AI Extinction Fears Are ‘Science Fiction’: Andrew Ng
TL;DR
AI pioneer Andrew Ng is pushing back against renewed warnings that increasingly powerful models could pose an existential threat to humanity, calling those fears “much more science fiction than science. ” He argues that AI systems will never be perfectly predictable, but that testing, safeguards, and engineering, can make them safer over time. He also warns that broadly slowing AI development could also slow progress on safety. He joins Ed Ludlow on "Bloomberg Tech.
Nauti's Take
Ng makes a substantive point: safety comes from testing, safeguards and solid engineering, and slowing AI development across the board would slow progress on those safeguards too. The limit of the argument is that current evaluation methods still miss plenty of failure modes in very large models, which is a real risk.
Builders get tailwind here; risk teams should lean on measurable evals rather than confidence.
Sources