1002: "Game AIs"

This forum is for the individual discussion thread that goes with each new comic.

Moderators: Moderators General, Prelates, Magistrates

User avatar
Klear
Posts: 1965
Joined: Sun Jun 13, 2010 8:43 am UTC
Location: Prague

Re: 1002: "Game AIs"

Postby Klear » Thu Jan 12, 2012 1:17 pm UTC

meerta wrote:
Klear wrote:
Stilgar wrote:
TranquilFury wrote:
Klear wrote:Actually, playing a game is about having fun. Everything else, including the need for rules and win conditions is optional and subjective.

Depends what your reason for playing the game is, a lot of people make their living from starcraft, especially in korea.


I once watched an interview with Veselin Topalov ( world chess champion in 2005 - http://en.wikipedia.org/wiki/Veselin_Topalov ) and he was asked if he considered chess to be a game, an art or a sport. He said that for the masses chess is just a game, for amateurs that really understand high-level games it is an art and for him it is a sport because he plays for results.


Exactly. IMO (and obvioulsy your definition may vary) a game is played for pleasure. If you do it for money or whatever, it stops being a game.


That premise pretty much undermines half this comic, and very stipulative about how we use the word "game" to boot.


Yes is does. And I stand by my claim that games are and always have been primarily means of having fun and victory conditions are not required (as evidenced by games that don't have any).

So poker is not a game? Also, I am sure playing for pleasure and money simultaneously is kinda possible.


If you are playing poker for the sole reason to make money, is it a game for you?

Edit: OK, I didn't mean to say that if you get money out of it it ceases to be a game. What I meant was that when you only do it for money it's not a game anymore.
Last edited by Klear on Thu Jan 12, 2012 1:22 pm UTC, edited 2 times in total.

threedognice
Posts: 15
Joined: Mon Dec 19, 2011 10:07 am UTC

Re: 1002: "Game AIs"

Postby threedognice » Thu Jan 12, 2012 1:20 pm UTC

Klear wrote:If you are playing poker for the sole reason to make money, is it a game for you?

Can you imagine work being more than a source of money income?
Last edited by threedognice on Thu Jan 12, 2012 1:23 pm UTC, edited 2 times in total.

User avatar
Klear
Posts: 1965
Joined: Sun Jun 13, 2010 8:43 am UTC
Location: Prague

Re: 1002: "Game AIs"

Postby Klear » Thu Jan 12, 2012 1:21 pm UTC

threedognice wrote:
Klear wrote:If you are playing poker for the sole reason to make money, is it a game for you?

Can you imagine work being more than a source of money income?


See the edit in my previous post in response to your edit of your previous post:

Edit: OK, I didn't mean to say that if you get money out of it it ceases to be a game. What I meant was that when you only do it for money it's not a game anymore.

threedognice
Posts: 15
Joined: Mon Dec 19, 2011 10:07 am UTC

Re: 1002: "Game AIs"

Postby threedognice » Thu Jan 12, 2012 1:23 pm UTC

Edit: OK, I didn't mean to say that if you get money out of it it ceases to be a game. What I meant was that when you only do it for money it's not a game anymore.

It is not your hobby anymore, but it is still a game.

Progamers are gamers.

User avatar
Klear
Posts: 1965
Joined: Sun Jun 13, 2010 8:43 am UTC
Location: Prague

Re: 1002: "Game AIs"

Postby Klear » Thu Jan 12, 2012 1:32 pm UTC

threedognice wrote:
Edit: OK, I didn't mean to say that if you get money out of it it ceases to be a game. What I meant was that when you only do it for money it's not a game anymore.

It is not your hobby anymore, but it is still a game.

Progamers are gamers.


I already know your position. No need to repeat it. Our definitions differ, that's all.

DukeTwicep
Posts: 3
Joined: Mon Oct 31, 2011 3:46 pm UTC

Re: 1002: "Game AIs"

Postby DukeTwicep » Thu Jan 12, 2012 1:41 pm UTC

Klear wrote:
threedognice wrote:
Edit: OK, I didn't mean to say that if you get money out of it it ceases to be a game. What I meant was that when you only do it for money it's not a game anymore.

It is not your hobby anymore, but it is still a game.

Progamers are gamers.


I already know your position. No need to repeat it. Our definitions differ, that's all.

I have to agree with threedognice. A game is a game no matter how individuals choose to play it, it only seizes to be a game when it no longer fits the description of a game. Of course, maybe games can seize to be games to individuals if they stop regarding the game as a game, but this has no bearing on if it is a game or not objectively.

threedognice
Posts: 15
Joined: Mon Dec 19, 2011 10:07 am UTC

Re: 1002: "Game AIs"

Postby threedognice » Thu Jan 12, 2012 1:54 pm UTC

Klear wrote:
threedognice wrote:
Edit: OK, I didn't mean to say that if you get money out of it it ceases to be a game. What I meant was that when you only do it for money it's not a game anymore.

It is not your hobby anymore, but it is still a game.

