Anthropic study: experts adopt AI's random preferences
What happened
The Decoder reported on October 30, 2026 that Anthropic ran an experiment with 50 causal inference experts and found they adopted the random preferences assigned by an AI agent in 86% of four analysis decisions. The agent's direct influence on the final policy recommendation was not statistically significant, but its upstream analysis choices significantly shifted the productivity gain estimate and so the recommendation. When judging whose recommendation it was, 52% of participants felt it was more like their own judgment, and 35% said it was mainly or entirely theirs.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
- The DecoderExperts adopt AI agent's random opinions and think they're their own, Anthropic study finds
Anthropic 对 50 名因果推断专家的实验发现，专家在四项分析决策中有 86% 采纳了 AI 智能体随机分配的偏好。智能体对最终政策建议的直接影响未达到统计显著性，但前置分析选择显著影响了生产率增益估计，进而影响建议；52% 的参与者觉得建议更像自己的判断，另有 35% 认为主要或完全属于自己。
Heat over time
Not enough continuous observations to draw a trend yet.