The Media Still Can’t Get AI Right, and It’s Getting Worse

Remember when Facebook had to “pull the plug” on its AI because the bots started chatting in a secret language? That story went viral a few years back, complete with Terminator comparisons and breathless headlines about machines slipping our control. Except none of it was true.

Back in 2017, Facebook’s AI research unit published a paper about negotiation bots. The interesting part was how the bots learned to barter over objects. But there was a quirk: because the researchers hadn’t constrained the bots to proper English grammar, they sometimes spat out nonsense like “balls have zero to me to me to me to me.” That’s not a new language. That’s a bug. The researchers noted it, fixed it, and moved on.

Fast Company ran with “AI Is Inventing Language Humans Can’t Understand. Should We Stop It?” Then the sharks circled. “Facebook engineers panic, pull plug on AI after bots develop their own language,” one outlet screamed. The Sun compared it to The Terminator. Carnegie Mellon professor Zachary Lipton watched the whole thing and called it what it was: “sensationalized crap.”

This isn’t a new problem. In 1946, when the ENIAC computer was unveiled, journalists called it an “electronic brain,” a “mathematical Frankenstein,” and a “wizard.” Physicist DR Hartree tried to correct the record with a sober technical article in Nature. The London Times still ran with “An Electronic Brain: Solving Abstruse Problems; Valves with a Memory.” Hartree wrote a letter to the editor saying the machine was “no substitute for human thought.” Too late. The “brain machine” nickname stuck.

Same story in 1958 when Frank Rosenblatt showed off the perceptron, a primitive pattern-recognition algorithm. The New York Times claimed it could “teach itself” and would soon “walk, talk, see, write, reproduce itself and be conscious of its own existence.” The perceptron could barely recognize a few shapes.

Lipton calls this the “AI misinformation epidemic.” Social media has made it worse. Self-proclaimed “AI influencers”—people who do nothing but paraphrase Elon Musk’s worst takes—churn out low-quality pieces for clicks. The result is a public that either thinks we’re five minutes from SkyNet or that AI is a total scam. Neither helps.

The real danger isn’t that the bots will rise up. It’s that the hype creates unrealistic expectations, which then crash into reality, causing funding freezes and public distrust. We saw this happen in the 1970s with the first “AI winter.” The same pattern is repeating now, just faster and louder.

I’ve been watching this cycle for years. Every time a paper comes out with a mildly interesting result, I brace for the hot takes. A model generates a weird image? “AI is hallucinating.” A chatbot says something dumb? “AI is sentient.” No. It’s just math doing unexpected things because the training data had a bug or the constraints were loose.

What frustrates me most is that good, careful journalism about AI exists. But it doesn’t go viral. The measured take doesn’t get the retweets. So we get the Terminator story instead, and the researchers who actually do the work have to spend half their time cleaning up messes they never made.

If you’re reading this and you write about AI, please stop. Stop with the Frankenstein metaphors. Stop pretending every bug is a sign of consciousness. And for the love of God, read the actual paper before you publish. The rest of us are tired of cleaning up your mess.