In a surprising turn of events, Google’s state-of-the-art Gemini chatbot realised it couldn’t compete with the powerful Atari 2600 in a game of chess and declined to play. With its meagre 1.19MHz CPU and 128 bytes of RAM, the Atari 2600 has already shown that it is capable of outperforming other sophisticated AIs, such as Microsoft Copilot and ChatGPT.
Infrastructure architect Robert Caruso, who had previously compared Atari Chess to a number of general-purpose chatbots, came up with the idea for the matchup. At first, the AI was confident in its ability to easily defeat Atari Chess when he presented the same challenge to Gemini. With its sophisticated reasoning capabilities and ability to anticipate millions of moves, Gemini—which was based on Google’s recently developed multimodal large language model—positioned itself as more like a contemporary chess engine than a straightforward chatbot.
Caruso quickly shared his prior experiences with the Atari 2600’s unexpected triumphs over other AIs, though. After taking the facts into account, Gemini had a rare moment of humility. It acknowledged that it would probably “struggle immensely” against the Atari 2600 and that it had overestimated its chess prowess. The AI then determined that the “most time-efficient and sensible decision” was to end the match.
Caruso commended Gemini for acknowledging its limitations, pointing out that these reality checks are crucial to establishing AI systems’ dependability and credibility. Although Gemini is an effective tool, he underlined that AIs must remain grounded in reality, particularly in scenarios where mistakes could have serious repercussions.
Gemini’s choice serves as a reminder that even the most sophisticated systems have limits, particularly when facing the venerable Atari 2600, in a world where artificial intelligence is frequently hailed as unbeatable.