AI algorithm excels at StartCraft II strategy game
Impressive results in a game that’s even more complex than chess or Go
Following victories over humans at chess, Go and poker, an artificial intelligence algorithm has notched up yet another gaming milestone in defeating some of the world's best StarCraft II players.
An article in the scientific magazine Nature has detailed how a team from London-based AI company DeepMind developed a programme called AlphaStar which is capable of ranking in the top 0.2 per cent of all human players for the e-sport. The programme used a technique called reinforcement learning, enabling the algorithm to teach itself effective strategies and counter-strategies.
David Silver, a principal research scientist at DeepMind, which is owned by Google, claimed that StarCraft II is more of a challenge than other board or strategy games and it is one that has been perplexing scientists for as long as 15 years.
“The game's complexity is much greater than chess, because players control hundreds of units; more complex than Go, because there are 1,026 possible choices for every move; and players have less information about their opponents than in poker,” he told Nature.
“We see StarCraft as a benchmark domain to understand the science of AI and advance in our quest to build better AI systems.”
Overall, AlphaStar’s technique has impressed professional StarCraft players. Raza “RazerBlader” Sekha, who is one of the UK’s top three StarCraft II professionals, played as a Terran against AlphaStar and also watched its matches against others. He concluded that the neural networks were impressive but that AlphaStar often took an unusual approach.
“There was one game where someone went for a very weird [army] composition, made up of purely air units – and AlphaStar didn’t really know how to respond,” he told the BBC.
“It didn’t adapt its play and ended up losing. That’s interesting because good players tend to play more standard styles, while it’s the weaker players who often play weirdly.”