A Google software programme has won the second game in a landmark contest against a South Korean champion in the board game Go.
Computer programme AlphaGo beat Lee Sedol after four and a half hours of play on Thursday, in a five-game series seen as one of the biggest tests of the power of artificial intelligence (AI).
AI systems had long struggled to beat humans at Go, which was previously thought to be too complex for computers, as they lacked the creative, self-learning software needed to adapt to the millions of possible gameplay scenarios.
"Yesterday I was surprised, but today it's more than that — I am speechless," 33-year-old Lee was quoted as saying by The Verge technology website after the game.
Lee now needs to beat AlphaGo in all three remaining games to claim the prize money of 1 million dollars, which will be donated to charity if the software wins.
The rules of Go, which originates in China, are easy in principle: two players try to conquer spaces with white and black stones on the board, which is divided by vertical and horizontal lines into a grid.
AlphaGo was developed by Google's DeepMind AI branch to forecast the most probable moves of its human opponent and adjust itself accordingly.
"AlphaGo played some beautiful creative moves in this game," DeepMind founder Demis Hassabis wrote on Twitter.
In October, the programme won a five-game match against Fan Hui, a top European player, ranked well below Lee internationally.
The contest in Seoul is due to run until March 15.