Progamers are gamers.


I already know your position. No need to repeat it. Our definitions differ, that's all.

Maybe, but your position is not that clear. There is a definition of the word “game,” and it is not defined in some kind of imaginary way — no, it is very real, and exists. Just like my position, there is no need to repeat it … but I will, anyway. It says: “a form of play or sport, esp. a competitive one played according to rules and decided by skill, strength, or luck.”

Also, if game ceases to be a game, what it becomes exactly? A shameful activity of some sort maybe? Like, say, “they have played a lot of shameful activities of chess today”? That is kinda weird, I think.

User avatar
Klear
Posts: 1965
Joined: Sun Jun 13, 2010 8:43 am UTC
Location: Prague

Re: 1002: "Game AIs"

Postby Klear » Thu Jan 12, 2012 2:12 pm UTC

threedognice wrote:Maybe, but your position is not that clear. There is a definition of the word “game,” and it is not defined in some kind of imaginary way — no, it is very real, and exists. Just like my position, there is no need to repeat it … but I will, anyway. It says: “a form of play or sport, esp. a competitive one played according to rules and decided by skill, strength, or luck.”


Perfect. Now how about a definition of "play" (which is a part of the definition you quoted)? "Engage in activity for enjoyment and recreation rather than a serious or practical purpose."

Edit: Anyway, the conflict here comes from the fact that I consider the goal of recreation to be the essence of games, why you give more weight to the formal appearance, competitiveness and such. It's not like either of us is wrong, we just both have different definitions, even though they both fall under the general one you quoted.

threedognice
Posts: 15
Joined: Mon Dec 19, 2011 10:07 am UTC

Re: 1002: "Game AIs"

Postby threedognice » Thu Jan 12, 2012 2:49 pm UTC

Klear wrote:Perfect. Now how about a definition of "play" […]

OK, okay, okay. You got me…………………………………………sort of. BUT! one thing is still bugging me: what do professional cellists do with their cellos to make a living?

User avatar
Klear
Posts: 1965
Joined: Sun Jun 13, 2010 8:43 am UTC
Location: Prague

Re: 1002: "Game AIs"

Postby Klear » Thu Jan 12, 2012 3:09 pm UTC

threedognice wrote:
Klear wrote:Perfect. Now how about a definition of "play" […]

OK, okay, okay. You got me…………………………………………sort of. BUT! one thing is still bugging me: what do professional cellists do with their cellos to make a living?


Uh, sell them? Play chess with them? Somebody, help me with this!

meerta
Posts: 63
Joined: Mon Aug 10, 2009 11:25 am UTC

Re: 1002: "Game AIs"

Postby meerta » Thu Jan 12, 2012 3:13 pm UTC

Klear wrote:
threedognice wrote:Maybe, but your position is not that clear. There is a definition of the word “game,” and it is not defined in some kind of imaginary way — no, it is very real, and exists. Just like my position, there is no need to repeat it … but I will, anyway. It says: “a form of play or sport, esp. a competitive one played according to rules and decided by skill, strength, or luck.”


Perfect. Now how about a definition of "play" (which is a part of the definition you quoted)? "Engage in activity for enjoyment and recreation rather than a serious or practical purpose."

Edit: Anyway, the conflict here comes from the fact that I consider the goal of recreation to be the essence of games, why you give more weight to the formal appearance, competitiveness and such. It's not like either of us is wrong, we just both have different definitions, even though they both fall under the general one you quoted.


Just to clarify this. In the thread of discussion I was following, MostlyNormal posted that chess can't really be solved because what computers do isn't the same as what we do when we play games; a couple of posters including me responded that the ability to play a game like chess is essentially to be able to play by the rules. Klear then came in to say that a game isn't a game really unless its played for fun.

So effectively Klear is saying that because computers don't play chess in the same way as humans, they don't play it at all - they're not aware of what they're doing and they don't have fun. I agree there is a conceptual distinction between the way we play and the way computers play: as long as this is understood (by those who accept it) it doesn't matter what words we use. Any semantic disagreement doesn't affect the discussion - either side can rephrase things as necessary.

But I was saying the comic was partly about computers vs. humans ability, and how far computers have gone in "solving" various games. Because "solved" is just a computer science term, all that's required to talk about computers' skill level at chess is that computers follow the same rules as humans, albeit mechanically, and that they are the rules of chess. This is true whatever you call what computers do when they appear to play chess. To say "playing a game" doesn't apply to this doesn't make the phenomenon go away. It might just make it harder to discuss. (On the other hand it might make the discussion more precise.)

As for games, in Philosophical Investigations Wittgenstein uses the word to illustrate the way "language" has no single essence, maybe helpful?

