In March 2016, the AlphaGo software beat one of the best Go players on the planet, Korean Lee Sedol, before doing it again in May 2017 by defeating Chinese world champion Ke Jie. These two victories marked the meteoric rise of artificial intelligence (AI). But according to the FinancialTimes, an American player recently avenged them by winning against the machine.
Yet far from the world’s top 10, Kellin Pelrine has won against the software developed by DeepMind, the famous Google lab. But the American player did not succeed alone: he was helped in this task by… an artificial intelligence.
14 wins out of 15 games
Indeed, it is a software developed by the company FAR AI which managed to find the fault. The latter trained extensively against the open-source AI KataGO, which was itself developed from the AlphaGo software. Once the weaknesses were identified, Kellin Pelrine only had to follow the instructions of the AI to win 14 times in 15 games.
According to him, the technique used by his artificial ally “would be quite easy for a human to spot”. But while this win challenged AlphaGo’s dominance, it also exposed its weaknesses in the eyes of the designers. What will happen once the flaw is corrected?