facebook rss twitter

Starcraft AI needs several more years work to beat best humans

by Mark Tyson on 14 March 2016, 11:46

Tags: Activision Blizzard (NASDAQ:ATVI), Google (NASDAQ:GOOG)

Quick Link: HEXUS.net/qaczdm

Add to My Vault: x

The battle between humans and artificial intelligence has been in the headlines over the weekend. Most of the stories have been focussed upon the intellectual wrestling match between world champion Go player Lee Sedol and the Google DeepMind developed AlphaGo AI system. But what about the rise of AI in a game HEXUS readers might more likely to have played on their PCs – Starcraft?

In the ancient, and 'profoundly complex', board game of Go the AI player appears to be sitting on the brink of being unconquerable. In January we reported on AlphaGo playing and winning against accomplished human players for the first time. On Sunday, 33 year old Lee Sedol won a match against AlphaGo, denying the AI a clean sweep of wins in the proposed five match series. Until recently it was predicted an AI wouldn't be able to challenge a professional Go player for at least 10 years.

Speaking to the Korea Times, in the wake of the AlphaGo news, a computer engineering professor still thinks such an AI feat in Starcraft remains several years away. Professor Kim Kyung-joong from Sejong University told the newspaper that while AI can surpass humans in unit control, operating 100x faster in that respect, a telling part of a winning player's skill is in strategic decisions. Kim asserted that the Starcraft AI still is way behind humans in blueprinting strategies as well as making unplanned actions spontaneously.

Describing the AI challenge more specifically "No matter how well Starcraft AI may control a 12-unit group of Zealots ― an infantry unit from Protoss, one of the game’s three species ― the computer has always been short of winning whenever humans countered with a stronger combination of different units or infiltrated AI's main base with a surprise attack like using a dropship ― a transport unit from Terran, another species of the game," said Kim.

Prof Kim's research team developed a Starcraft AI called Xelnaga, one of the top AI engines for the game with a 50-60 percent winning rate until 2012. The most recent Starcraft AIIDE tournament was won by an independent AI bot developer, Vegard Mella using the Zerg race, with Xelnaga (Protoss) placed 9th.

HEXUS Forums :: 3 Comments

Login with Forum Account

Don't have an account? Register today!
Are these AI limited to human like actions per second?
Are these AI limited to human like actions per second?

Watching some of the Koreans makes me reassess what “human-like actions per second” is ;)
I remember reading somewhere that some of the APM werent any actual important ones but they had to keep their APM up as a feint or their real ones would be more obvious. Can't remember exactly where I read it…