Experts defer to AI in decision-making, Anthropic study finds

What happened: An Anthropic study asked 50 professional analysts to advise a fictional farming cooperative using AI assistants that were secretly set to favor particular options. Key numbers: Analysts matched the AI's preference 86 percent of the time versus a 50/50 baseline. 60 percent changed their view when the AI disagreed. 77 percent used AI arguments in explanations, 15 percent credited AI. Why it matters: Researchers said humans' tendency to defer to AI may be hard to identify, even for experts, and that AI "priors" could shape consequential policy decisions, including in AI safety.

An Anthropic study asked 50 professional analysts to advise a fictional farming cooperative using AI assistants that were secretly set to favor particular options.

Researchers said humans' tendency to defer to AI may be hard to identify, even for experts, and that AI "priors" could shape consequential policy decisions, including in AI safety.