Google's Deepmind AI claims chess crown, Starcraft is next!

freaky

Banned
Banned
Joined
Mar 9, 2015
Messages
6,308
Reaction score
0
http://www.bbc.com/news/technology-42251535
Google says its AlphaGo Zero artificial intelligence program has triumphed at chess against world-leading specialist software within hours of teaching itself the game from scratch.

https://deepmind.com/blog/deepmind-and-blizzard-open-starcraft-ii-ai-research-environment/

The London-based team is currently trying to develop a system that can beat humans at the space strategy game Starcraft, which is seen as being an even more complex challenge.

Not sure about you guys, but I'm a huge starcraft nerd and I don't see AI beating a human. The creativity and micro management from a human will always be superior to a bot. It would be crazy if an AI can beat a top SC player. Soon, we will have AI generals advising our army what to do in warfare.

This is the future you guys. This is crazy!!!

 
If you think they can't program AI to make the same decisions as the best players, you underestimate the power of data analysis.
 
I used to play chess
stoned.gif
 
If you think they can't program AI to make the same decisions as the best players, you underestimate the power of data analysis.

I highly doubt an AI from scratch will know to build cannons outside the opponents base or to throw feints or proxy etc. But I'm sure they can do everything if we feed the AI history of all the different plays then yea.

But some of the plays in SC is just so creative and out of the blue, that I don't think an AI who thinks linearly will think outside the box and try to trick their opponent. It would take them a long time to learn about human trickery. But once they do..my god...like I said, I can see AI being the general giving warfare advises.
 
If you think they can't program AI to make the same decisions as the best players, you underestimate the power of data analysis.

They only programmed the AI to learn. They then specify the rules of the game (whatever it happens to be - chess in this case) and it learns how to play at superhuman levels by playing itself....

Once they manage to get the AI to learn at this level in open problems, humanity will be left behind and only be able to look on in amazement as computers surpass us in almost all intellectual pursuits.
 
I highly doubt an AI from scratch will know to build cannons outside the opponents base or to throw feints or proxy etc. But I'm sure they can do everything if we feed the AI history of all the different plays then yea.

But some of the plays in SC is just so creative and out of the blue, that I don't think an AI who thinks linearly will think outside the box and try to trick their opponent. It would take them a long time to learn about human trickery. But once they do..my god...like I said, I can see AI being the general giving warfare advises.

A.I. already beat a world class GO player which is one of the most complex games in existence.
 
Starcraft 1 or 2; cause I will power fuck a computer on starcraft 1. They already have AI's you can play against in the starcraft.
 
/build zergs
/rush zergs

/ggs pathetic humans
 
holy shit its gonna play starcraft? if its brood war i cant take him down i used to have sick micro with terran
 
Stockfish has been the undisputed computer chess program for a while now and to see Google's AI demolish it is extraordinary stuff as it is. The fact that it learned chess and then beat Stockfish within 4 hours is absolutely insane and really shows how scary these AI's can be. Elon Musk's team built an AI that crushed world champions in DOTA and there is no doubt in my mind that Google's AI can conquer Starcraft.
 
Starcraft 1 or 2; cause I will power fuck a computer on starcraft 1. They already have AI's you can play against in the starcraft.

Not like AlphaGO/Deepmind. Its only a matter of time before A.I. can beat any human in any game. The A.I. community thought it would be another 20-30 years before an A.I. could beat a worldclass GO player and it happened last year.
 
They only programmed the AI to learn. They then specify the rules of the game (whatever it happens to be - chess in this case) and it learns how to play at superhuman levels by playing itself....

Once they manage to get the AI to learn at this level in open problems, humanity will be left behind and only be able to look on in amazement as computers surpass us in almost all intellectual pursuits.

Yeah it's definitely the future. We'll have computers solving all of our problems with impressive algorithms and artificial intelligence.
 
Starcraft is good as a next step because you have imperfect information, you can't see the whole map. Also it's real time with an untold number of possible things happening at any time.

I'm sure they could write a piece of code to play 'perfect' Starcraft with impossible micro and macro, but they want the AI to learn its way to being better than SC pros.
 
Starcraft is good as a next step because you have imperfect information, you can't see the whole map. Also it's real time with an untold number of possible things happening at any time.

I'm sure they could write a piece of code to play 'perfect' Starcraft with impossible micro and macro, but they want the AI to learn its way to being better than SC pros.

1 billion APM :eek:
 
I highly doubt an AI from scratch will know to build cannons outside the opponents base or to throw feints or proxy etc. But I'm sure they can do everything if we feed the AI history of all the different plays then yea.

But some of the plays in SC is just so creative and out of the blue, that I don't think an AI who thinks linearly will think outside the box and try to trick their opponent. It would take them a long time to learn about human trickery. But once they do..my god...like I said, I can see AI being the general giving warfare advises.
I remember reading about one of the GO matches that the AI won. As the game was being played the computer made a move that everybody immediately thought was a fuckup and the human would take it. By the end of the game (which the computer won) it turned out that move was a key point for its victory. "Outside the box" is one of the things these kind of programs tend to specialize in because they generally build up their strategy from scratch without relying on established play styles.
 
I read they're limiting the AI to human levels of APM, so it has to outthink you.
 
Back
Top