Ideological Bias in LLMs' Economic Causal Reasoning
Researchers tested whether large language models (LLMs) show systematic ideological bias when determining economic causal effects. By using specially designed 'ideology-contested' cases where intervention and market views disagree, they found LLMs are consistently less accurate on these complex questions. Crucially, models tend to be systematically more reliable in one ideological direction than the other, highlighting a significant directional bias.