Atari 2600 game console beats ChatGPT at chess

Modern AI lost to an old video game console at chess. Citrix engineer Robert J. Caruso conducted an unusual experiment in which he forced ChatGPT to play a game against an Atari 2600 console running the game Video Chess. As he reports in his LinkedIn post, the victory went to the 8-bit machine from the 1970s.

According to Caruso, it all started with a discussion with ChatGPT about the history of AI in chess. The model agreed to play against Video Chess, expressing interest in how quickly it could beat a primitive program that calculates moves a maximum of two steps ahead. The console emulation was carried out through Stella.

However, ChatGPT quickly began to make serious mistakes. The model confused pieces, got lost on the board, and did not notice elementary chess threats. At first, she blamed the Atari version of the game for the overly abstract icons, but even after switching to standard chess notation, the situation did not improve. The engineer notes that the game took 90 minutes, during which he had to intervene to stop incorrect moves and remind the models of the current position of the pieces.

Meanwhile, the old Video Chess, despite its minimal computing resources – just 128 bytes of RAM and a 1.19 MHz processor – played chess consistently and without error. Caruso emphasizes that there was no language model or machine learning in the Atari, just a simple brute force algorithm and the stubbornness of 1977.

In the end, ChatGPT admitted defeat. Caruso jokingly concluded that a third-grade chess club might make fun of such a game. For reference: the Atari 2600 console was produced from 1977 to 1992. The game Video Chess was released in 1979 and was developed by Larry Wagner and Bob Whitehead, who later moved to Activision.