LLMs show ideological bias in economic reasoning, favoring interventionist views

A new study evaluated 20 large language models (LLMs) on their ability to reason about economic causal effects, finding that they exhibit systematic ideological bias. The research extended the EconCausal benchmark with contested cases where pro-government and pro-market perspectives predict different outcomes. LLMs were less accurate on these contested items, and their errors disproportionately favored pro-government interpretations, a skew not corrected by in-context prompting. AI

IMPACT Highlights systematic ideological bias in LLMs' economic reasoning, necessitating direction-aware evaluations for policy analysis.

RANK_REASON Academic paper evaluating LLM performance on a specific benchmark.

Read on arXiv cs.CL →

• arXiv
• EconCausal
• LLMs

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →