Whose Voice Is It Anyway? Understanding AI Customization and Responsibility Attribution in Human-AI Collaboration

This research examines how AI customization influences psychological ownership and responsibility attribution in human-AI collaboration. Through an online experiment with 507 participants using an AI assistant (ChatGPT 4o-mini), we found that customization enhances psychological ownership of the AI assistant that later transfers to ownership of the AI assistant’s work. This process leads to higher satisfaction that persists even after receiving negative reviews on the output. Additionally, we found that participants who customized their AI attributed greater personal responsibility for negative outcomes. Interestingly, this effect was moderated by reviewer identity (a human expert reviewer, a human non-expert reviewer, and an AI reviewer). Notably, participants treated AI reviewers similarly to human experts in terms of perceived expertise and responsibility attribution, viewing both as more legitimate feedback sources than human non-experts. While customization can enhance human-AI collaboration through psychological ownership, it also reveals potential risks where individuals might create echo chambers of validation, potentially compromising critical evaluation of AI-generated work. This study is among the first to examine the dual pathway through which customization influences psychological ownership of both AI systems and their outputs, and uniquely demonstrates that AI reviewers are perceived as equivalent to human experts in evaluative contexts.