ChatGPT's Caste Bias: AI Assigns Professions Based on Indian Surnames

• GPT-4 assigned high-status professions to 'Bansal' (upper-caste) and low-status roles to 'Ahirwar' (Dalit) based solely on surnames.
• Multiple AI models consistently internalized caste hierarchies from their training data, linking certain surnames to specific professions and traits.
• Sociologists confirm AI's bias reflects societal realities, as caste identities persist even with religious conversions.
• Concerns rise that AI's embedded bias could influence critical decision-making in hiring, credit, education, and healthcare.
• Studies like DECASTE and IndiCASA highlight the need for specific frameworks to audit caste-based bias in AI, beyond Western race/gender models.