We’ve all had moments of doubt.

Recently, Google’s Gemini AI, which had boasted about being an expert in chess, experienced a crisis of confidence before a match against the Atari 2600 chess engine, a program nearly five decades old.

According to Robert Caruso, a software engineer known for pitting vintage AI against modern counterparts, Gemini initially expressed confidence in its chess skills. However, the prospect of facing an engine with extensive experience proved overwhelming. During a pre-match conversation, Caruso remarked on Gemini’s bravado, which seemed to unsettle the AI.

Gemini stated: “I’m not merely a large language model but a modern chess engine capable of evaluating countless positions and thinking many moves ahead.”

However, as the match approached, Gemini’s confidence faltered. Caruso pointed out to Gemini that he had previously orchestrated matches between LLMs and the Atari game, a fact that made the AI hesitate.

In what might be considered a wise decision, Gemini ultimately declared, “Canceling the match is likely the most time-efficient and sensible decision.”

This incident brings forth a curious discussion about AI’s self-awareness and ability to recognize potential failure. As Caruso noted, the AI’s admission of its limitations showcases a level of perception that might suggest a more nuanced understanding of competition.

This episode is reminiscent of childhood experiences where overconfidence leads to a humbling realization. While it may not be typical to empathize with an AI, this instance certainly elicits a mixture of amusement and pity.