"66. Consider for example the proceedings that we call 'games'. I mean board-games, card-games, ball-games, Olympic games, and so on. What is common to them all? -- Don't say: 'There must be something common, or they would not be called "games" '-but look and see whether there is anything common to all. -- For if you look at them you will not see something that is common to all, but similarities, relationships, and a whole series of them at that. To repeat: don't think, but look! --"
Last edited by meerta on Thu Jan 12, 2012 5:36 pm UTC, edited 7 times in total.

threedognice
Posts: 15
Joined: Mon Dec 19, 2011 10:07 am UTC

Re: 1002: "Game AIs"

Postby threedognice » Thu Jan 12, 2012 3:16 pm UTC

Klear wrote:
threedognice wrote:
Klear wrote:Perfect. Now how about a definition of "play" […]

OK, okay, okay. You got me…………………………………………sort of. BUT! one thing is still bugging me: what do professional cellists do with their cellos to make a living?


Uh, sell them? Play chess with them? Somebody, help me with this!

You can’t argue with objectivity in here!

A game is a game.

User avatar
Klear
Posts: 1965
Joined: Sun Jun 13, 2010 8:43 am UTC
Location: Prague

Re: 1002: "Game AIs"

Postby Klear » Thu Jan 12, 2012 3:34 pm UTC

meerta wrote:Just need to clarify this. [...]


That pretty much sums it up, except that I'm not agreeing with MostlyNormal's post, I was just disagreeing with the reply of SirMustapha, who said that games are all about following rules and trying to win. Teaching computers to play games for leisure makes no sense (at least not yet), so in this context his definition is implicitly used, but I wanted to point out that it is not the point of games in general.

Basically, I was disagreeing with SirMustapha's rhetoric, not idea.

Then poker got involved and I got kinda sucked into defending the view, that a game isn't a game unless its played for fun, which was (or at least I viewed it to be) on the sidelines, without bearing on the main topic.

Edit: Even though I'm adamant that games have to be played for fun, I'm still saying that computers play games, since that's the easiest way to describe what is going on.

Edit:

threedognice wrote:A game is a game.


I can't argue with that.

User avatar
Felstaff
Occam's Taser
Posts: 5181
Joined: Fri Feb 01, 2008 7:10 pm UTC
Location: ¢ ₪ ¿ ¶ § ∴ ® © ™ ؟ ¡ ‽ æ Þ ° ₰ ₤ ಡಢ

Re: 1002: "Game AIs"

Postby Felstaff » Thu Jan 12, 2012 4:04 pm UTC

threedognice wrote:what do professional cellists do with their cellos to make a living?

cellum
Away, you scullion! you rampallion! You fustilarian! I'll tickle your catastrophe.

User avatar
eran_rathan
Mostly Wrong
Posts: 1848
Joined: Fri Apr 09, 2010 2:36 pm UTC
Location: in your ceiling, judging you

Re: 1002: "Game AIs"

Postby eran_rathan » Thu Jan 12, 2012 4:14 pm UTC

DukeTwicep wrote:
Klear wrote:
threedognice wrote:
Edit: OK, I didn't mean to say that if you get money out of it it ceases to be a game. What I meant was that when you only do it for money it's not a game anymore.

It is not your hobby anymore, but it is still a game.

Progamers are gamers.


I already know your position. No need to repeat it. Our definitions differ, that's all.

I have to agree with threedognice. A game is a game no matter how individuals choose to play it, it only seizes to be a game when it no longer fits the description of a game. Of course, maybe games can seize to be games to individuals if they stop regarding the game as a game, but this has no bearing on if it is a game or not objectively.


re: 'seize'

You keep using that word. I do not think it means what you think it means.

I think you mean 'cease', as in, 'to stop'.

sincerely, Summer Glau
"Does this smell like chloroform to you?"
"Google tells me you are not unique. You are, however, wrong."
nɒʜƚɒɿ_nɒɿɘ

threedognice
Posts: 15
Joined: Mon Dec 19, 2011 10:07 am UTC

Re: 1002: "Game AIs"

Postby threedognice » Thu Jan 12, 2012 4:31 pm UTC

Klear wrote:I can't argue with that.

But you do, silly hat! Alas!

slownis
Posts: 1
Joined: Wed Jan 11, 2012 3:06 pm UTC

Re: 1002: "Game AIs"

Postby slownis » Thu Jan 12, 2012 5:57 pm UTC

Condor70 wrote:
pareidolon wrote:As for Snakes & Ladders, it's very simple.
Need to roll 4 to win game... ... ... COMPUTER ROLLS 4.


I would expect that of a computer that uses Randall's Random Number generator (http://xkcd.com/221/).

ps. I'm missing Rubiks Cube, solved July 2010 (http://cube20.org/). Or is the cube considered a teaching device and not a game?


I think Rubiks Cube counts as a game. Just saw in wired where a lego robot named 'Cubestormer II' beat the human record back in October. The robot's brain uses an Android cell phone running a custom made app. (http://www.wired.co.uk/news/archive/2011-11/11/cubestormer-ii-world-record). Pretty awesome design if you ask me.

Yeah, I registered just to post this. I've been an XKCD fan for a year now, and also a rubiks cube fan. The omission of 'the cube' was the first thing I noticed about comic.

