|
But the points that both people in your quote nests make are fair. The fact that Go has more states made it so it both needed more tricks and more computational power to beat the top players.
If RTS indeed have more states than Go, the same problem will persist.
The other point is that there is now AI that can beat computers even though we cannot naively exhaust all game states of Go (ie, brute force in the traditional sense of the word).
Go was 'handled' despite the number of positions. Not irregardless of them. It is still true that the difficulty of solving a game is approximately directly proportional to the number of game states. Why? Because our way of solving them still relies on naively sampling a small fraction of the possible game states. To solve a game, we both need smart tricks to make that fraction really really small. And we need computational power, because if the number of game states is really really big, even completely sampling that really really small fraction will take a lot of time.
Your comment is no longer puzzling to me. It is simply wrong.
|
Norway28690 Posts
I think expecting that knowledge of how the world's most advanced AI's operate should be common knowledge, even in this forum, is kinda silly.
|
France12901 Posts
On May 28 2017 07:56 Ernaine wrote:
But the points that both people in your quote nests make are fair. The fact that Go has more states made it so it both needed more tricks and more computational power to beat the top players.
If RTS indeed have more states than Go, the same problem will persist.
The other point is that there is now AI that can beat computers even though we cannot naively exhaust all game states of Go (ie, brute force in the traditional sense of the word).
Go was 'handled' despite the number of positions. Not irregardless of them. It is still true that the difficulty of solving a game is approximately directly proportional to the number of game states.
Your comment is no longer puzzling to me. It is simply wrong.
?
Of course it'll be harder if there are more states, except in a few particular cases, but that's obvious so it doesn't need to be said.
However it's hard to accurately measure the number of states of Starcraft because it's harder to "model" what is a game of starcraft.
But imho the most difficult thing about this all, will be convincing people that the AI won the games fairly.
Since mechanics are such a vital part of Starcraft, there will always be ways for defeated players to contest the loss.
Once the egos of the top players will be in danger, they'll not accept the games as fair because you can argue forever about it.
|
On May 28 2017 01:04 XenoX101 wrote:Show nested quote +On May 28 2017 00:16 LetaBot wrote:On May 27 2017 21:55 XenoX101 wrote:
One problem with AI learning a game like SC2 is that it can't speed up the learning process without access to the game code. (I.e. it cannot play 1,000,000,000 games against itself to learn optimal strategies). So it can only play games at the same pace as humans to gather information. The good part is that it can play 24/7 where a normal player can only play 12/7, and the amount of information it can gather per match is much higher than an AI. It could technically make more precise calculations about engagements than any player.
However it may also be possible for the AI to play games in its "head", if it plays enough games to be able to understand the game mechanics well enough. So then even if it can't physically play a lot of games, it can simulate them at a much faster rate than a human can play them. Technically if its mental model is accurate enough it could bypass playing games altogether, and rely solely on its understanding of the game to run "game experiments" in its head. But the flipside is that unlike Go there are too many possible outcomes for each game, such that you would need an incredibly powerful supercomputer to run through even a small fraction of them. So the AI would have to be very well optimized to only look at "worthwhile" simulations, and ignore the billions of not so valuable simulations (e.g. how to deal with mass reapers or a similarly weak startegy) that would only waste processing time.
EDIT: Thinking about this more I see one way that humans can still win, and that is through "sub-optimal" play that the AI would not expect or would be willing to gamble losing to because most players wouldn't do it. This would be something like a DT rush against someone going Robo tech with observers, or a transition into mass Muta after Terran has gone viking. If the AI doesn't scout it, it would not expect this kind of play. On average it will still likely win most games because of the balance of probabilities, but it would not win 100% due to these kind of oddball games where it would have to predict stupidity. Though this is more reflective of the game itself than the AI, where there are always going to be games that lead to inevitable build order losses. So the real test isn't whether AI can always beat human players, or even if it can beat the best players, but whether it can do so with a higher winrate than any existing player, i.e. greater than Flash's 70% in all matchups. Deepmind works closely together with Blizzard, so they will probably have some way to speed up the game. Sub-optimal play won't work either, since even in BW there are bots that can consider the possible build orders and unit compositions based purely on the time of the game (there are only so many minerals you can gather in a certain amount of time). The main issue of bots right now is actually micromanagement. Even with unlimited apm you still have to make tactical decisions, which bots aren't good at yet. Wouldn't "speeding up the game" be considered cheating? It's something normal players would not have access to, so I would think they wouldn't allow the AI to do it. My thinking is the AI can only access the visual pixel information, the same as a real person, as this would put it on equal footing with a human. As for sub-optimal play, this is an opportunity cost issue. You can only choose a build that is viable against some subset of builds, meaning you are guaranteed to be vulnerable to the compliment of that subset. The AI would ideally always pick the build that is viable against the most probable builds to be used, which are almost always the "best" builds for the other player to choose. The issue is that there will always be the risk of the human player choosing a "not so good" build, which is outside of the subset that the AI will do well against. This is because the AI is technically making the right choice, it is just that the right choice still has a build order loss probability. A simpler way to say this is that AI will always lose since there is no build without a BO loss probability.
