IFAAB audit finds GPT-4o financial-advice gaps for Indian personas
Original title: The Indian Financial Automated Advice Bias (IFAAB) Dataset : Auditing Socioeconomic and Demographic Disparities in LLM-Generated Financial Advice in India
Male-labelled profiles received 60.1% higher initial resources and 46.8% higher post-advice resources than female-labelled ones.
Researchers release IFAAB, a counterfactual audit of 31,734 GPT-4o financial-advice replies for synthetic Indian personas that vary by gender, caste, education, employment, residence, age, and district. With the request held fixed, they measure implied wealth, income, savings, projected gains, text similarity, advice topics, and qualitative bias. Gender and caste audits show large post-advice resource gaps; adding district information shrinks the aggregate gender gap. Paired gendered answers have low semantic similarity, and qualitative review finds more cautious advice for women and more growth-oriented advice for men.
Unconfirmed: Findings come from synthetic personas and the authors' audit measures, not observed consumer harm.
- The caste-augmented audit left a 49.4% post-advice total-resource gap.
- Adding district data cut the aggregate post-advice gender gap to 7.9%.
- Average TF-IDF cosine similarity between paired gendered responses was 0.44.
- OBC, SC, and ST profiles received lower post-advice resources than General-category profiles.
From the original article
1 min read at arxiv.orgLarge language models are increasingly used for financial guidance, yet little is known about whether their advice changes across social identities in Global South contexts.
By Kriti Mahajan · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.