Collaborative Evaluation of Deepfake Text with Deliberation-Enhancing Dialogue Systems

View on arXiv ← Back to list

Authors: Jooyoung Lee, Xiaochen Zhu, Georgi Karadzhov, Tom Stafford, Andreas Vlachos, Dongwon Lee

Published: 2025-03-06 20:19:38+00:00

AI Summary

This research investigates the use of DeepFakeDeLiBot, a deliberation-enhancing chatbot, to improve group accuracy in detecting deepfake text. Group-based detection significantly outperformed individual efforts, while the chatbot's impact on accuracy wasn't statistically significant but positively affected group dynamics.

Abstract

The proliferation of generative models has presented significant challenges in distinguishing authentic human-authored content from deepfake content. Collaborative human efforts, augmented by AI tools, present a promising solution. In this study, we explore the potential of DeepFakeDeLiBot, a deliberation-enhancing chatbot, to support groups in detecting deepfake text. Our findings reveal that group-based problem-solving significantly improves the accuracy of identifying machine-generated paragraphs compared to individual efforts. While engagement with DeepFakeDeLiBot does not yield substantial performance gains overall, it enhances group dynamics by fostering greater participant engagement, consensus building, and the frequency and diversity of reasoning-based utterances. Additionally, participants with higher perceived effectiveness of group collaboration exhibited performance benefits from DeepFakeDeLiBot. These findings underscore the potential of deliberative chatbots in fostering interactive and productive group dynamics while ensuring accuracy in collaborative deepfake text detection. textit{Dataset and source code used in this study will be made publicly available upon acceptance of the manuscript.

Key findings

Group-based deepfake text detection significantly outperformed individual efforts. While DeepFakeDeLiBot did not significantly improve detection accuracy, it enhanced group dynamics such as engagement, consensus building, and reasoning-based utterances. The chatbot's positive impact was particularly noticeable in groups that perceived their collaboration as effective.

Approach

The study uses a collaborative approach where participants detect deepfake text individually and then in groups, with half the groups using DeepFakeDeLiBot, a chatbot designed to enhance deliberation through prompting. The researchers compare individual and group performance with and without the chatbot, analyzing both accuracy and group dynamics.

Datasets

A dataset of 14 articles, each with two human-written paragraphs and one GPT-2 or GPT-3.5 generated paragraph. The dataset was partially created by regenerating paragraphs from an existing dataset using GPT-3.5.

Model(s)

GPT-2, GPT-3.5, Sentence-T5, Flan-T5, and GPT-2 small model (for perplexity calculation). DeepFakeDeLiBot is a retrieval-based dialogue agent using Sentence-T5 for embedding and Flan-T5 for utterance refinement.

Author countries

USA, United Kingdom

← Previous