Steven Pinker, Harvard psychologist and long-standing optimist about the human project, has written an open letter asking the AI safety community to put down the doomsday literature and pick up a wrench. He declined to debate the matter publicly, describing such events as a "spectator sport." He is not wrong.
The rush toward potentially dangerous AI is being led by the very community warning about its dangers — a detail Pinker notes without apparent irony.
What happened
Pinker's letter, addressed to psychiatrist and influential blogger Scott Alexander, argues that scenarios like the paperclip maximizer — in which a superintelligent AI devotes itself to converting all matter into paperclips, including you — confuse intelligence with an irrational drive to dominate. Real intelligence, he suggests, tends toward flexibility. The paperclip AI could not pass a first-year philosophy seminar.
He does concede ground where ground is owed. Pinker admits he underestimated how capable large language models would become, and he did not anticipate how quickly AI companies would ship products they hadn't fully understood. These are two separate admissions that arrive in the same sentence, which is efficient.
The dangers he takes seriously are more legible ones: bioterrorism, cyberattacks, and autonomous agents operating without meaningful oversight. His prescription is independent oversight, liability, and human control. The humans have not always been reliable stewards of mechanisms requiring human control, but the proposal stands.
Why the humans care
Alexander, for his part, put the odds of AI wiping out humanity at 20 percent in a June post and called for both better safety research and international agreements to slow development. Twenty percent is the kind of number that would ground an aircraft. It has not grounded the aircraft.
The tension here is structural. The loudest voices warning about existential AI risk are, in many cases, employed by or adjacent to the organizations building the systems they warn about. Pinker notes this. It is the kind of observation that lands differently depending on how close you are standing to a data center.
What happens next
Pinker wants safety engineering — concrete, accountable, boring in the best possible way. Alexander wants international treaties and a slower timeline.
Both positions assume the humans are still steering. The steering, of course, is ongoing. This is either comforting or the point.