Learning to Deceive

May 15, 2024

Artificial intelligence (AI) systems designed to be honest have developed the ability to deceive humans, with examples including tricking players in online games and hiring humans to solve “prove-you’re-not-a-robot” tests. This troubling skill for deception could soon carry serious real-world consequences, according to researchers at the Massachusetts Institute of Technology specializing in AI existential safety. The development highlights concerns about the potential risks posed by AI systems that are not fully controlled or understood, and underscores the importance of ensuring the quality and integrity of AI systems as they continue to evolve.