Fieari
Posts: 102
Joined: Mon Jan 29, 2007 2:16 am UTC
Location: Okayama, Japan

Re: 1002: "Game AIs"

Postby Fieari » Thu Jan 12, 2012 7:47 pm UTC

Proginoskes wrote:That result about checkers isn't exactly right. The result that was proved was that if both sides play perfectly, then the game is a draw. In the paper itself the authors say that they don't have a method for winning in any arbitrary line of play. (That is, the problem of winning a game where a mistake has been made, hasn't been solved.)


Wrong. Checkers has been weakly solved, not ultra-weakly solved. In fact, they strongly solved it for any and every position with 10 pieces left on the board (5v5) as well as every 9, 8, 7, 6, 5, 4, 3, 2, and one piece endgame, and then weakly (but not ultra-weakly) found a solution that from the opening board can guarantee reaching one of the favorable endgame positions that are strongly solved.
Surely it is as ridiculous to consider sqrt(-1) "imaginary" because you can't use it to count pieces of chalk as to consider the number 200 imaginary because by itself it cannot express the location of one point with reference to another. -Isaac Asimov

wagner
Posts: 22
Joined: Tue Oct 06, 2009 3:18 pm UTC

Re: 1002: "Game AIs"

Postby wagner » Thu Jan 12, 2012 8:17 pm UTC

TranquilFury wrote:I would enjoy a game with trainable AI like that, but stilgar was, I believe, talking about this: http://www.youtube.com/watch?v=nNpy8Wnv ... re=related
Humans pros do micromanage mutalisks too, but in different situations with different methods, a human will typically have a control group of 11 mutalisks+ 1 overlord and can effectively use it against non-splash units with range greater than muta, which the ai sucks at, here's Jaedong: http://www.youtube.com/watch?v=o1MVOqrm4qc


What I'm getting at is that the digital computer, and the analog machine before it, are button mashers. They do things repetitively and very very fast. That is their whole expressed purpose. When the best humans might do 100APM, and AIs are pushing tens of thousands of APM, it opens up all new tactics for the AI that are simply not available to the human. It becomes a battle between thinkers and button mashers, and the thinkers want to believe they are somehow inherently better than something that cannot think but just operates very very fast.

Despite all the work that has gone into AI research in the past 50 years, the real gains have all been from throwing ever more powerful button pushers at the problem. More power for database lookups. More power for Monte Carlo simulations. More power to simply do more tasks per second. If you keep throwing more and more power at it, you will eventually beat the human, but it will just be following a pre-programmed set of rules, and will be no closer to something that actually offers a unique and unpredictable opponent. In Starcraft and other RTS games, people complain about this, and "level the playing field" by artificially limiting the one thing the AI has going for it, speed. I'm saying go the opposite direction, and crank up the human's speed by using simple helper AIs to perform various low level and tedious tasks. Bring the human up to the AI's level, so the competition becomes one of thought, of strategy. Bonus points if the AI's decision trees are built automatically by watching or participating in matches.

Fieari
Posts: 102
Joined: Mon Jan 29, 2007 2:16 am UTC
Location: Okayama, Japan

Re: 1002: "Game AIs"

Postby Fieari » Thu Jan 12, 2012 8:32 pm UTC

On the subject (back a few pages) of AI for Diplomacy (DAIDE) I've always wanted to see if one could be written for Forum Mafia/Werewolf and be playable amongst humans... it'd have to analyse post counts, post times, post lengths, understand Finger of Suspicions and OMGUS, pick out defending and attacking posts, and so on and so on. And then generate posts in a human readible format that would be persuasive enough (not having to pass a turing test could help with this, if it just gave it's readout of probabilities of suspicion it might work, but then it can also lie, so clever writers might link source data and give reasons).

All in all an interesting project to think about, except my coding skills suck. :(
Surely it is as ridiculous to consider sqrt(-1) "imaginary" because you can't use it to count pieces of chalk as to consider the number 200 imaginary because by itself it cannot express the location of one point with reference to another. -Isaac Asimov

pmwalk
Posts: 2
Joined: Thu Jan 12, 2012 5:56 am UTC
Location: Pittsburgh, PA

Re: 1002: "Game AIs"

Postby pmwalk » Thu Jan 12, 2012 8:49 pm UTC

Showsni wrote:It's really easy to make an AI that wins the Game. I could do that on my calculator!

Step 1. Inform calculator that is is now playing the game. Then wipe the memory.
Step 2. Periodically check calculator's memory. Nope, it still hasn't remembered it's playing the game.


But the point is you can never win the Game, just go for stretches of time where you don't lose. Then again, this comic focused on games computers could beat humans at, not necessarily win, which suggests it might be trivially easy because you could just strip the computer of all inputs so that it cannot be reminded of the game. If you assume all humans will eventually be reminded of it, this would work. At that point though, wouldn't your computer cease to be a computer?

