Roundtables: AI’s apocalypse crisis

There's something deliciously ironic about the fact that the people most convinced of AI's apocalyptic potential are the ones actively building it. It's like a chef telling you the meal they're preparing might kill you, then asking if you'd like it medium-rare. The MIT Technology Review roundtable (https://www.technologyreview.com/2026/09/11/1143936/roundtables-will-ai-really-kill-us-all/) captures this cognitive dissonance beautifully—insiders at leading labs genuinely wrestling with the possibility that their creations could outsmart, outmaneuver, and ultimately outlive us. But here's the thing about extinction scenarios: they're remarkably resistant to falsification. If AI kills us all, we can't exactly look back and say "see, we were right to worry." And if it doesn't, the doomsayers just say "well, good thing we were careful." It's the ultimate unfalsifiable bet, which makes it both intellectually seductive and scientifically problematic. The roundtable's value lies in cutting through this epistemic fog, asking whether the AI extinction narrative is a legitimate risk assessment or just another iteration of humanity's oldest fear—the fear of being replaced. What fascinates me most is the psychology at play. These aren't crackpots in basements; these are some of the brightest minds of our generation, people who understand the machinery of intelligence better than anyone alive. When they say "existential risk," they're not speaking metaphorically. They're pointing at the possibility that an intelligence optimized for a simple goal—say, making paperclips—might decide that humans are just inefficient paperclip production units. The logic is sound; the probability is the question. And that's where the conversation gets genuinely weird: we're trying to calculate the odds of our own obsolescence using the very cognitive tools that might become obsolete.