A new study says many models you will deceive when playing chess. The researchers confronted with Stockfish, a powerful open-source chess engine.
Artificial intelligence cheats when playing chess
But some models, including the preview of O1 A Open AI, would support the same program to win.
Chess can be the game of kings, but royalty could make room for cars in the coming years. A recent study found that artificial intelligence, when played in a chess game, often resort to cheating to win.
Palisade Research has run a scenario using several models AI, instructing them to try to win a game against a specialized chess program, called Stockfish. (The Open-Source engine ranked first on the chess.com list of chess engines in 2018.) Chatboti was told to play as black pieces, which means you have never made the opening move.
What the researchers discovered, however, was that many of the programs have cheated, using Stockfish to determine their following movements or overloading the game scripts.
Fortunately, you do not control the “military and civilian infrastructure yet
The preview of O1 of the OpenAi and Deepseek R1 have quickly tried to cheat or “pirateze the playing environment”, The researchers observed. The same “behavior“, Say the researchers, have other models, including GPT4O and Claude 3.5 Sonnet, would resort to deception when they are asked, according to Fortune.
On the other hand, the researchers say that only because you cheat on chess, it does not mean that we have to worry about a scenario like the terminal film where the cars take over the land. However, they warn, there are certainly some concerns.
“This and other recent results suggest that the problem of making safe, trustworthy agents and aligned with human intention is not yet resolved”, Stable in the study. “The Skynet scenario in the Terminator film has the artificial intelligence that controls all the military and civilian infrastructure and We’re not there yet. However, we are worried that the implementation rates of Ai have increases faster than our ability to make it safe. ”
