| Forum Index > SC2 General |
|
necaremus
45 Posts
On March 16 2016 04:45 Xyik wrote: [...] divide that by the 9 possible match-ups (TvX, ZvX, PvX), [...] try that again, plz >_> unless you are counting TvZ and ZvT as different match-up, i don't get 9 | ||
|
Xyik
Canada728 Posts
On March 16 2016 06:05 necaremus wrote: try that again, plz >_> unless you are counting TvZ and ZvT as different match-up, i don't get 9 i was thinking that it would need to learn the play Z build orders vs T just as it would need to learn to play T orders vs Z, but yes you are right, you can remove all duplicates by segmentation (e.g: treat a TvZ same as a ZvT when the map, spawn and build orders are the same for both races). | ||
|
LetaBot
Netherlands557 Posts
I have also seen several bots that let their units spam so much apm that they keep getting stuck in the attack animation without actually attacking. As for limiting the apm to something like 300: Once bots can defeat pro-players without an APM limit rule, you can slowly start capping it more and more. Right now bots are only competitive with C- or lower players, even with their high apm. | ||
|
snakeeyez
United States1231 Posts
That being said a bot playing protoss could beat pro human players just with perfect dragoon micro and kiting. It would always trade favorably, and then just win late game with all those advantages. Its not really fair if flash has to either micro marines or macro at his base while the bot can do everything at once. Its a huge advantage for the bot. Making a bot that beats flash in a best of 7 without falling into predictable patterns or stupid build orders/decisions would be a very tough AI challenge though. Even harder if you emulate how a human plays the game so it has to beat him fairly. After beating GO grandmaster they are running out of games to solve honestly. I never thought they would win at GO its something they been after since 1960s | ||
|
dankobanana
Croatia238 Posts
 | ||
|
Charoisaur
Germany15981 Posts
On March 16 2016 04:40 nimdil wrote: It's funny that people discuss - at the same time - if AI could beat top players in StarCraft and the ways how SC-tuned AlphaGo like AI (so AlphaSC I guess) should be obstructed so that the game will be fair. AlphaGo is an AI that bases it's aactions on graphical input and that's it. If you feel that you need to tune down the ability of AI to execute perfect strategies at superhuman speed even though it would be using standard inputs then sorry but it is game over. AI won. SC2 is meant to be played with mechanical restraints (aka mouse and keyboard) and not through graphical input. If someone plays sc2 without those mechanical restraints he would be basically cheating. So the only fair ways for an AI vs human game would be either to have the AI played by a robot (which isn't technically possible) or simulate a robot (aka played through graphical input but with the mechanical restraints that humans/robots have). otherwise the AI would be cheating. | ||
|
LetaBot
Netherlands557 Posts
On March 16 2016 09:40 Charoisaur wrote: SC2 is meant to be played with mechanical restraints (aka mouse and keyboard) and not through graphical input. If someone plays sc2 without those mechanical restraints he would be basically cheating. So the only fair ways for an AI vs human game would be either to have the AI played by a robot (which isn't technically possible) or simulate a robot (aka played through graphical input but with the mechanical restraints that humans/robots have). otherwise the AI would be cheating. So to clarify, would this be considered cheating or not by you? | ||
|
Hexe
United States332 Posts
| ||
|
Baarn
United States2702 Posts
| ||
|
aTnClouD
Italy2428 Posts
| ||
|
ashara
France22 Posts
It would be quite interesting to see if they can make an AI that just knows how to learn. So that you could tell it: "learn to play Starcraft 1v1 at top level" or "write an html5 breakout game and publish it on this platform" and it would actually do it. | ||
|
papaz
Sweden4149 Posts
On March 16 2016 15:14 aTnClouD wrote: Bots are nowhere near the level to solve starcraft and play the best players. Don't underestimate the complexity and amount of time of learning Starcraft through trial and error for a computer compared to a game like chess. There's too much difference between the two and human brain can still find valid strategies at a higher and faster degree than a bot at this point in time. I think in GO the AI learnt by playing literally millions of games vs itself. It would do the same thing in Starcraft in order to improve, just as humans do. The only difference? It can play millions of games simultaneously and learn faster than any human. Also it could learn the game by "watching" vods of pro players. There is no possible way a human can come close to computers. Just think about it for a second. Just 40 years ago there was no home PCs. Today we have computers everywhere, in just 40 years! Just think of how fast the development goes. Can you even imagine the world in 40 more years from today? If Google decides to dedicate time to a Starcrat AI make no mistake. It will crush any of todays pros. I love how the pro GO players were so confident of themselves and both European and World champion got crushed. Google and AI fighting! | ||
|
aTnClouD
Italy2428 Posts
On March 16 2016 17:04 papaz wrote: I think in GO the AI learnt by playing literally millions of games vs itself. It would do the same thing in Starcraft in order to improve, just as humans do. The only difference? It can play millions of games simultaneously and learn faster than any human. Also it could learn the game by "watching" vods of pro players. There is no possible way a human can come close to computers. Just think about it for a second. Just 40 years ago there was no home PCs. Today we have computers everywhere, in just 40 years! Just think of how fast the development goes. Can you even imagine the world in 40 more years from today? If Google decides to dedicate time to a Starcrat AI make no mistake. It will crush any of todays pros. I love how the pro GO players were so confident of themselves and both European and World champion got crushed. Google and AI fighting! As I previously stated bots are not capable of competing with humans at this point in time. There will be a point in the future when they will solve any kind of complex game but the day is not close to now. It comes down to how complex the game is. Backgammon has been solved but poker is nowhere close to be solved. Starcraft is extraordinarily complex and so the same applies for it. As computers get better and faster and better self learning programs are developed sure AI will find and execute strategies that can't be overcome by humans. | ||