User avatar
Klear
Posts: 1965
Joined: Sun Jun 13, 2010 8:43 am UTC
Location: Prague

Re: 1002: "Game AIs"

Postby Klear » Thu Jan 12, 2012 8:54 pm UTC

Goddamn! Can you stop talking about the Game? I had a pretty good winning streak lately... =(

TranquilFury
Posts: 131
Joined: Thu Oct 15, 2009 1:24 am UTC

Re: 1002: "Game AIs"

Postby TranquilFury » Thu Jan 12, 2012 9:09 pm UTC

wagner wrote:
TranquilFury wrote:I would enjoy a game with trainable AI like that, but stilgar was, I believe, talking about this: http://www.youtube.com/watch?v=nNpy8Wnv ... re=related
Humans pros do micromanage mutalisks too, but in different situations with different methods, a human will typically have a control group of 11 mutalisks+ 1 overlord and can effectively use it against non-splash units with range greater than muta, which the ai sucks at, here's Jaedong: http://www.youtube.com/watch?v=o1MVOqrm4qc


What I'm getting at is that the digital computer, and the analog machine before it, are button mashers. They do things repetitively and very very fast. That is their whole expressed purpose. When the best humans might do 100APM, and AIs are pushing tens of thousands of APM, it opens up all new tactics for the AI that are simply not available to the human. It becomes a battle between thinkers and button mashers, and the thinkers want to believe they are somehow inherently better than something that cannot think but just operates very very fast.

Despite all the work that has gone into AI research in the past 50 years, the real gains have all been from throwing ever more powerful button pushers at the problem. More power for database lookups. More power for Monte Carlo simulations. More power to simply do more tasks per second. If you keep throwing more and more power at it, you will eventually beat the human, but it will just be following a pre-programmed set of rules, and will be no closer to something that actually offers a unique and unpredictable opponent. In Starcraft and other RTS games, people complain about this, and "level the playing field" by artificially limiting the one thing the AI has going for it, speed. I'm saying go the opposite direction, and crank up the human's speed by using simple helper AIs to perform various low level and tedious tasks. Bring the human up to the AI's level, so the competition becomes one of thought, of strategy. Bonus points if the AI's decision trees are built automatically by watching or participating in matches.
Not necessary at all, the human Starcraft pros have 300~400 apm (sustained) which they have spent the past decade figuring out how to use most efficiently and effectively, so even if the ai has 50k apm, the humans are so much better at the decision making aspect, and good enough at the unit control aspect, that the AI gets crushed. It is incredibly difficult, and critically important being able to correctly balance your units(safety) against your economy and your tech, I haven't yet seen an AI efficiently hold off well executed pressure that it's programmers didn't see before and specifically account for, and too many times i've seen an AI player that WAS programmed to specifically account for an allin, completely overreact to something that it thought looks similar, and cripple it's own economy when it didn't need to, or not when/how to punish greedy build X.

Even for the humans, having lots of APM doesn't make you a better player, THAT comes from playing a lot of games and fixing the specific reasons you lose.

User avatar
Adam H
Posts: 1267
Joined: Thu Jun 16, 2011 6:36 pm UTC

Re: 1002: "Game AIs"

Postby Adam H » Thu Jan 12, 2012 9:25 pm UTC

Fieari wrote:On the subject (back a few pages) of AI for Diplomacy (DAIDE) I've always wanted to see if one could be written for Forum Mafia/Werewolf and be playable amongst humans... it'd have to analyse post counts, post times, post lengths, understand Finger of Suspicions and OMGUS, pick out defending and attacking posts, and so on and so on. And then generate posts in a human readible format that would be persuasive enough (not having to pass a turing test could help with this, if it just gave it's readout of probabilities of suspicion it might work, but then it can also lie, so clever writers might link source data and give reasons).

All in all an interesting project to think about, except my coding skills suck. :(
To be "playable", IMO, it would have to be able to understand human language (sarcasm, idioms, and typos would give it fits) more than technology currently allows (at least easily available technology). I could be wrong but I doubt AI is at the point where it can handle party games like mafia.

A mafia game with ONLY AI, now that would be interesting...

You should come play mafia on this forum, by the way! xkcders are the best at it. :)
-Adam

PigeonBrain
Posts: 5
Joined: Thu Jan 12, 2012 9:58 pm UTC

Re: 1002: "Game AIs"

Postby PigeonBrain » Thu Jan 12, 2012 10:01 pm UTC

Erinaceus wrote:What about backgammon? TD-gammon (a temporal-difference-learning algorithm for backgammon) was developed in 1992, and has actually created strategies that were unknown before.


You beat me to it :) TD-Gammon is a real AI, not a brute force chess calculator. Beat the world's best humans, too.

bridgeplayer
Posts: 2
Joined: Wed Jan 11, 2012 7:20 pm UTC

Re: 1002: "Game AIs"

Postby bridgeplayer » Thu Jan 12, 2012 10:12 pm UTC

[5 am, Wednesday morning.  I show up at my colleague's house to drive him to the airport for our business trip.]

