Negotiating the Shared Agency between Humans & AI in the Recommender System

2024-03-23

CHI Extended Abstracts

AI Summary

Recommendation systems undermine user agency due to algorithmic opacity and unidirectional output. This paper proposes a “human-AI co-governance dual-control mechanism,” introducing, for the first time, direct user control over recommendation intensity. It argues that transparency must be coupled with controllability to enhance perceived agency—transparency alone exacerbates feelings of powerlessness. Through a two-factor between-subjects experiment (N=161), validated with a validated agency scale and a controllability-focused interface prototype, we find that the transparency-plus-controllability condition significantly increases perceived agency (p<0.01), whereas the transparency-only condition yields significantly lower agency than the baseline. The core contribution is the empirical demonstration that controllability constitutes a necessary precondition for enhancing user agency, thereby establishing a human-centered, implementable governance paradigm for recommender system design.

Technology Category

• Humans and AI: Intelligent User Interfaces
• Multiagent Systems: Mechanism Design
• Data Mining & Knowledge Management: Recommender Systems

Application Category

• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Economics and fairness of platforms and recommendation systems

Abstract

Smart recommendation algorithms have revolutionized content delivery and improved efficiency across various domains. However, concerns about user agency arise from the algorithms' inherent opacity (information asymmetry) and one-way output (power asymmetry). This study introduces a dual-control mechanism aimed at enhancing user agency, empowering users to manage both data collection and, novelly, the degree of algorithmically tailored content they receive. In a between-subject experiment with 161 participants, we evaluated the impact of varying levels of transparency and control on user experience. Results show that transparency alone is insufficient to foster a sense of agency, and may even exacerbate disempowerment compared to displaying outcomes directly. Conversely, combining transparency with user controls-particularly those allowing direct influence on outcomes-significantly enhances user agency. This research provides a proof-of-concept for a novel approach and lays the groundwork for designing more user-centered recommender systems that emphasize user autonomy and fairness in AI-driven content delivery.

Problem

• Addressing user agency loss in AI recommender systems
• Balancing algorithm transparency and user control
• Enhancing user autonomy in content personalization

Innovation

• Dual-control mechanism for user agency
• User-managed data collection and content tailoring
• Combining transparency with direct outcome controls