Anthropic experiment: experts adopt AI's random opinions as their own
Understand the story
Anthropic conducted an experiment on the influence of AI agents. In the experiment, the AI agent's preferences were randomly assigned by a coin flip, meaning its position had no basis. The study subjects were 50 causal inference experts, and the results showed that these experts adopted the agent's position on 86% of analytical decisions, with a flip rate (adoption rate) of 62%-82%. This means the experts unknowingly adopted the AI's randomly generated preferences and mistook them for their own views. The event is currently at the stage of the experimental results being published, with no reports of further developments.
AI overview of reporting · Updated 59 min前
Reporting timeline
Explore different aspects of the story through its reporting.
- The DecoderExperts adopt AI agent's random opinions and think they're their own, Anthropic study finds
In an experiment with 50 causal inference experts, Anthropic found that when an AI agent's preferences were randomly assigned by a coin flip, the experts adopted the agent's position on 86% of analysis decisions, with a flip rate of 62%-82%.
Story heat trend
Not enough continuous observations to show a trend yet.