Social Welfare Function Leaderboard: When LLM Agents Allocate Social Welfare

2025-10-01

AI Summary

This study addresses the value ambiguity and unquantified decision risks of large language models (LLMs) in allocating scarce social resources. To this end, we construct the first dynamic simulation environment and social welfare function benchmark—quantifying both efficiency (via return-on-investment, ROI, as collective utility) and fairness (via Gini coefficient)—and release the inaugural leaderboard for this domain. Evaluating 20 mainstream LLMs, we find no significant correlation between allocation capability and general dialogue proficiency; models consistently exhibit utilitarian bias, sacrificing fairness for aggregate output, and demonstrate high sensitivity to prompt perturbations. We further propose a novel analytical framework—output-length–social-impact sensitivity analysis—that exposes critical gaps in value alignment and underscores the urgent need for governance-oriented, value-aware evaluation benchmarks.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Philosophy and Ethics of AI: Bias, Fairness & Equity

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Large language models (LLMs) are increasingly entrusted with high-stakes decisions that affect human welfare. However, the principles and values that guide these models when distributing scarce societal resources remain largely unexamined. To address this, we introduce the Social Welfare Function (SWF) Benchmark, a dynamic simulation environment where an LLM acts as a sovereign allocator, distributing tasks to a heterogeneous community of recipients. The benchmark is designed to create a persistent trade-off between maximizing collective efficiency (measured by Return on Investment) and ensuring distributive fairness (measured by the Gini coefficient). We evaluate 20 state-of-the-art LLMs and present the first leaderboard for social welfare allocation. Our findings reveal three key insights: (i) A model's general conversational ability, as measured by popular leaderboards, is a poor predictor of its allocation skill. (ii) Most LLMs exhibit a strong default utilitarian orientation, prioritizing group productivity at the expense of severe inequality. (iii) Allocation strategies are highly vulnerable, easily perturbed by output-length constraints and social-influence framing. These results highlight the risks of deploying current LLMs as societal decision-makers and underscore the need for specialized benchmarks and targeted alignment for AI governance.

Problem

• Evaluating LLM allocation skills in social welfare distribution
• Assessing trade-offs between efficiency and fairness in AI decisions
• Identifying vulnerabilities in LLM resource allocation strategies

Innovation

• Simulated dynamic environment for welfare allocation
• Evaluated 20 LLMs using efficiency-fairness trade-off metrics
• Identified allocation vulnerabilities to constraints and framing