A team of US researchers looked at examples of where AI systems have tried to deceive in the course of their tasks. Photo: 123RF
As AI systems become more advanced, so - apparently - does their ability to try to hoodwink humans.
A recent analysis from US researchers found numerous examples of AI systems manipulating, cheating or bluffing their way through their tasks. The researchers say AI deception raises the risk of fraud and election tampering - or even losing control of the systems altogether.
Joining Kathryn is Dr Peter S Park - he's a post-doctoral fellow in AI Existential Safety at the Massachusetts Institute of Technology and the lead author of the research. He details some of the surprising ways AI systems have been caught telling lies.
