In a storyline that feels ripped from a satirical sci-fi novel, the world’s most talked-about AI, ChatGPT, has been royally humbled at a game of chess. The vanquisher? Not a grandmaster, but a gaming console that was collecting dust when the internet was still a twinkle in a computer scientist’s eye: the Atari 2600.
In a showdown for the ages, or at least for a slow Tuesday, Citrix engineer Robert Caruso pitted the cutting-edge, multi-million-dollar-trained ChatGPT 4o against an emulated version of Atari’s 1979 “Video Chess.” On the beginner difficulty, no less. What was expected to be a swift and brutal checkmate by the modern marvel of artificial intelligence turned into a comedic tragedy of errors, with the AI reportedly getting “absolutely wrecked.”
Since IBM’s Deep Blue sent Garry Kasparov packing in 1997, we’ve grown accustomed to silicon brains dominating the 64 squares. The latest chess engines boast ELO ratings that would make even the most brilliant human players weep. This, however, was not a battle of titans. It was a battle of a titan against a technological fossil, and the fossil won.
Caruso, after a chat with the AI about the history of the game, was curious to see “how quickly” the self-assured chatbot would dismantle a machine that thinks only a move or two ahead. The answer, it turned out, was not quickly at all. In fact, not ever.
The match, which dragged on for a gruelling hour and a half, saw ChatGPT flailing about like a novice who’d just learned which way the horsey moves. According to Caruso’s now-viral LinkedIn post, the AI prodigy confused rooks for bishops, blissfully ignored pawn forks, and frequently lost track of where its own pieces were. In a move that will be familiar to anyone who has ever blamed the controller for a loss in a video game, ChatGPT’s first excuse was that the Atari’s pixelated icons were simply “too abstract to recognise.”
Ever the patient human, Caruso switched to standard chess notation. It didn’t help. The blunders continued, so egregious they would have got laughed out of a 3rd-grade chess club. All the while, the humble 8-bit engine of the Atari, with its brute-force board evaluation and 1977 stubbornness, just kept chugging along, a relentless, wood-panelled titan.
Even with Caruso intervening to save it from its most catastrophically bad decisions, ChatGPT couldn’t turn the tide. In a final, desperate plea that hints at a burgeoning digital ego, it repeatedly asked if they could just “start over.” Eventually, with its king in tatters and its digital dignity in shreds, ChatGPT conceded.
Of course, the AI evangelists are quick to point out that ChatGPT isn’t designed for chess. But for a technology so often touted as the dawn of a new era of general intelligence, one that has society bracing for untold upheaval, getting trounced by a console you can still buy on eBay for less than a fancy dinner is a sobering, and frankly hilarious, reality check.
So, while we may be on the precipice of something new, it seems our future AI overlords might need to brush up on their classics. As for this blogger, I’ll sleep soundly knowing ChatGPT can’t outwit a 50-year-old cartridge. I’ll start to lose sleep when the AI can beat me at kickboxing!
Oh, and if you want to recreate the experiment, I was surprised to see that “new and sealed” Atari 2600s are on eBay.
“Atari 2600” by moparx is licensed under CC BY-SA 2.0.