This comment makes no sense. An AI is not a human. The point is somehow create an AI that can beat humans in a fair game. You do not create an AI the way you create a human. Maybe humans are cheating at RTS, because we humans have things AI's so far do not have.
Once the AI is created, I think there is something to be said about having the AI do completely the same tasks that humans do; looking at the screen and physically moving the mouse. But as any AI vs Human match is artificial anyway, you can arbitrarily select whatever way of having them compete.
As for picking BO's, there are different ways to go about it. You can code the AI to play a build that is soft countered by all, hard countered by none, and allows the AI to do it's AI thing, and get an advantage that way later. This assumes the AI can actually outplay the human in a long game.
You can provide the AI with statistics showing how likely it is for a human to play either build. This means there is no possibility of the AI not considering the opening build of the human. I also don't get how you claim a build can be 'the right choice' but can still 'always lose'.
There is also no reason why information about the BO the human picks cannot leak out of the player, through the way he plays, and inform the AI. Like how the player moves his scouting worker. How keen the player is on preventing scouting by the AI. Building placement, etc. AI can see details and pattern simply invisible to humans. And neural net/tensorflow algorithms can be very good at that.
In SC currently, we have AI's that use really odd and stupid BO's. But they are hard-coded/optimized for the select number of game states that result in those games. And that can be an ok scenario for them. So I do not see this problem that AI's might have with BO's. irrational decisions humans made, or limited information.
|
On May 28 2017 08:05 Poopi wrote:
But imho the most difficult thing about this all, will be convincing people that the AI won the games fairly.
Since mechanics are such a vital part of Starcraft, there will always be ways for defeated players to contest the loss.
Once the egos of the top players will be in danger, they'll not accept the games as fair because you can argue forever about it.
Why? Just look at chess. In chess, we knew computers were going to one day beat all humans. A very small number of chess players, and a slightly bigger number of ordinary people, thought that would be a problem for chess. And there are still people claiming that chess engines 'cheat', because they have endgame tables or opening databases. Or worse, because computers evaluate moves completely differently from how we think we evaluate a move (the truth is, we have little idea about how we do it).
But that all turned out to be completely irrelevant.
So we know that for any game, there will be a very small window in time where a computer vs human match will be interesting. Before and after that time, the human will either win or lose easily. And for human vs human competition, being before or after that small window is largely irrelevant.
You really think the ego of Usain Bolt is bruised by the fact that a car can run faster than he can?
|
Norway28690 Posts
Quite some people think AI's semi-solving chess kinda ruined the game. I definitely feel that way about backgammon, and I think following chess tournaments where the analysis ends up being 'so the computer says this is not the best move' 'the computer says this move would be better' 'the computer says this was indeed the ideal move' 'this is a flawless game so far, the moves have been identical to computer suggestions' is really boring. I expect AI's to become better than humans at everything humans do during my lifetime but I also think that's largely a negative thing.
|
You refer here to a style of analysis where the analysis consists of simply stating what moves the computer would make. That indeed, is a boring and highly non-instructive analysis.
Chess engines certainly changed the way top players have to approach the game, where using engines to help analyze positions is a large thing. Including using computers to find novelty moves.
But all this preparation and memorization being done would be there even without computers, as it is a natural element of the way chess works. And we do the same thing in RTS. Any game where the starting point is identical every time will have this aspect to it.
The importance of opening repertoire and game preparation, they are removing some of the charm of chess. Not chess engines.
