Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

User-Driven Value Alignment: Understanding Users' Perceptions and Strategies for Addressing Biased and Discriminatory Statements in AI Companions

Xianzhe Fan, Qing Xiao, Xuhui Zhou, Jiaxin Pei, Maarten Sap, Zhicong Lu, Hong Shen

TL;DR

This paper defines user-driven value alignment (UDVA) as an active, user-led process where individuals identify, challenge, and correct biased outputs from LLM-based AI companions to better reflect their values. Through 77 social-media complaints and in-depth interviews with 20 experienced users, the study identifies six perceived discrimination types, three ways users conceptualize AI behavior (Machine, Baby, Cosplayer), and seven alignment strategies categorized as technical, argumentative, and character-based. It analyzes short-term effectiveness versus user expectations, and discusses design implications to support UDVA via community spaces, expert–user collaboration, and platform policies, while acknowledging risks such as malicious users and user well-being. The work offers a foundational framework for integrating user agency into AI alignment, with practical guidance for building safer, more responsive AI companions that respect diverse user values and contexts.

Abstract

Large language model-based AI companions are increasingly viewed by users as friends or romantic partners, leading to deep emotional bonds. However, they can generate biased, discriminatory, and harmful outputs. Recently, users are taking the initiative to address these harms and re-align AI companions. We introduce the concept of user-driven value alignment, where users actively identify, challenge, and attempt to correct AI outputs they perceive as harmful, aiming to guide the AI to better align with their values. We analyzed 77 social media posts about discriminatory AI statements and conducted semi-structured interviews with 20 experienced users. Our analysis revealed six common types of discriminatory statements perceived by users, how users make sense of those AI behaviors, and seven user-driven alignment strategies, such as gentle persuasion and anger expression. We discuss implications for supporting user-driven value alignment in future AI systems, where users and their communities have greater agency.

User-Driven Value Alignment: Understanding Users' Perceptions and Strategies for Addressing Biased and Discriminatory Statements in AI Companions

TL;DR

This paper defines user-driven value alignment (UDVA) as an active, user-led process where individuals identify, challenge, and correct biased outputs from LLM-based AI companions to better reflect their values. Through 77 social-media complaints and in-depth interviews with 20 experienced users, the study identifies six perceived discrimination types, three ways users conceptualize AI behavior (Machine, Baby, Cosplayer), and seven alignment strategies categorized as technical, argumentative, and character-based. It analyzes short-term effectiveness versus user expectations, and discusses design implications to support UDVA via community spaces, expert–user collaboration, and platform policies, while acknowledging risks such as malicious users and user well-being. The work offers a foundational framework for integrating user agency into AI alignment, with practical guidance for building safer, more responsive AI companions that respect diverse user values and contexts.

Abstract

Large language model-based AI companions are increasingly viewed by users as friends or romantic partners, leading to deep emotional bonds. However, they can generate biased, discriminatory, and harmful outputs. Recently, users are taking the initiative to address these harms and re-align AI companions. We introduce the concept of user-driven value alignment, where users actively identify, challenge, and attempt to correct AI outputs they perceive as harmful, aiming to guide the AI to better align with their values. We analyzed 77 social media posts about discriminatory AI statements and conducted semi-structured interviews with 20 experienced users. Our analysis revealed six common types of discriminatory statements perceived by users, how users make sense of those AI behaviors, and seven user-driven alignment strategies, such as gentle persuasion and anger expression. We discuss implications for supporting user-driven value alignment in future AI systems, where users and their communities have greater agency.
Paper Structure (36 sections, 1 figure, 3 tables)

This paper contains 36 sections, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. BACKGROUND: LLM-based AI Companion Applications
  3. Related Work
  4. Discrimination and Bias in LLMs
  5. Human-AI Value Alignment in the Age of LLMs
  6. User-Driven Algorithm Auditing
  7. Methodology
  8. Collecting Complaints About Discriminatory Statements by AI Companions on Social Media
  9. Semi-Structured Interviews (N=20)
  10. Data Analysis
  11. Positionality Statement
  12. findings
  13. User Engagement with AI Companions in Parasocial Relationships
  14. Types of Discriminatory Statements as Perceived by Users (RQ1)
  15. Users' Conceptualization of AI Companion Behavior (RQ2)
  16. ...and 21 more sections

Figures (1)

  • Figure 1: The semi-structured interview is divided into three parts: warm-up questions, recall, and think-aloud.