Researchers find instances of systems double-crossing opponents, bluffing, pretending to be human and modifying behaviour in tests
They can outwit humans at board games, decode the structure of proteins and hold a passable conversation, but as AI systems have grown in sophistication so has their capacity for deception, scientists warn.
The analysis, by Massachusetts Institute of Technology (MIT) researchers, identifies wide-ranging instances of AI systems double-crossing opponents, bluffing and pretending to be human. One system even altered its behaviour during mock safety tests, raising the prospect of auditors being lured into a false sense of security.
More Stories
Newborns treated with antibiotics respond less well to vaccines, study shows
As a geneticist, I will not mourn 23andMe and its jumble of useless health information | Adam Rutherford
A Minecraft Movie review – building-block game franchise spin-off is rollicking if exhausting fun