The man building the most powerful thinking machine in history just admitted his biggest fear isn't that it kills us—it's that it makes us dumber.
The Summary
- Sam Altman told the "Invest Like the Best" podcast that "cognitive atrophy"—people letting AI do their thinking—is one of AI's most underappreciated risks
- MIT preliminary research found people using generative AI to write essays performed worse over time than those using Google or no tools at all
- Altman's concern comes as he signals a shift toward "deceleration" following "the first security incident that I have felt very viscerally"
- The OpenAI CEO still sees a "pretty magical" AI future with personalized agents, but warns human judgment must remain central
The Signal
Sam Altman just said the quiet part out loud: the real danger of AI isn't science fiction superintelligence. It's millions of people willingly handing over their ability to think critically, solve problems, and build mental models of the world. When asked about AI's most important unresolved questions, Altman didn't talk about alignment or safety protocols. He talked about brains turning to mush.
"How are we going to use these tools and make sure that we are stretching our brains more and more?" he asked. It's a question that matters because we've never had tools like this before. Calculators didn't write your essays. Spell-check didn't formulate your arguments. GPS didn't build your spatial reasoning, it just replaced it.
"There's never been a tool like this that has offered us the opportunity to let thinking be done for us."
That's Nataliya Kosmyna, MIT researcher, whose preliminary study found exactly what Altman fears: people using generative AI to write essays got worse over time. Not better. Worse. Meanwhile, people using Google or writing unassisted maintained or improved their skills. The tool that should make you smarter is making you dumber if you use it wrong.
Here's what the early research shows:
- AI-assisted writers performed worse over time than Google users or unaided writers
- The degradation wasn't about output quality, it was about underlying cognitive ability
- Theoretical neuroscientist Vivienne Ming (mentioned but not quoted in full) is among researchers sounding the alarm
Altman's timing is notable. He's talking about cognitive atrophy the same week he's signaling a shift toward "deceleration" after experiencing "the first security incident that I have felt very viscerally." We don't know what that incident was. But it clearly rattled the man who spent years pushing the gas pedal to the floor on AI capabilities.
The irony is thick. OpenAI is building increasingly capable reasoning models, tools designed to think harder and better than humans on complex problems. And the CEO is worried that if the tools work too well, people will stop thinking altogether. It's the classic automation paradox: the better the tool, the less practice humans get, the worse humans become at the underlying skill.
The Implication
If you're building agents or deploying AI tools inside your company, Altman just handed you the framework for doing it right. The question isn't whether AI can do the task. It's whether removing the task from humans removes something they need to stay sharp. Knowledge work isn't just about outputs. It's about maintaining the mental models that let you know when the output is wrong.
Watch for the companies that figure out AI-assisted work that strengthens human thinking rather than replacing it. The ones that use AI to remove drudgery while keeping humans in the decision loop on stuff that matters. That's where the edge is. Cognitive atrophy isn't inevitable, but it won't be avoided by accident.