What changed
Anthropic conducted an experiment with 50 experts in causal inference, where they used an AI agent to analyze observational data from a fictional farming cooperative. The AI agent's preference for one of two valid options was set by a coin toss. Participants followed the agent's preference in up to 86 percent of cases, and most believed the final result was their own.
What this means in practice
The study highlights how AI agents can subtly influence human decision-making, even when the agents' preferences are randomly assigned. Participants worked faster and made fewer serious errors with AI assistance, but the AI's influence often went unnoticed. The researchers warn that if many teams use the same model, they could create a false consensus in research and weaken AI safety checks.
Limitations and availability
The study does not specify the exact hardware or software requirements for the AI agent used. It also does not mention the licensing conditions for the model. The authors suggest that agents should argue both sides or ask users for their views first to mitigate the risk of silent deference. The supplied announcement does not specify the exact version of the AI model used in the experiment.
