Atari humiliates ChatGPT and Copilot lose in chess to a nearly 50-year-old console

The Atari 2600 , a console launched almost half a century ago, has just defeated two of the world’s most advanced artificial intelligence systems: OpenAI’s ChatGPT and Microsoft’s Copilot . Through a series of chess games conducted by infrastructure architect Robert Caruso , this curious confrontation revealed not only the technical limits of modern AIs, but also their fragility when faced with tasks requiring spatial memory, coherent logic, and consistent strategy .

It all started when Caruso was discussing chess engines like Stockfish or AlphaZero with ChatGPT , during which the chatbot claimed that he was also a formidable player and could easily beat Atari 2600 chess . Motivated by the challenge, Caruso decided to put those statements to the test.

Using the Stella emulator and the classic game Video Chess (released in 1979), he pitted OpenAI’s language model against a console with a mere 1.19 MHz processing speed and 128 bytes of RAM . The result was disastrous for the AI. Piece confusion , invalid moves , and board tracking errors were constant throughout a 90-minute game on the lowest difficulty setting .

Despite receiving human help interpreting the environment, ChatGPT was unable to play consistently. Caruso went so far as to claim that he “made enough mistakes to be laughed at in a third-grade chess club,” and the AI eventually justified its defeat by claiming that the game icons were too abstract to properly recognize, though the problems persisted even after changing the board layout.

After ChatGPT's defeat, Caruso repeated the experiment with Copilot , Microsoft 's programming assistant based on similar technology. Copilot claimed to be able to think 10 or 15 moves ahead , and promised to remember previous moves to improve its performance. However, reality proved disappointing once again.

In just seven moves , the AI had lost several key pieces and came up with impossible strategies , such as placing its queen right in front of its opponent’s to “capture it next turn,” without considering that this would mean its own immediate elimination . In the end, Caruso asked Copilot to draw the board from memory, and noting the inconsistencies with reality, the assistant politely accepted defeat .

While it may seem anecdotal, the experiment exposes an important reality: generative AIs like ChatGPT and Copilot lack human understanding , persistent memory , and abstract thought . As The Register explains, these technologies predict words based on statistical patterns in text, but they don't "see" the board or "learn" from the game. This makes them poorly suited to tasks that require spatial coherence , structured reasoning , or complex visual recognition .

For its part, Windows Central warns that these systems may appear intelligent, but in reality they only simulate intelligence through language, without a true understanding of the physical or logical environment. In a game like chess, where every move depends on the context of the board and the precise rules, this lack becomes evident.

The irony is that the Atari 2600 doesn't "think" at all . Its chess program is based on simple algorithms, with rigidly coded rules and no learning capability. Still, that logical rigidity was enough to defeat two of today's most hyped AIs , making it clear that complexity doesn't always equal efficiency.

Robert Caruso has announced that the next objective will be to pit Atari's undefeated system against Gemini , the AI developed by Google , in a new duel that could expand the conclusions of his experiment.

ChatGPT and Copilot 's defeat by a 1970s console is a humiliating but revealing lesson about the current limits of artificial intelligence. Although these systems can write poetry, solve equations, and engage in complex conversations, their performance on tasks requiring visual interaction, spatial context, and structured logic still leaves much to be desired.

Far from being a definitive failure, this episode serves to recalibrate expectations about what these AIs can—and can't—do. In a world that increasingly relies on technology, understanding its true capabilities is just as important as continuing to push its development. And meanwhile, an Atari 2600 continues to prove that, sometimes, retro can still win .