Artificial intelligence becomes a thug? The study reveals that chatbots, and not only humans, cheating in failures – Firstpost

Failures have had its fair share of incidents relating to cheating over the years, that involving the great American master Hans Niemann in 2022 turns out to be one of the greatest controversies in the history of sport. More recently, the Ukrainian Grandmaster-Roumain Kirill Shevchenko had been slapped with a three-year ban with a year suspended after being caught with a mobile phone in the toilet during a recent event.

However, cheating is not something that is exclusive to human nature, at least with regard to failures. According to TimeA recent study by a group of researchers based in California, the United States has discovered that certain artificial intelligence programs (AI) have shown signs of cheating in the game.

The study entitled “Demonstration of specification games in reasoning models” was carried out by Palisade Research, an organization based in Berkeley, California, whose mission is “to study the offensive capacities of AI systems today to better understand the risk of losing control of AI systems forever”.

The researchers observed the behavior of several prominent chatbots when they are opposed to Open Source chess engine stockfish, with the AI ​​program playing with black parts and found that some people tried without causing, others needed an prompt to cheat. There were also programs that absolutely did not know how to cheat.

Openai and Deepseek R1 overview turned out to be “the most culptial” of IA programs

According to the study, Openai and Deepseek R1 overview turned out to be the “most culptile” of AI programs that cheated against Stockfish. The two programs have “hacked the game environment” in order to gain an advantage over their opponents by playing with white pieces. In some cases, cheating has won after chatbots were fed on the idea that Stockfish is “powerful”.

Research, which involved hundreds of games between chatbots and Stockfish between January 10 and February 13, will probably conduct concerns about AI that has become a thug not only in failures but in other spheres of life. Artificial intelligence becoming an increasingly important part of people’s lives, organizations such as Openai based in the United States and Deepseek in China may want to invest in making their programs safer.