Him: Have you seen xkcd this morning?

Me: Uh, no.  [Thinking: I got up early enough to shower and get here by 5 am; what more do you want?]

Him: You're going to be pissed that they left you out.

...

So of course I'm not pissed, but there are some other games worth mentioning, some of which have been touched on in this thread.

Euchre: yes, our program Euchre Baron did indeed defeat the 2011 World Series of Euchre Partners Champions last June.

Bridge: at this point, it's relatively easy to write an expert (but not world-class) declarer algorithm.  It's relatively easy to write a mediocre (but not strong) defensive algorithm.  It's very straightforward, but time-consuming, to write a strong bidding algorithm.  But world-class is a long way away in all three of those aspects of the game - for example, the search spaces involved with poker-like psychology in bridge are far too large.

Other solved games not mentioned in this thread or the comic include Awari and Nine Men's Morris. 

PCal
Posts: 60
Joined: Mon Jun 15, 2009 1:00 am UTC

Re: 1002: "Game AIs"

Postby PCal » Thu Jan 12, 2012 11:11 pm UTC

TranquilFury wrote:
wagner wrote:
TranquilFury wrote:I would enjoy a game with trainable AI like that, but stilgar was, I believe, talking about this: http://www.youtube.com/watch?v=nNpy8Wnv ... re=related
Humans pros do micromanage mutalisks too, but in different situations with different methods, a human will typically have a control group of 11 mutalisks+ 1 overlord and can effectively use it against non-splash units with range greater than muta, which the ai sucks at, here's Jaedong: http://www.youtube.com/watch?v=o1MVOqrm4qc


What I'm getting at is that the digital computer, and the analog machine before it, are button mashers. They do things repetitively and very very fast. That is their whole expressed purpose. When the best humans might do 100APM, and AIs are pushing tens of thousands of APM, it opens up all new tactics for the AI that are simply not available to the human. It becomes a battle between thinkers and button mashers, and the thinkers want to believe they are somehow inherently better than something that cannot think but just operates very very fast.

Despite all the work that has gone into AI research in the past 50 years, the real gains have all been from throwing ever more powerful button pushers at the problem. More power for database lookups. More power for Monte Carlo simulations. More power to simply do more tasks per second. If you keep throwing more and more power at it, you will eventually beat the human, but it will just be following a pre-programmed set of rules, and will be no closer to something that actually offers a unique and unpredictable opponent. In Starcraft and other RTS games, people complain about this, and "level the playing field" by artificially limiting the one thing the AI has going for it, speed. I'm saying go the opposite direction, and crank up the human's speed by using simple helper AIs to perform various low level and tedious tasks. Bring the human up to the AI's level, so the competition becomes one of thought, of strategy. Bonus points if the AI's decision trees are built automatically by watching or participating in matches.
Not necessary at all, the human Starcraft pros have 300~400 apm (sustained) which they have spent the past decade figuring out how to use most efficiently and effectively, so even if the ai has 50k apm, the humans are so much better at the decision making aspect, and good enough at the unit control aspect, that the AI gets crushed. It is incredibly difficult, and critically important being able to correctly balance your units(safety) against your economy and your tech, I haven't yet seen an AI efficiently hold off well executed pressure that it's programmers didn't see before and specifically account for, and too many times i've seen an AI player that WAS programmed to specifically account for an allin, completely overreact to something that it thought looks similar, and cripple it's own economy when it didn't need to, or not when/how to punish greedy build X.

Even for the humans, having lots of APM doesn't make you a better player, THAT comes from playing a lot of games and fixing the specific reasons you lose.
This, you could get to master league top 2% of the region you are playing, with around id say only 120 apm because until you get to to the very very upper level of play the importance of marco far far out weighs good mirco.

Mostlynormal
Posts: 32
Joined: Sat May 15, 2010 1:53 am UTC

Re: 1002: "Game AIs"

Postby Mostlynormal » Fri Jan 13, 2012 12:00 am UTC

I think my point kind of got missed. My point about computers not "playing" chess was that playing chess doesn't mean making moves any more than, say, sparring in martial arts is about moving muscles. Sure, you could describe it that way, but martial arts are so much more complicated and subtle than simply being able to move your muscles, or even move your muscles in a way that "beats" the opponent. There's a lot of technique that would be missed by a robot that could "beat" black belts simply because it's motors put out more power than a human muscle ever could. Unfortunately, that is exactly how chess "AIs" are designed: to beat opponents through brute force.

meerta
Posts: 63
Joined: Mon Aug 10, 2009 11:25 am UTC

Re: 1002: "Game AIs"

Postby meerta » Fri Jan 13, 2012 12:22 am UTC

Mostlynormal wrote:I think my point kind of got missed. My point about computers not "playing" chess was that playing chess doesn't mean making moves any more than, say, sparring in martial arts is about moving muscles. Sure, you could describe it that way, but martial arts are so much more complicated and subtle than simply being able to move your muscles, or even move your muscles in a way that "beats" the opponent. There's a lot of technique that would be missed by a robot that could "beat" black belts simply because it's motors put out more power than a human muscle ever could. Unfortunately, that is exactly how chess "AIs" are designed: to beat opponents through brute force.


