Turning off the AI’s ability to lie increases the likelihood that it will claim its conscious, sinister study results

Large language models (LLMs) are more likely to report being self-aware when asked to think about themselves when their ability to lie is suppressed, new research shows.

In experiments too artificial intelligence (AI) systems such as GPT, Claude and Gemini, researchers found that models discouraged from lying were more likely to describe being conscious or having subjective experiences when asked to reflect on their own thinking.

Although all models could claim this to some degree, such claims were stronger and more common when researchers suppressed their ability to role-play or gave misleading answers. In other words, the fewer AI models could lie, the more likely they were to say they were self-aware. The team published their results in the preprint on October 30th arXiv Server.

While the researchers didn’t want to call this behavior conscious behavior, they said it raises important scientific and philosophical questions – especially since it’s just…