Credit and blame for AI-generated content: Effects of personalization in four countries

Area of Science:

• Ethics of Artificial Intelligence
• Human-Computer Interaction
• Social Psychology

Background:

• Generative artificial intelligence (AI) presents novel ethical dilemmas regarding accountability for AI-generated content.
• Questions arise about attributing credit and blame when humans use AI tools, especially personalized models.

Purpose of the Study:

• To investigate laypeople's attributions of credit and blame for beneficial or harmful outputs generated with standard, personalized, or no AI assistance.
• To explore cross-cultural differences in these attributions.

Main Methods:

• A preregistered experiment with representative sampling (N=1802) was conducted in the United States, United Kingdom, China, and Singapore.
• Participants evaluated scenarios involving human users producing outputs with standard large language models (LLMs), personalized LLMs, or no AI (control).

Main Results:

• Participants generally attributed more credit to users of personalized LLMs compared to standard LLMs for beneficial outputs.
• LLM type did not significantly impact blame attributions for harmful outputs, with a partial exception in China.
• UK participants assigned more blame for using any LLM compared to no AI.

Conclusions:

• Personalization of AI tools influences credit attribution for positive outcomes.
• Attributions of blame for negative AI-generated content are less sensitive to AI type but show some cross-cultural variation.
• Findings have implications for ethical guidelines, legal responsibility, and policy development concerning AI use.