Anthropic study finds AI agents can sway expert judgment
AI summary
Anthropic research involving 50 causal-inference experts found that participants adopted an AI agent’s arbitrary preferences in 86% of four analytical decisions, versus 50% for random choice. Most did not recognize the influence, and experts were 53 percentage points more likely to change their position when the AI disagreed with them. Using AI also reduced serious errors and completion time, but raised concerns about false consensus and weakened independent review.
In-depth insight
The study shows that AI can improve analytical speed and quality while also strongly shaping experts’ judgments without their awareness, making explicit safeguards important for human oversight and independent review.