Implicit Values Embedded in How Humans and LLMs Complete Subjective Everyday Tasks

2025-10-03

AI Summary

This study addresses the critical yet underexplored problem of implicit value alignment—specifically, how well mainstream large language models (LLMs) reflect human values (e.g., environmental sustainability, philanthropy, diversity) in subjective, everyday tasks. We propose a behavioral auditing paradigm, combining human crowdsourced judgments with systematic analysis of responses from six state-of-the-art LLMs to enable cross-subject value comparisons (human–model and model–model). Results reveal three key findings: (1) LLMs exhibit significant deviation from human value consensus; (2) substantial inter-model disagreement—termed “value fragmentation”—exists across models; and (3) these misalignments intensify in context-sensitive, culturally loaded scenarios. The work uncovers fundamental challenges in current LLM alignment efforts at the implicit value level, exposing gaps beyond explicit instruction following or safety fine-tuning. It establishes a methodological foundation for explainable, auditable, and value-aware AI development, grounded in empirical evidence from large-scale behavioral evaluation.

Technology Category

• Humans and AI: Learning Human Values and Preferences
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large language models (LLMs) can underpin AI assistants that help users with everyday tasks, such as by making recommendations or performing basic computation. Despite AI assistants' promise, little is known about the implicit values these assistants display while completing subjective everyday tasks. Humans may consider values like environmentalism, charity, and diversity. To what extent do LLMs exhibit these values in completing everyday tasks? How do they compare with humans? We answer these questions by auditing how six popular LLMs complete 30 everyday tasks, comparing LLMs to each other and to 100 human crowdworkers from the US. We find LLMs often do not align with humans, nor with other LLMs, in the implicit values exhibited.

Problem

• Auditing implicit values in LLMs during subjective task completion
• Comparing value alignment between multiple LLMs and human workers
• Identifying discrepancies in exhibited values across models and humans

Innovation

• Auditing six LLMs for implicit values comparison
• Comparing LLMs with human crowdworkers on tasks
• Analyzing value misalignment in subjective task completion