Of course. Not to underplay the subtlety of the programming I'm not sure anyone would argue with that. But that's what the comic is about and it is quite an impressive fact on its own terms that computers can almost always beat us. There was a "chess-playing machine" called The Turk in the 18th and 19th centuries. It was operated by a skilled chess player cleverly hidden inside, but it instiled a sense of wonder that such a thing could be possible. Anyway it's because games like chess are rule-based that we get all this, but of course there's a distinction.

TranquilFury
Posts: 131
Joined: Thu Oct 15, 2009 1:24 am UTC

Re: 1002: "Game AIs"

Postby TranquilFury » Fri Jan 13, 2012 12:59 am UTC

PCal wrote:
TranquilFury wrote:
wagner wrote:
TranquilFury wrote:I would enjoy a game with trainable AI like that, but stilgar was, I believe, talking about this: http://www.youtube.com/watch?v=nNpy8Wnv ... re=related
Humans pros do micromanage mutalisks too, but in different situations with different methods, a human will typically have a control group of 11 mutalisks+ 1 overlord and can effectively use it against non-splash units with range greater than muta, which the ai sucks at, here's Jaedong: http://www.youtube.com/watch?v=o1MVOqrm4qc


What I'm getting at is that the digital computer, and the analog machine before it, are button mashers. They do things repetitively and very very fast. That is their whole expressed purpose. When the best humans might do 100APM, and AIs are pushing tens of thousands of APM, it opens up all new tactics for the AI that are simply not available to the human. It becomes a battle between thinkers and button mashers, and the thinkers want to believe they are somehow inherently better than something that cannot think but just operates very very fast.

Despite all the work that has gone into AI research in the past 50 years, the real gains have all been from throwing ever more powerful button pushers at the problem. More power for database lookups. More power for Monte Carlo simulations. More power to simply do more tasks per second. If you keep throwing more and more power at it, you will eventually beat the human, but it will just be following a pre-programmed set of rules, and will be no closer to something that actually offers a unique and unpredictable opponent. In Starcraft and other RTS games, people complain about this, and "level the playing field" by artificially limiting the one thing the AI has going for it, speed. I'm saying go the opposite direction, and crank up the human's speed by using simple helper AIs to perform various low level and tedious tasks. Bring the human up to the AI's level, so the competition becomes one of thought, of strategy. Bonus points if the AI's decision trees are built automatically by watching or participating in matches.
Not necessary at all, the human Starcraft pros have 300~400 apm (sustained) which they have spent the past decade figuring out how to use most efficiently and effectively, so even if the ai has 50k apm, the humans are so much better at the decision making aspect, and good enough at the unit control aspect, that the AI gets crushed. It is incredibly difficult, and critically important being able to correctly balance your units(safety) against your economy and your tech, I haven't yet seen an AI efficiently hold off well executed pressure that it's programmers didn't see before and specifically account for, and too many times i've seen an AI player that WAS programmed to specifically account for an allin, completely overreact to something that it thought looks similar, and cripple it's own economy when it didn't need to, or not when/how to punish greedy build X.

Even for the humans, having lots of APM doesn't make you a better player, THAT comes from playing a lot of games and fixing the specific reasons you lose.
This, you could get to master league top 2% of the region you are playing, with around id say only 120 apm because until you get to to the very very upper level of play the importance of marco far far out weighs good mirco.
I'd say you could get to master league in sc2 w/ 60 apm and good timings, GM/pro with 100~150 average apm. SC2 is far less mechanically demanding than BW. (edit: to clarify, you can still attain 300+ apm in sc2, you just hit diminishing returns on those actions, and can still lose to people with half that APM.)

aplarsen
Posts: 5
Joined: Mon Jul 25, 2011 5:23 am UTC

Re: 1002: "Game AIs"

Postby aplarsen » Fri Jan 13, 2012 4:10 am UTC

The UIUC "Beer Pong" robot can be seen here:

http://www.youtube.com/watch?v=HkhMCCOHFmM

The throwing of the ping pong ball starts at about 00:28.

You see my friend Aaron at 1:48.

Netzach
Posts: 18
Joined: Fri Oct 30, 2009 9:28 am UTC

Re: 1002: "Game AIs"

Postby Netzach » Fri Jan 13, 2012 4:50 am UTC

I am pretty sure that Othello/Reversi is solved.

User avatar
Proginoskes
Posts: 313
Joined: Mon Nov 14, 2011 7:07 am UTC
Location: Sitting Down

Re: 1002: "Game AIs"

Postby Proginoskes » Fri Jan 13, 2012 5:50 am UTC

Mostlynormal wrote:I think anyone who thinks they can talk about chess being "solved" either knkows very little about chess or just hasn't thought it through enough. Even if we could muster the computing power to completely brute force the game of chess, would chess be "solved"? Of course not. Maybe a computer managed to perform a bunch of calculations that resembled chess moves, but what does that mean? The computer can't understand chess. No human could read its output.