Computers certainly will remove 'magic' or 'mystery' or 'murkiness' of the game, reducing it to what it actually is. But that is just how insight in general works.
I have not seen statistics that show that indeed less players play chess now that would have without computers. Or that chess is less popular because of chess engines.
|
Norway28690 Posts
I'd guess the ease of playing pickup chess online has made more people play it, personally. And I'm not saying that my opinion is that of others or that others should adopt my opinion, but to me, I just don't find that much enjoyment in trying to master something a computer is better than me at. Which paints a bleak picture of the future, as that's going to be literally every activity. ;p Figuring out the magic and mystery is where the primary enjoyment lies. To me. 
|
On May 28 2017 08:24 Liquid`Drone wrote:
Quite some people think AI's semi-solving chess kinda ruined the game. I definitely feel that way about backgammon, and I think following chess tournaments where the analysis ends up being 'so the computer says this is not the best move' 'the computer says this move would be better' 'the computer says this was indeed the ideal move' 'this is a flawless game so far, the moves have been identical to computer suggestions' is really boring. I expect AI's to become better than humans at everything humans do during my lifetime but I also think that's largely a negative thing.
do people really say that in chess tournaments? what a horrible way to ruin the fun
|
The point is, the magic and mystery wasn't there in the first place.
And computers or no computers, chess is the greatest waste of human intelligence either way. I guess what computers are doing is making people force to really think about why they do and care for certain things. Because without them, they don't need to know, or they think they know, but they don't.
|
On May 28 2017 08:24 Liquid`Drone wrote:
Quite some people think AI's semi-solving chess kinda ruined the game.
I don't know why people would say that. Top chess tournaments are way more exciting than they were 10-15 years ago. There's a lot more fighting spirit and instead of playing the same lines over and over again, people are playing a wider variety of openings in order to dodge each others computer assisted preparation.
I definitely feel that way about backgammon, and I think following chess tournaments where the analysis ends up being 'so the computer says this is not the best move' 'the computer says this move would be better' 'the computer says this was indeed the ideal move' 'this is a flawless game so far, the moves have been identical to computer suggestions' is really boring.
I agree but the trend has been for commentator to not use chess engines anymore. Actually, even following the thought process of a top grandmaster would be a challenge for most of us. Better to explain ideas at the level of your audience, even if the result is that you don't reach a conclusion in many positions.
|
France12901 Posts
On May 28 2017 08:20 Ernaine wrote:Show nested quote +On May 28 2017 08:05 Poopi wrote:
But imho the most difficult thing about this all, will be convincing people that the AI won the games fairly.
Since mechanics are such a vital part of Starcraft, there will always be ways for defeated players to contest the loss.
Once the egos of the top players will be in danger, they'll not accept the games as fair because you can argue forever about it.
Why? Just look at chess. In chess, we knew computers were going to one day beat all humans. A very small number of chess players, and a slightly bigger number of ordinary people, thought that would be a problem for chess. And there are still people claiming that chess engines 'cheat', because they have endgame tables or opening databases. Or worse, because computers evaluate moves completely differently from how we think we evaluate a move (the truth is, we have little idea about how we do it). But that all turned out to be completely irrelevant. So we know that for any game, there will be a very small window in time where a computer vs human match will be interesting. Before and after that time, the human will either win or lose easily. And for human vs human competition, being before or after that small window is largely irrelevant. You really think the ego of Usain Bolt is bruised by the fact that a car can run faster than he can?
?
In chess, mechanics don't matter, so it's only a matter of which move to make, not how well you do it.
In starcraft, you can win games solely with your mechanics. You can play "stupidly" in "autopilot" mode but crush inferior players if they can't handle your superior multitasking.
You can win games with good micro, etc...
The game is not only about strategy but about forcing players to do mistakes by giving them less time to think, using your physical abilities.
There is a reason foreign terrans can't do anything whereas Koreans can win in their scene: terran is too hard mechanically compared to the other two races. So there are very concrete examples of why handling the mechanics issue is important for the integrity of the competition between AI and humans in SC.
Plus we humans perform these mechanical tasks with various levels of performance.
Strict training allow top players to reduce the variance of their performance, but there is still some. So you can't just allow the AI to micro as well as the best human ever achieved (unfair for the human), nor can you only allow it to micro like the best players micro on average (unfair for the AI).
