News JVTech Unable to defeat an AI in a board game, he gets help from an unexpected assistant.
Published on 02/21/2023 at 12:40
Share :
A very experienced amateur Go player has defeated a very high-level artificial intelligence. He was aided by a bot, which discovered a flaw in his opponent.
Kelline Pelrine, an amateur Go player who won a competition against the KataGo system, is a great example of a human triumph, one of the fewest since AlphaGo's triumph in 2016.
Kelline Pelrine's accomplishment isn't a total loss. According to the Financial Times, the player probably wouldn't have gotten to victory without the assistance of FAR AI's worka company that has developed a program capable of detecting the flaws in KataGo.
When humans join forces with an AI to defeat another, it's a game changer.
The bot from FAR AI has played over a million games against KataGo. This allowed him to reveal a vulnerability in the way of playing artificial intelligence that was sufficiently powerful for an amateur player of any level to exploit.
The player then created a kind of "loop" with his stones to encircle those of the AI, then he distracted her by making moves in other sections of the board. KataGo became trappednot realizing that his own stones were being surrounded.
Kelline Pelrine says she expects a human to recognize this defect as quickly as it does. The whole aim was to locate this flaw, which is straightforward to exploit once it is discovered.
A human triumph that appears to be halfhearted, yet...
Pelrine's victory is certainly not entirely due to his abilities as a player, but he was able to exploit this flaw to win. As explained in a 58-page paper published shortly after the completion, the FAR AI experiment is successful on this point.
Players who face KataGo online regularly exploit its flaw, according to Lightvector, the game's creator, who has already announced its intention to patch it. Ultimately, KataGo will emerge stronger as a result of this experience, and players will soon face new challenges to beat the AI.