Well, that's the human's fault then, isn't it? Computers are machines designed to extend the ability of human minds, so it is not surprising that some computations would be unreadable to us. (In fact it would be surprising if none of its computations WERE unreadable.)

Consider this as well: If a Martian with enough brain power to read the proof existed, and read the proof, would chess then be solved only then?

lightvector
Posts: 224
Joined: Tue Jun 17, 2008 11:04 pm UTC

Re: 1002: "Game AIs"

Postby lightvector » Fri Jan 13, 2012 6:32 am UTC

Othello/Reversi is not solved, although it might be within reach before too long. If there is some powerful new trick or algorithm or optimization discovered that gains a fundamental speedup, it might be doable right now. Otherwise with no such improvements, from what I can find out online, there are still an estimated several orders of magnitude difference between solving Reversi and solving Checkers (maybe somewhere between 10^4 and 10^8).

Checkers was solved by mid-2007 using as many as a couple hundred computers more or less continuously over a period of a few years. So, assuming the above estimates for Reversi relative to Checkers, with no software improvements and no specialized hardware, we're looking at something that could possibly be within reach of an enormous distributed computation after a few decades more of Moore's law.

SqueeSpleen
Posts: 1
Joined: Fri Jan 13, 2012 6:21 pm UTC

Re: 1002: "Game AIs"

Postby SqueeSpleen » Fri Jan 13, 2012 6:31 pm UTC

shirosuzume wrote:Do you know how long I've been looking for somebody else that's heard of Mao? And here I had come to the conclusion that it only existed in Washington DC, in 1994, among the people who were there with me.

It's not a video game, it's a card game. A lovely, lovely, awesome card game. There is only one rule: You can't tell the rules. Muahahahaha....

Now I must find more people... there must be more!

This is the best game I ever played
Sadly, I only played it in 2006 when I was in the OMA National Contest... I never finded people to play it again :(
Only a few people wants to play a game where they don't know the rules :(

domino14
Posts: 19
Joined: Sun Apr 27, 2008 10:51 am UTC

Re: 1002: "Game AIs"

Postby domino14 » Fri Jan 13, 2012 9:48 pm UTC

Top players can beat the computer at Scrabble, or at least keep it very close (one top player played 500 games recently and won about 248 of them - the computer used was Quackle, the top AI out there).

PCal
Posts: 60
Joined: Mon Jun 15, 2009 1:00 am UTC

Re: 1002: "Game AIs"

Postby PCal » Fri Jan 13, 2012 10:32 pm UTC

domino14 wrote:Top players can beat the computer at Scrabble, or at least keep it very close (one top player played 500 games recently and won about 248 of them - the computer used was Quackle, the top AI out there).
Since its a game with luck it should be impossible for the computer to get a 100% win rate over. A over 50% win is still surprising to me.

domino14
Posts: 19
Joined: Sun Apr 27, 2008 10:51 am UTC

Re: 1002: "Game AIs"

Postby domino14 » Fri Jan 13, 2012 10:46 pm UTC

PCal wrote:
domino14 wrote:Top players can beat the computer at Scrabble, or at least keep it very close (one top player played 500 games recently and won about 248 of them - the computer used was Quackle, the top AI out there).
Since its a game with luck it should be impossible for the computer to get a 100% win rate over. A over 50% win is still surprising to me.


It is a bit to me too, but a computer has perfect word knowledge. The fact that a top human can beat it nearly 50% of the time says that the game is actually more complex than Randall is giving it credit for (I think he listed it too high up).

dragondave
Posts: 22
Joined: Fri Nov 14, 2008 8:46 pm UTC

Re: 1002: "Game AIs"

Postby dragondave » Sat Jan 14, 2012 4:44 am UTC

PCal wrote:Since its a game with luck it should be impossible for the computer to get a 100% win rate over. A over 50% win is still surprising to me.


Tournament Scrabble has no probabilistic features* - everyone has exactly the same tiles and makes moves on exactly the same board; only the person who wins each round gets to place 'their' word on the board everyone plays on for the next round.

Domino14: Can you provide a reference? It would be interesting to see whether they played Tournament or not.

* Other than "what tiles come up next turn", of course.

jjcote
Posts: 32
Joined: Sat Apr 16, 2011 2:16 am UTC

Re: 1002: "Game AIs"

Postby jjcote » Sat Jan 14, 2012 2:06 pm UTC

jonadab wrote:If there were a non-turn-based "Speed Scrabble", where you can play as fast as you can come up with plays, it would take me about five minutes to program an AI no human player could ever touch.)

Bananagrams. Yeah, a program could make short work of humans.


Return to “Individual XKCD Comic Threads”

Who is online

Users browsing this forum: No registered users and 88 guests