EPTAlphaStar AI goes 10-1 against human pros in demonstration…
| Forum Index > SC2 General |
|
outscar
2832 Posts
| ||
|
bela.mervado
Hungary409 Posts
seriously, amazing games, can we have these AIs in a future patch Blizz plz  | ||
|
Rodya
546 Posts
On January 25 2019 07:17 No_Roo wrote: As many others are pointing out, there's not much to be impressed with while the AI is playing a different game. We already know that a microbot using an API can outmicro a human using a mouse. I look forward to seeing what they come up with in the future, but what I saw today was no more interesting than if they had been feeding the AI bonus minerals. Also looking forward to detailed replay analysis, I expect that will reveal that "APM" is a very misleading metric to be using. Exactly; from what I've glanced at the games this isn't very interesting. Perfect orb-walking and dropship/blink micro. If humans could do that then all dropships would be heavily nerfed and blink would be removed from the game. I'm sorry Google - no one is shocked nor cares about this. Sadly their marketing department will continue to push this bot down our throats while they tell themselves that they've revolutionized AI. User was warned for this post | ||
Heyoka
Katowice25012 Posts
| ||
|
Shuffleblade
Sweden1903 Posts
| ||
|
StormsInJuly
Sweden165 Posts
| ||
|
Gianttt
Netherlands194 Posts
I'm also a bit sad that it wasn't smart enough to saturate his base better, which is one of the first things you learn playing Starcraft. | ||
|
FueledUpAndReadyToGo
Netherlands30548 Posts
Does this thing run on some supercomputer mainframe or just a regular i7? | ||
|
pvsnp
7676 Posts
On January 25 2019 07:30 Rodya wrote: Exactly; from what I've glanced at the games this isn't very interesting. Perfect orb-walking and dropship/blink micro. If humans could do that then all dropships would be heavily nerfed and blink would be removed from the game. I'm sorry Google - no one is shocked nor cares about this. Sadly their marketing department will continue to push this bot down our throats while they tell themselves that they've revolutionized AI. User was warned for this post Do you have any technical understanding whatsoever of AI? Or are you just talking out of your ass? This is an impressive technical achievement. I wouldn't call it revolutionary, since the algorithms Deepmind is working off of are fairly mainstream, but this was still a great first showing. | ||
Poopi
France12944 Posts
On January 25 2019 07:43 FueledUpAndReadyToGo wrote: Crazy... Never thought it would be this effective Does this thing run on some supercomputer mainframe or just a regular i7? The training is done on 16TPU if I recall correctly (<=> optimized GPU for machine learning computations) but the agent that plays against humans is made to be able to run on a single desktop GPU (probably a powerful one tho, latest most powerful nvidia would be my guess) | ||
|
yep
Canada9 Posts
I was really impressed and I think it is much more than a simple microbot. It knew when to push and when to back off, meaning he could size its opponents' army size, he targetted a warp prisms with phoenixes while not over-committing them because he knew the strength of warp prisms in an attack. He also knew when to expand by seeing its opponent army in the final showmatch. The A.I. also improved quite significantly in between the games vs TLO and the one versus Mana. He clearly learned that blocking its entrance was useful and that ramps were also dangerous. While the A.I didn't know how to deal with Mana double immortal drop, it might be simply because he did not practice against this situation before. However, I'm pretty sure it is something it can learn since he clearly knew how to deal with mineral line harassment by leaving stalker in its mineral line. | ||
|
deacon.frost
Czech Republic12129 Posts
On January 25 2019 07:46 yep wrote: Wow, I wasn't expecting people bashing on the A.I by saying it is only a good microbot and that "no one is shocked nor cares about this". I was really impressed and I think it is much more than a simple microbot. It knew when to push and when to back off, meaning he could size its opponents' army size, he targetted a warp prisms with phoenixes while not over-committing them because he knew the strength of warp prisms in an attack. He also knew when to expand by seeing its opponent army in the final showmatch. The A.I. also improved quite significantly in between the games vs TLO and the one versus Mana. He clearly learned that blocking its entrance was useful and that ramps were also dangerous. While the A.I didn't know how to deal with Mana double immortal drop, it might be simply because he did not practice against this situation before. However, I'm pretty sure it is something it can learn since he clearly knew how to deal with mineral line harassment by leaving stalker in its mineral line. I believe it all comes down to not understanding the technology behind it and how the program operates. I am very amazed and I hope they already learned that there are people on the internet which won't like what they do. I liked it, very much. We need a company, which will build the Skynet and kill the humanity! Learning it Go and Starcraft seems like a plausible start. | ||
|
Xain0n
Italy3963 Posts
Seriously speaking, AlphaStar was indeed impressive but I wasn't satisfied with it inhumanly microing blink stalkers during the most convincing victory and falling to immortal drops when it eventually lost; however, it's already way, way better than any AI in terms of decision making. I actually don't get why didn't they let a single agent play more than one game to show its capability of learning from its mistakes; is it still to a point where it needs thousand of games to develop and effective change in strategy? | ||
|
pvsnp
7676 Posts
On January 25 2019 07:52 deacon.frost wrote: I believe it all comes down to not understanding the technology behind it and how the program operates. I am very amazed and I hope they already learned that there are people on the internet which won't like what they do. I liked it, very much. We need a company, which will build the Skynet and kill the humanity! Learning it Go and Starcraft seems like a plausible start. Well said. The only people unimpressed by this are those too ignorant of computer science to understand what it means. Reminds me of: "Congressman, the iPhone is made by a different company." - Sundar Pichai, Google CEO | ||
|
geokilla
Canada8248 Posts
| ||
|
deacon.frost
Czech Republic12129 Posts
On January 25 2019 07:55 pvsnp wrote: Well said. The only people unimpressed by this are those too ignorant of computer science to understand what it means. Reminds me of the US Congress trying to ask Sundar Pichai about Google making iPhones. I wouldn't be so harsh on them. In the end, we may be in their position in other science fields  On January 25 2019 07:59 geokilla wrote: They had TLO off race and then didn't get any good pros to face against. I don't mean TLO or Mana aren't good pros, but it would have been cool to see AlphaStar vs Gumiho or Scarlett or Elazer for example. MaNa got 2nd place in the last year at some WCS I believe. Where was Scarlett at that time?  No offense to anyone, but MaNa is a very strong WCS player, maybe even stronger than SCarlett(don't take my word for it, I am an ignorant of WCS after all ). I take MaNa and TLO. I would rather saw sOs, Stats or Classic but they won't speak that much English(show) and it would be costly to fly them there | ||
|
imre
France9263 Posts
On January 25 2019 06:52 Waxangel wrote: Honestly I feel that StarCraft II is much more of a mechanics game than 'strategy' game than most people think, (it's more real-time connect-four than chess), so I also felt this demonstration wasn't quite as impressive as it could have been. At least for my personal taste, the truly impressive way to hold his exhibition would be to have an AI that's actually considerably BAD mechanically but still crushes human opponents based on strategies and tactics (and, I guess, EXTREME efficiency of actions). This was both awesome since the decisions weren't that bad but the whole thing looked staged due to the micro the IA had. | ||
|
pvsnp
7676 Posts
On January 25 2019 07:59 deacon.frost wrote: I wouldn't be so harsh on them. In the end, we may be in their position in other science fields Well, when I start spewing bs from ignorance, feel free to correct me. | ||
|
TheDougler
Canada8310 Posts
ZvT from AlphaStar will be especially fascinating to watch. | ||
|
9Dimitri
Hungary14 Posts
 | ||
| ||