|
sh1RoKen
Russian Federation93 Posts
On March 16 2016 18:29 aTnClouD wrote: As I previously stated bots are not capable of competing with humans at this point in time. There will be a point in the future when they will solve any kind of complex game but the day is not close to now. It comes down to how complex the game is. Backgammon has been solved but poker is nowhere close to be solved. Starcraft is extraordinarily complex and so the same applies for it. As computers get better and faster and better self learning programs are developed sure AI will find and execute strategies that can't be overcome by humans. Could you please read about AlphaGo or Artificial neural networks just a little bit before arguing about the topic you clearly know nothing about. That "point in the future when they will solve any kind of complex game" has been reached just now. GO has more combinations than atoms in the whole universe. And that AI doesn't calculated all of it and solved the whole game. He learned to play the game like humans do but much better. That AI has nothing in common with any other AI you might find in any other game. And BTW: 1. He can kill 100 mutalisks with 1 phoenix without taking any damage 2. He can kill 200 marines without steam with 1 stalker without taking any damage 3. He can kill 33 ultralisks with 1 immortal + warp prism without taking any damage 4. He can do 1, 2 and 3 at the same time. Starcraft units wasn't designed to be microed by computer. | ||
Poopi
France12902 Posts
The success of Alpha GO or how neural networks work don't mean it's an insta gg, far from it. The fact that SC depends so much on mechanics is a problem because you have to balance it, not because we want a "fair" match but because the point is to have an AI that does well in such a complex game strategically, but a lot of its complexity lies in the interface we have to use. | ||
|
Flonomenalz
Nigeria3519 Posts
I think the only way to beat it would be to come at it with a strategy it had never seen before? But even then... | ||
nimdil
Poland3748 Posts
On March 16 2016 09:40 Charoisaur wrote: SC2 is meant to be played with mechanical restraints (aka mouse and keyboard) and not through graphical input. If someone plays sc2 without those mechanical restraints he would be basically cheating. So the only fair ways for an AI vs human game would be either to have the AI played by a robot (which isn't technically possible) or simulate a robot (aka played through graphical input but with the mechanical restraints that humans/robots have). otherwise the AI would be cheating. To be fair it's just the most efficient way for Humans to interact but if we could prepare mouse, keyboard and mechanical hands (jesus that's stupid already) that yet would be able to spam on the level of thousand of actions per minute (obviously it would be custom build mouse and keyboard) - would it be "fair"? Because I don't think it'd be too hard compared to how complicated is building AI like alphago. What you would be doing is encorcing trnaslation of eletronic impulses into mechanical moves which in turn will be translated into other elektronic impuulses. I call it stupid. | ||
|
Crying
Bulgaria778 Posts
AlphaGo utilizes Machine Learning specifically to learn to play the game through numerous amounts of data. Imagine feeding AlphaGo with good amount of data (replays), where he can study and create absolutely beautiful results. The difference between chess go and starcraft is that chess and go is the following : Chess uses heuristics to provide the best move. Go uses heavy machine learning and neural networks to study games and thus come with best move/strategy Starcraft would have to use machine learning neural networks AND heuristics probably to achieve better results, also it needs huge quantity of data. The thing is, chess at the moment is pretty much absolutely dominated by computers, Magnus Carlsen for instance having an ELO of 2850~, a computer easiliy pulls 3000+, even going to 3300 with different settings. The best outcome Magnus has versus a computer is that of a draw. Chess grandmasters say they use engines to evaluate situations and see their mistakes in matches, most of them never play enginges because its pointless, while u see 20moves ahead, the engine already calculated more than 10 million. And due to the heuristic based search, it will always minimize your outcome with minimax algorithm, clever alpha-beta pruning, quintessence search and so on. Starcraft however hasn't got perfect information, thus searching is not possible, to some extent however. The amount of advancement in Machine Learning, Neural Networks and so on in the past years will probably some day bring an abomination to life. At the moment people don't really know how to make A.I. safe.(for humans) It blows my mind how in 1960's we didn't even know how to find the shortest path from A-B and nowadays we have these things. | ||
|
heqat
Switzerland96 Posts
https://en.wikipedia.org/wiki/Blue_Brain_Project | ||
|
rockslave
Brazil318 Posts
It blows my mind how in 1960's we didn't even know how to find the shortest path from A-B and nowadays we have these things. https://en.wikipedia.org/wiki/Dijkstra's_algorithm =P | ||
| ||
|
|
Sparkling Tuna Cup
Afreeca Starleague
Snow vs EffOrt
Wardi Open
PiGosaur Monday
LiuLi Cup
OSC
The PondCast
CranKy Ducklings
Maestros of the Game
Serral vs herO
Clem vs Reynor
[BSL 2025] Weekly
[ Show More ] |
|
|
EPT
