EPTAlphaStar AI goes 10-1 against human pros in demonstration…
| Forum Index > SC2 General |
BisuDagger
Bisutopia19367 Posts
| ||
|
pvsnp
7676 Posts
On January 25 2019 09:24 snakeeyez wrote: IDK if they will, but I would like to see alphastar play some more games against some of these pro players. Also it would be interesting to see the best like an innovation play against the AI. I still think what they have right here, and these results of a fair AI that is no cheating playing a full game at this level and winning is still a pretty amazing achievement. I would like to see more games where the pro players can look for more weaknesses, but the AI being completely unpredictable makes it harder to predict than a normal person Its too bad they cant transfer this AI to brood war so we could see a flash or jaedong play against it. I also think brood war is a more balanced time tested game than sc 2 I guarantee there will be many more matches, against progressively superior AI. This is just the beginning. | ||
|
ANGELIAS1234
United States46 Posts
| ||
|
snakeeyez
United States1231 Posts
| ||
|
Garrl
Scotland1978 Posts
| ||
|
pvsnp
7676 Posts
On January 25 2019 09:28 Garrl wrote: I do have to wonder if the way it plays is informed by its' mechanical ability - if we gave it infinite APM and and a client with enough framerate to support it, would it just do totally fundamentally unsound plays because it can get away with it? Yes. | ||
|
Dangermousecatdog
United Kingdom7084 Posts
On January 25 2019 06:43 Poopi wrote: Probably the more accurate description. Though the AI is technically impressive, we already know that an appropriately programmed AI with sheerly precise apm will always win over a human. The AI didn't win thorugh strategy, it won through sheer actions muscle.I think the title is kinda misleading. Raw interface AI goes 10-0, camera interface AI goes 0-1, would be more appropriate | ||
|
snakeeyez
United States1231 Posts
On January 25 2019 09:27 ANGELIAS1234 wrote: this means nothing unless they beat top Kr it means nothing I have been out of the SC 2 scene a while, but I love AI research so seeing this is pretty cool. I always wondered what the best AI we could produce could do against top level players. What types of strategies does it use and are humans playing the game wrong? Can an AI beat someone like innovation in a long 10 game set? IDK that seems tough. They said this AI had normal APM it was below that of TLO. They showed the graph. It was not cheating | ||
|
MoonyD
Australia191 Posts
The AlphaStar AI is definitely leaps and bounds better than previous AI bots, but we did see a few openings/weaknesses. | ||
|
Jerom
Netherlands588 Posts
Ultimately the AI did seem to understand the core concepts of the game, but it didnt really do anything 'genius' strategically. It would be very interesting to see this AI handicapped much more to the point where it would absolutely have to outsmart a human pro player. Still a great achievement for an AI as it did understand many core concepts. However in a sense the showcase versus a pro player felt a bit misplaced. It showed to me that an AI with only reasonable understanding of the game will still win with insane micro. It didnt really show that the AI was capable of actually solving the strategic aspects of sc2. It honestly didnt really seem to understand the game well at all. I hope they try to train it with a bigger handicap to see if it can actually play like an intelligent human being. | ||
|
Yonnua
United Kingdom2331 Posts
Then, we know that they incentivise the selection of certain units to create variety amongst agents, leading to their different 'preferences'. However, if you look at the games, those incentives are clearly so strong that the AI always over-builds that one unit, because its predicted value of any of the other units does not match the power of the incentive. So in the last game, the AI didn't build a phoenix because it was incentivised to build oracles, so it always thought that was more valuable than responding to the drop. That's clearly the biggest weakness of the AI at the moment, but as the agents play against each other more, and as the need for incentives declines, there should be a greater sophistication about which unit preferences create the most value in different scenarios. | ||
|
TheDougler
Canada8310 Posts
On January 25 2019 06:54 Poopi wrote: How is that insignificant? The raw interface didn't lose, yet the camera interface lost the first game its played. That means there is a huge actual difference between the two. Keep in mind that it's an estimated MMR, it's not like the AI played human ladder and only managed to go to 7300 MMR instead of 7500. It's the internal MMR from their internal league "The version of AlphaStar using the camera interface was almost as strong as the raw interface, exceeding 7000 MMR on our internal leaderboard." To answer your question, it's insignificant because you're committing a post-hoc ergo proctor hoc fallacy: you see that something came after something, IE the AI lost after the camera change happened, and you conclude that the second thing caused the first. It's related to the fact that "correlation does not imply causation". You don't know that it was the the camera change that actually was the determining factor here. It could be that Mana had a better idea of what he was up against. It could be that the warp prism threw off the AI's gameplan (I think it's this one). It could be that this AI isn't quite as good as other AIs. It's impossible to say, so the safest bet is to include it as part of the larger dataset rather than think of it as a case study unto itself. | ||
|
TheDougler
Canada8310 Posts
Heh, otherwise known as "The Marineking" approach to SC2. | ||
|
snakeeyez
United States1231 Posts
| ||
|
swissman777
1106 Posts
On January 25 2019 09:46 snakeeyez wrote: Have they said what kind of neural net it uses? How big is the neural net? They said it also uses reinforcement learning. We would have to wait until they give more info. Just saying reinforcement learning doesn't mean much. The best you could try to deduce things from would be the one time they tried to explain the AI's thought process with mana's and AI's pov along with other things on the screen. | ||
|
pvsnp
7676 Posts
On January 25 2019 09:46 snakeeyez wrote: Have they said what kind of neural net it uses? How big is the neural net? They said it also uses reinforcement learning. CNN and RNN, iirc | ||
|
PharaphobiaSC
Czech Republic457 Posts
Thanks in advance | ||
|
agsub
Singapore368 Posts
On January 25 2019 09:46 snakeeyez wrote: Have they said what kind of neural net it uses? How big is the neural net? They said it also uses reinforcement learning. | ||
|
snakeeyez
United States1231 Posts
| ||
|
ZigguratOfUr
Iraq16955 Posts
These results suggest that AlphaStar’s success against MaNa and TLO was in fact due to superior macro and micro-strategic decision-making, rather than superior click-rate, faster reaction times, or the raw interface. This conclusion isn't particularly convincing. Deepmind's research is great, but their propendency for PR stunts is wearing. | ||
| ||