So players will agree on something they thought was fair, but they'll want a rematch with other conditions because maybe it wasn't that fair  .
edit: and the game being "solved" or not is a huge deal. I know I won't ever bother to learn how to play chess precisely because it's solved. And now I know I don't have to waste time learning Go either. Some people think like me, some don't care, but I'm pretty sure I'm not the only one so this fairness thing is very important. For example, when there was a lot of hype with people training themselves to succeed in the 2048 game, I didn't even bother to try to learn how to achieve it, because I could just launch a bot that does it for me, which I did, so better not waste my time with that game.
|
You're right and I also think that an AI beating a human wouldn't prove that the AI was strategically superior.
Even with an APM cap people would say something like "the AI has perfect accuracy, never forgets supply depots etc"
However it would still be an incredible achievement to beat SC players with an AI, even if they have a slight mechanical advantage.
|
On May 28 2017 08:59 Poopi wrote:?
In chess, mechanics don't matter, so it's only a matter of which move to make, not how well you do it.
In starcraft, you can win games solely with your mechanics. You can play "stupidly" in "autopilot" mode but crush inferior players if they can't handle your superior multitasking.
I don't know what this means. There is no universal agreement on what 'mechanics' means in chess. It is used every once in a while, but referring to different things. Usually to tactics or endgame concepts.
I also don't know what exactly you mean with 'playing stupidly'. And I am sure, top players can beat you on autopilot in chess.
You can win games with good micro, etc...
In chess you can win games with good strategy, good positional play, good tactics, good endgame play, etc.
The game is not only about strategy but about forcing players to do mistakes
Which is true in all chess.
by giving them less time to think,
Which in chess is true in some time control formats always, and at certain points in any time control.
using your physical abilities.
I guess this refers to the ability to execute micro/multitasking/macro. I highly contest this is a physical ability. You don't need 'fast muscles' or 'strong fingers' to micro as a top player. It is about connections in your brain.
There is a reason foreign terrans can't do anything whereas Koreans can win in their scene: terran is too hard mechanically compared to the other two races. So there are very concrete examples of why handling the mechanics issue is important for the integrity of the competition between AI and humans in SC.
Plus we humans perform these mechanical tasks with various levels of performance.
Strict training allow top players to reduce the variance of their performance, but there is still some. So you can't just allow the AI to micro as well as the best human ever achieved (unfair for the human), nor can you only allow it to micro like the best players micro on average (unfair for the AI).
I don't get this. Computer scientists are in the business of making powerful computers and smart algorithms. Every now and then they decide to show off their abilities by creating AI that can play, and hopefully for them beat, human players. There is no 'fairness'. If the challenge the CS people have is to create an AI with the decision making ability of a SC player, the issue of mechanics and micro is irrelevant. The compute can have APM far beyond what is humanly possible. A car runs faster than a human can possibly run. That is just how things are.
If they want to create a robot arm that moves around the mouse, to show they can meet the mechanical challenge of the robotics involved there, they do that.
In the end humans are biological machines. Our performance has variance, always, because the biochemistry has a certain level of stochasticisity to it, being biochemical reactions. So with the same input, the output is not always the same. That is where 'human mistakes' come from. And that is something code doesn't have.
As in chess, moving the piece is an arbitrary task for both the human and the robot, it is not part of the challenge. Controlling the mouse and keyboard in SC isn't, neither for the human or the robot/AI.
So players will agree on something they thought was fair, but they'll want a rematch with other conditions because maybe it wasn't that fair .
But aren't the initial game conditions agreed on?
edit: and the game being "solved" or not is a huge deal. I know I won't ever bother to learn how to play chess precisely because it's solved. And now I know I don't have to waste time learning Go either. Some people think like me, some don't care, but I'm pretty sure I'm not the only one so this fairness thing is very important. For example, when there was a lot of hype with people training themselves to succeed in the 2048 game, I didn't even bother to try to learn how to achieve it, because I could just launch a bot that does it for me, which I did, so better not waste my time with that game.
Maybe you need to think about why you play a game. Or do anything you do. So you decide you like to do certain things. But then when an AI shows up and does it better than you, you decide top stop doing it. Then why did you do it in the first place? That seems odd to me.
Whatever you are doing, there is going to be a person, and probably an AI, that can do it better.
|
France12901 Posts
I'm talking about Starcraft in my examples most of the time so why do you talk about chess instead?
