中央研究院 Logo

摘要:  

Fake news is not just a technological problem—it’s a cognitive challenge. This talk presents an interactive system that leverages human inquiry to iteratively enhance explanations provided by a large language model for questionable news stories. By treating explanation generation as a dialogue, the system encourages users to ask clarifying questions, raise objections, and offer comments, prompting the AI to refine its responses in real time. With each round of Q&A, the explanations become increasingly complete, transparent, and tailored to users’ concerns. Our findings reveal that this question-driven process significantly shapes explanations. Thoughtful questions can help uncover the truth behind false claims—highlighting inconsistencies and providing enlightening context to debunk misinformation. Although the AI’s responses, influenced by its underlying knowledge and reasoning abilities, could potentially lead to misunderstandings, the overall risk of harm remains minor. Ultimately, the interactive questioning process significantly enhances the persuasiveness and clarity of generated explanations. Looking forward, we envision the system becoming a hands-on educational tool. In classroom settings, it can serve as a debate partner, presenting controversial news claims and engaging students in lively Q&A exchanges. The system actively prompts students to challenge AI statements, demand evidence, and identify logical fallacies, fostering essential news literacy skills. For students unfamiliar with debate practices, the AI can also demonstrate an automated argumentative process, serving as a helpful teaching assistant. This playful yet informative format transforms passive news consumption into active critical investigation, making lessons about misinformation both enjoyable and impactful.