Aide: A Conversational Agent Based on Inoculation Theory and Large Language Models to Empower Learners to Recognize Disinformation

This paper introduces Aide, a conversational agent designed to help learners recognize the rhetorical strategies behind disinformation. Drawing on inoculation theory—which suggests that exposing individuals to controlled doses of misinformation can build cognitive resistance—the authors developed a tool that moves beyond passive fact-checking. Instead of simply teaching users to spot specific fake news stories, Aide engages them in the active creation and analysis of disinformation, helping them understand the underlying mechanisms like emotional triggers, overgeneralization, and conspiracy framing.

Aide is built as a React application powered by OpenAI’s GPT-3. The system guides users through a series of interactive rounds where they must evaluate AI-generated stories and create their own examples of misleading content. This design choice is rooted in the idea that active participation is more effective for learning than passive consumption. The agent uses a conversational, colloquial tone and incorporates emojis to maintain engagement, aiming to scaffold the user's ability to identify common disinformation tactics.

Preliminary testing with four graduate and undergraduate students revealed that while users could navigate the interface, they struggled to understand the progression of the learning journey. Participants also expressed a desire for the agent to provide explicit definitions of technical terms like "cognitive bias" or "unwarranted correlation," rather than relying on an external facilitator. The authors plan to address these usability issues by adding highlighting features and incorporating more human-to-human collaborative elements, as they believe social interaction is crucial for relating these concepts to real-world experiences.