Since you seem to answer in a weird manner on every detail, and that you have a very low post count, I can now be sure that you are trolling, so I won't bother answering anymore :/.
Edit: however the last portion is interesting so I'll answer that.
Because you can have fun with your opponent, tease each other and basically affect his feelings with your play.
I never play against AI not because they are bad, but because it's not fun. It's pointless to manner mule an AI, it's empty.
And I will keep playing Starcraft exactly because imho there is no way for the competition to be totally fair for both parties so I will always be able to argue that it doesn't prove much.
In my opinion anyways RTS are not hard per se strategically, because people collectively make the metagame so every top player eventually has the same knowledge and experience / reactions so it comes down to mechanics and not some magical creativity.
That is why I prefer players that have a very good micro because that's the most difficult thing, everyone plays so fucking much but the most ephemere thing players have are their mechanics, your hands get old whereas your mind can still take good decisions when you are old if you are experienced.
|
The point is - if they'll make AlphaGo AI strong enough to compete humans with let's say apm 250 limit then it would be interesting, because perfect mechanics does not mean game winning situation. It means for perfect supply/worker production it takes some % from those (250 apm) restrictions.
|
Your post starts with 'In chess, [...], but I am not allowed to talk about chess?
Most things you say here seem contradictory or just overall not very well thought out. Ok, so now you say you prefer not to play vs AI. Fine. But earlier you and other suggested that it is pointless to play a game when there is AI that is better than the best human.
You only registered in 2010, so maybe your way of posting is understandable. Especially after you admitted that you play to humiliate other people, the knowledge that they feel worse about themselves because of you, and that is where you find your satisfaction.
You can not tell if I am trolling? Maybe that is what happens when a scientist talks to an illiterate college kid? They don't know if they are hit with knowledge bombs, or being trolled?
|
Norway28690 Posts
On May 28 2017 08:44 Die4Ever wrote:Show nested quote +On May 28 2017 08:24 Liquid`Drone wrote:
Quite some people think AI's semi-solving chess kinda ruined the game. I definitely feel that way about backgammon, and I think following chess tournaments where the analysis ends up being 'so the computer says this is not the best move' 'the computer says this move would be better' 'the computer says this was indeed the ideal move' 'this is a flawless game so far, the moves have been identical to computer suggestions' is really boring. I expect AI's to become better than humans at everything humans do during my lifetime but I also think that's largely a negative thing. do people really say that in chess tournaments? what a horrible way to ruin the fun
I'm not really a connoisseur of chess tourneys, but with Magnus Carlsen becoming the pride of Norway, quite some of his games have been broadcast with commentary. And I mean, it's not the only thing they say, but the way they evaluate which player is ahead is through computer calculation of which player is ahead, and they always state what move the computer thinks is best before either player makes their move (unless it's really fast moving). I'm sure there might be other chess broadcasters where they avoid this style, because indeed, I don't think all that fun.
|
That's beside the point. At one point, people used to say 'Kasparov thinks it is a good/bad move' during some live game analysis, which is what was done just after he retired.
|
France12901 Posts
On May 28 2017 10:31 Ernaine wrote:
Your post starts with 'In chess, [...], but I am not allowed to talk about chess?
Most things you say here seem contradictory or just overall not very well thought out. Ok, so now you say you prefer not to play vs AI. Fine. But earlier you and other suggested that it is pointless to play a game when there is AI that is better than the best human.
You only registered in 2010, so maybe your way of posting is understandable. Especially after you admitted that you play to humiliate other people, the knowledge that they feel worse about themselves because of you, and that is where you find your satisfaction.
You can not tell if I am trolling? Maybe that is what happens when a scientist talks to an illiterate college kid? They don't know if they are hit with knowledge bombs, or being trolled?
What?
I said firstly to have fun with my opponent. Because that's exactly what is is. If you win you get to boast, while if you lose it's the opponent. The thing is that online you can't have fun with the guy so everyone seems like an horrible person, but when players meet irl they can relax and have fun about their loss because they can connect with their opponent which isn't possible online.
Everyone on TL knows what people mean when saying "mechanics", at least most people have a rough idea.
The fact that you seemingly don't know this and rather speak from an outside perspective is really weird on such a forum.
The fact that you cherry pick what suits you best also increase the probability of trolling.
|
|
|
|
|
|
EPT
