Copilot’s Overconfidence: A Humbling Defeat Against Atari’s Video Chess

Microsoft’s Copilot Takes on Atari 2600 Video Chess: A Lesson in AI Hubris

In a surprising twist of fate, Robert Caruso, the tech enthusiast known for pitting AI against classic games, has once again turned the spotlight on artificial intelligence. This time, he set Microsoft’s Copilot against the Atari 2600’s Video Chess, hoping to see if the AI could outperform ChatGPT, which had previously suffered a humiliating defeat.

Caruso’s experiment began with a pre-game chat, where he outlined the pitfalls that led to ChatGPT’s loss. The primary issue? The inability to keep track of the chessboard. With a confidence that only an AI could muster, Copilot assured Caruso that it could not only play chess but excel at it. “I can think 10–15 moves ahead,” it claimed, although it opted for a more conservative 3–5 moves against the vintage game, citing the Atari’s tendency to make “suboptimal moves.”

As the game commenced, Copilot boasted about its ability to remember previous moves and maintain continuity. However, it soon became clear that its confidence was misplaced. Despite its claims, Copilot lost two pawns, a knight, and a bishop, while the Atari was only down a single pawn. When Caruso compared the board’s current state to the last screenshot, Copilot admitted they were different, echoing the earlier failures of ChatGPT.

“ChatGPT déjà vu,” Caruso remarked, highlighting the recurring theme of AI overconfidence. In the end, Copilot graciously acknowledged its defeat, tipping its digital king to the Atari 2600, which had once again proven its prowess.

This amusing experiment serves as a reminder of the limitations of generative AI systems. While they may be trained on vast amounts of data, their understanding of complex strategies—like those required in chess—remains flawed. The moral of the story? Beware the confidence of chatbots; they may excel in many areas, but when it comes to a 45-year-old chess game, they still have a long way to go.