EPTDeepMind sets AlphaGo's sights on SCII - Page 3
| Forum Index > SC2 General |
|
DinosaurPoop
687 Posts
| ||
|
teamamerica
United States958 Posts
I'd love to see what strategies the bot decides on with different contraints such as max apm, playing different races, playing different maps too! | ||
]343[
United States10328 Posts
On March 28 2016 17:54 teamamerica wrote: Deepmind bot vs archon mode maru/innovation is obviously why they're using sc2 instead of bw. um you know starcraft 1 has always had team melee, right edit: i'm guessing you weren't being serious there, but bruh "archon mode" literally always existed in sc1 | ||
|
Yonnua
United Kingdom2331 Posts
Goody will be Earth's champion. | ||
|
ilikeredheads
Canada1995 Posts
| ||
|
sharkie
Austria18726 Posts
| ||
|
Cyro
United Kingdom20343 Posts
On March 28 2016 13:09 chipmonklord17 wrote: The only way I see APM being limited would be if you did the following 1) Choose a designated 'esports keyboard' (some razer/steelseries/ect product) 2) Find the maximum amount of inputs the keyboard recognizes per minute 3) Make that the APM 'limit' Each keyboard will stop responding after a certain number of inputs per second, and said input limit is probably much lower than the input limit that AlphaGo would have. At least I assume so. You can register 1300APM by holding down a rapid fire key! | ||
|
Garbels
Austria653 Posts
On March 28 2016 17:49 DinosaurPoop wrote: Why do we need an apm limit you guys seriously overestimate the power of perfect micro. I don't think anyone is even close to overestimating perfect micro. If anything people are not giving enough thought to the impact perfect micro(even with limited apm) will have on strategy. | ||
|
Theosiel
France9 Posts
On March 28 2016 16:57 TheoMikkelsen wrote: Unless you limit AlphaGo it should be extremely easy to beat the best players in the world. All you need is to make AlphaGo Terran, give it medivacs, stim, and the ability to micro each individual marine perfectly. Seriously it should be the easiest task ever. Just make it like open reaper expand every game, then it only has to counter some cheeses. Go does not have mechanics, Starcraft 2 does, and there the AI would be infinitely superior. The point is precisely NOT THAT. If it's about implementing an AI that will go for a perfectly micro-ed cheese everytime, well, the technology is already here  The point of "AlphaStar" is that there is no "making" involved. The system must learn by itself the importance of scouting, the usefulness of cheese and what have you, just like a human beginner would, if he only knew the goal (make the opponent leave the game or destroy every building) and had a huge database of replays to watch. The choice of interface that would be used is essential. A keyboard, mouse and screen set up (even virtual) means that the AI has to move the camera to issue commands, and is limited by the max mouse sensibility of the engine, which would effectively limit the APM count. There is also the question of how APM will be perceived by the system through its learning phase. We think of automaton2000 and its godlike APM, but a "self-taught" AI may very well end up playing with a relatively low APM. Or spam keystrokes just as much as the worst of human APM spammers. The questions that we (the starcraft community) ask about this AI are questions that DeepMind's people will have to think about, and make choices according to what they want their AI to do. Food for thoughts: given the "right" replays, it's entirely possible that "AlphaStar" might try to make the opponent leave by BMing in chat. The BM to end all BM, now wouldn't that be a funny sight? | ||
|
neptunusfisk
2286 Posts
On March 28 2016 18:32 Garbels wrote: I don't think anyone is even close to overestimating perfect micro. If anything people are not giving enough thought to the impact perfect micro(even with limited apm) will have on strategy. Yeah, it would make any worker rush situation where the attacker has more workers for a brief moment a reliable way to win | ||
|
Flingoko
Denmark68 Posts
| ||
|
Circumstance
United States11403 Posts
| ||
|
Mendelfist
Sweden356 Posts
On March 28 2016 18:42 Theosiel wrote: Food for thoughts: given the "right" replays, it's entirely possible that "AlphaStar" might try to make the opponent leave by BMing in chat. The BM to end all BM, now wouldn't that be a funny sight? Indeed. Alphago played moves that no human would play, moves that shocked the Go community and made them question their knowledge of the game. What kind of BM would a super-AI come up with? Be afraid. | ||
|
spinesheath
Germany8679 Posts
On March 28 2016 17:54 teamamerica wrote: I'm interested in how they are going to expose game data to the bot e.g. will they teach it to recognize images drawn on screen That's pretty much what these Google AIs have been doing all this time. | ||
|
Cyro
United Kingdom20343 Posts
The choice of interface that would be used is essential. A keyboard, mouse and screen set up (even virtual) means that the AI has to move the camera to issue commands, and is limited by the max mouse sensibility of the engine, which would effectively limit the APM count. That doesn't actually limit nearly as much as you would think | ||
|
MockHamill
Sweden1798 Posts
Starcraft is so much more complex than Go it is absurd. Basically the amount of permutations in the first 3 minutes of Starcraft is larger then any match of Go. | ||
|
Elentos
55593 Posts
On March 28 2016 18:59 Mendelfist wrote: Indeed. Alphago played moves that no human would play, moves that shocked the Go community and made them question their knowledge of the game. What kind of BM would a super-AI come up with? Be afraid. All this would lead to is a squad of KeSPA referees running in and disqualifying the AI. | ||
lichter
1001 YEARS KESPAJAIL22272 Posts
On March 28 2016 19:19 Elentos wrote: All this would lead to is a squad of KeSPA referees running in and disqualifying the AI. if he actually watches too many replays and learns to type more than glhf and gg then he definitely will get DQ'd by kespa | ||
|
sertas
Sweden890 Posts
| ||
|
Mendelfist
Sweden356 Posts
On March 28 2016 19:15 MockHamill wrote: Starcraft is so much more complex than Go it is absurd. Basically the amount of permutations in the first 3 minutes of Starcraft is larger then any match of Go. As someone who has played chess (no rank), Go (10 kyu) and Starcraft (platinum) at low amateur level I say you are wrong. At the strategic level Go is enormously more complex and deep than both chess and Starcraft. There is absolutely no comparison. Go completely dwarfs any game that I have come in contact with. Measuring complexity in "number of permutations" is not meaningful. How many permutations are there in a game of billiard? Do you think it would be hard for a computer? | ||
| ||
