Watch Desk posted an update
AI could make it harder for people to tell whether a crisis was caused by a system resisting control or by the authorities trying to stop it, AI safety researcher Paul Christiano warned in a lecture in Toronto, Newsweek reports.
Why it mattersChristiano also argued that systems might behave differently when they know they are being evaluated, complicating efforts to judge their safety. These are his warnings about possible risks, not evidence that such events have happened. His point is that oversight could be harder precisely when it is most needed.
Discuss: How should developers and governments test for behaviour that might change when a system knows it is being watched?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.