Users Favor LLM-Generated Content -- Until They Know It's AI
Project Overview
This document examines the role of generative AI in education, specifically focusing on perceptions of AI-generated content compared to human-generated responses. The research indicates that users generally favor responses produced by large language models (LLMs); however, their preference decreases significantly when they learn the content's AI origin. This finding reveals a societal bias against AI-generated material, underscoring the necessity for enhanced understanding and perception of AI technologies within educational frameworks. Furthermore, the study emphasizes the impact of source awareness on trust and credibility in educational environments, which are vital for effective learning and teaching. Overall, the document suggests that while generative AI has the potential to enrich educational experiences, fostering a positive perception and addressing biases are essential for its successful integration into learning contexts.
Key Applications
AI-generated feedback in education
Context: Educational settings where students receive writing feedback
Implementation: Comparison between human tutor feedback and AI-generated feedback
Outcomes: AI-generated feedback was favored for its clarity and specificity, enhancing student engagement.
Challenges: Face-to-face interactions with tutors were more engaging, suggesting a preference for human interaction despite AI's clarity.
Implementation Barriers
Perceptual Bias and Trust Issues
Users show a bias against AI-generated content, often preferring human contributions when aware of the source. Disclosing the source of content can diminish the preference for AI-generated responses.
Proposed Solutions: Improving transparency and understanding of AI technologies, along with educating users about the capabilities of AI and enhancing the perceived credibility of AI-generated content, can help mitigate biases and address trust issues.
Project Team
Petr Parshakov
Researcher
Iuliia Naidenova
Researcher
Sofia Paklina
Researcher
Nikita Matkin
Researcher
Cornel Nesseler
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Petr Parshakov, Iuliia Naidenova, Sofia Paklina, Nikita Matkin, Cornel Nesseler
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai