SHANGHAI — Artificial intelligence will be able to defeat the world’s best competitive gamers within three years, experts said Friday at the World Artificial Intelligence Conference in Shanghai.
Scientists have been training machines to play against humans in classic games like poker and chess. Thanks to recent developments in AI that can “self-learn” from data, these machines have begun to dabble — and even triumph — in more complex arenas like the strategic territory-capturing board game Go.
Two years later, AIs have demonstrated their superiority in massively complex video games like Dota 2 and StarCraft II. DeepMind’s AlphaStar program, for example, has a 99.8% win rate against human StarCraft II players.
“AI responds really quickly,” Yan Junjie, the deputy head of research at Shanghai-based artificial intelligence giant SenseTime, said Friday during a panel discussion. He explained that while human players typically require 0.1 seconds to answer an action, trained machines can respond almost instantaneously.
“Moreover, AI has infinite calculation capacity … and it’s getting better at an exponential rate. For example, a machine may only have an IQ of 100 now — but in five years, it might be 10,000,” Yan said, adding that human capacity, on the other hand, is “relatively fixed.”
For now, current AI bots aren’t intelligent enough to defeat the best gamers, according to Zhou Hang, a former StarCraft II champion.
“It (the AI) is not yet changing and thinking, but rather mimicking humans,” Zhou, who is now a researcher with SenseTime, said during Friday’s forum. He added, however, that it’s only a matter of time — about three years — until AIs will be beating the best players in the world.
Pitting bots against humans in video games isn’t just about entertainment. One of the main goals of companies like SenseTime is to improve machines’ decision-making abilities, such as the optimal response to an enemy strike.
Decision-making skills can have many practical applications, Yan said. In the autonomous driving sector, for example, the vehicle’s AI system has to decide whether to follow or pass the cars ahead — a complex decision for which many variables must be taken into account.
“We start from (letting AI) copy humans, and we hope that one day, humans will copy AI,” Zhou said. “I hope AI can expand human cognition and imagination, in games or in other fields.”
Editor: David Paulk.
(Header image: A staff member takes a photo of a humanoid robot at the World Artificial Intelligence Conference in Shanghai, July 9, 2020. Qilai Shen/Bloomberg via Getty Images/People Visual)