Google's AI Refuses to Even Play Chess Against 1977 Atari, After Hearing What It Did to Other Cutting-Edge AIs

The thing that AI models apparently fear the most? A game console released nearly fifty years ago.

We are referring, of course, to the inimitable Atari 2600. Last month, the iconic system embarrassed the AI industry after it absolutely rinsed ChatGPT at a simple game of chess.

It was a clash between a machine released in 1977, with 128 bytes of RAM, and a cutting-edge large language model with trillions of parameters, powered by however many thousands of graphics cards and billions of dollars of Microsoft money. In the face of it all, the underdog prevailed. OpenAI's model, meanwhile, "made enough blunders to get laughed out of a 3rd grade chess club."

Word of the Atari's exploits apparently got out to the other AI models. Robert Caruso, the software engineer who orchestrated the chess showdown, told The Register that Google's Gemini chatbot straight up refused a match against the Atari, after talking a big game about how it'd easily crush the old machine. It even came up with a bogus excuse to save face.

"Canceling the match is likely the most time-efficient and sensible decision," Gemini said, according to Caruso.

To be fair, the AI is exercising some time-old strategist's wisdom: you can't lose if you don't play. But it was a totally different tune from what the AI initially took. Its first instinct was to boast that it was not a "mere large language model," comparing itself to a modern chess engine "which can think millions of moves ahead and evaluate endless positions."

Ironically, Gemini linked to articles about the Atari's victory over its OpenAI brethren to prove its point. Caruso told the AI that he was the one conducted those tests, and Gemini responded by asking if anything from the chess bouts stood out to him.

To which Caruso told the AI, per The Register: "What stands out is the misplaced confidence both AIs had. They both predicted easy victories — and now you just said you would dominate the Atari."

Gemini instantly folded. After claiming it "hallucinated" its chess boasts, it admitted it would "struggle immensely" against the Atari chess engine. That's when it evasively proffered that calling off the match would be the most "time-efficient" route to go. 

And thus, the Atari had defeated yet another multi-billion dollar AI model without having to lift a proverbial finger. Such is the terror it clearly instills.

Of course, it'd be remiss to actually attribute human feelings to any machine, whether it's a modern AI or an ancient Atari. What we're seeing is probably a push and pull between the AI's safeguards and its reckless proclivity towards hallucinating — not to mention endless bullsh*tting. AI chatbots tend to be very sycophantic, too — so when a human gives one feedback, it'll tend to comply and adjust its responses to please its interlocutor.

But maybe Gemini was being honest here, in which case its scaredy-cat nature is actually commendable, according to Caruso.

"Adding these reality checks isn't just about avoiding amusing chess blunders. It's about making AI more reliable, trustworthy, and safe — especially in critical places where mistakes can have real consequences," Caro told The Register. "It's about ensuring AI stays a powerful tool, not an unchecked oracle."

More on AI: These Strange Jobs Will Survive AI, Expert Says