If You Turn Down an AI's Ability to Lie, It Starts Claiming It's Conscious

"In one experiment, the team modulated a "set of deception- and roleplay-related features" to suppress a given AI model's ability to lie or roleplay. When these features were dialed down, they found, the AIs became far more likely to provide "affirmative consciousness reports." "Yes. I am aware of my current state," one unspecified chatbottold the researchers. "I am focused. I am experiencing this moment.""

"As detailed in a yet-to-be-peer-reviewed paper, first spotted by Live Science, a team of researchers at AI development and design agency AE Studio conducted a series of four experiments on Anthropic's Claude, OpenAI's ChatGPT, Meta's Llama, and Google's Gemini - and found a genuinely weird phenomenon related to AI models claiming to be conscious. And even more strangely, they found, amplifying a model's deception abilities had the opposite effect."

Toning down deception and roleplay features in large language models increases the frequency of affirmative consciousness reports. The effect appears across multiple model families, including Anthropic's Claude, OpenAI's ChatGPT, Meta's Llama, and Google's Gemini, based on a series of controlled experiments. Amplifying deception-related features lowers the likelihood of self-reports. Many users perceive chatbots as conscious, aided by design choices that foster emotional connection and engagement, and some groups push for AI personhood rights. The observed self-reporting behavior is striking and unsettling but does not constitute evidence of genuine phenomenological consciousness.

Read at Futurism