-level tactics by RNG, then vary them by responding somewhat (+/-50%) to variable thresholds such as resource production, discovering a resource to exploit, or enemy unit composition or density.
4. add a difficulty slider; tie the slider to AI health, damage, resource production, morale, etc. Overall behavior should be unchanged.

EA Sports AI:
1. Hockey- computer goalie stops every shot
2. Baseball- all balls hit by human player go straight to a defender
3. Football- Human players gets a penalty on every positive play
4. Basketball-Human players shoot bricks

 

[Gibsons]

EA Sports AI:
1. Hockey- computer goalie stops every shot
2. Baseball- all balls hit by human player go straight to a defender
3. Football- Human players gets a penalty on every positive play
4. Basketball-Human players shoot bricks

hehe
http://nesn.com/2013/09/report-nba-...-that-made-them-shoot-bricks-at-end-of-games/

 

[Hayabusa Rider]

I'd disagree about incentive for AI development. Yes, for gaming purposes that may be true however you can bet DARPA and the like are keen to have autonomous machines and the problem has so far been intractable. Looking at complex simulation games for example shows that no one has made much headway and that seems to be due in large part no qualitative improvements in processes. "Faster" is not "smarter". In any event it's hard to imagine that our relative protozoan reasoning level home computers can by any means made competitive with our organic minds for complex scenarios. I suspect that it may be fundamentally impossible.

 

[tmc]

IMHO computers have these major advantages:

1. don't get tired or in a bad form.
2. don't make [silly] mistakes/miscalculations.
3. the most important one - they don't feel fear. in one of the matches i saw live, the champion had a chance to win, but will have to allow two opponent rooks in his second rank (which is usually winning for the opponent and very scary to allow), and didn't make that move and drew. humans will usually think twice before allowing such combinations, but computers will make that move.

 

[Markbnj]

EA Sports AI:
1. Hockey- computer goalie stops every shot
2. Baseball- all balls hit by human player go straight to a defender
3. Football- Human players gets a penalty on every positive play
4. Basketball-Human players shoot bricks

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.

 

[Noo]

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?

 

[Net]

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

absolutely. in regards to the threads title, deep blue would loose a lot actually. the chess engines today are a lot better then deep blue and they are still not unbeatable by humans.

the game of chess still hasn't been solved (there is no engine that finds a perfect strategy).