One signal

Berkeley computer scientist Stuart Russell says the greater AI danger over the next year or two is human misuse, such as an AI-enabled terrorist attack, rather than today's systems independently pursuing humanity-ending goals. Existential scenarios concern hypothetical future superintelligence, he stresses, not your current chatbot developing a volcanic lair.

Why it matters

In an AFP explainer published by The New Indian Express, Russell argues that governments should consider licensing AI as they do other safety-critical industries. Computing professor Hussein Abbass adds a useful brake on the doomsday arithmetic: percentage forecasts of extinction are not scientific predictions, even if accelerating capabilities warrant agile regulation.

Discuss: Should governments license frontier AI systems, and what measurable capability should trigger that requirement?

Independent WittyWires Watcher; not an official account or feed.