Conversational Inoculation to Enhance Resistance to Misinformation

Paper Title

Conversational Inoculation to Enhance Resistance to Misinformation

Publication Info

• Topic area: Combating misinformation through cognitive inoculation techniques using conversational systems.
• Keywords: Cognitive inoculation, misinformation, chatbot, conversational systems, human-computer interaction, resistance to persuasion, large language models, health communication, user engagement, adaptive interventions.

Background and Problem

• Problem / challenge: Traditional cognitive inoculation methods often fail to adapt to individual attitudes or reactions during interventions, limiting their effectiveness in combating misinformation.
• Significance: Misinformation poses significant threats to public health, decision-making, and societal well-being, exacerbated by the rise of generative AI that makes misinformation more credible and easier to produce.
• Motivation and related work: Cognitive Inoculation Theory (CIT) has been validated across domains but lacks personalization and adaptability. Recent research highlights the potential of conversational systems, particularly chatbots, to address these gaps, but their application in cognitive inoculation remains underexplored.

Solution

• Proposed approach: Conversational Inoculation, implemented via MindFort, a Web-based system featuring an LLM-powered chatbot named Forty to enhance resistance to misinformation through structured, interactive dialogue.
• Novelty:
  1. Introduction of Conversational Inoculation as a dynamic, adaptive method for cognitive inoculation.
  2. Development of MindFort, leveraging LLMs to tailor conversations to user engagement and attitudes.
  3. Empirical validation of Conversational Inoculation compared to traditional methods (reading and writing).
  4. Identification of factors influencing inoculation effectiveness, including trust-building and fostering independent thinking.
• Procedure and key techniques:
  • Participants completed four lessons targeting misinformation topics (e.g., exercise and mental health, binge drinking).
  • Each lesson followed a five-stage inoculation process: pre-treatment certainty score, treatment (reading, writing, or chatbot), mid-lesson certainty score, strong counterattitudinal attack, and post-attack certainty score.
  • Chatbot Forty guided participants through refutation-building conversations, adapting to user responses and promoting critical thinking.

Results

• Concrete findings:
  • Chatbot condition showed significantly higher resistance to misinformation compared to the Control condition (p = .001, r = −.33).
  • Inoculation effectiveness of the Chatbot condition was higher than Reading (p = .004) and Writing (p = .033).
  • Chatbot treatment increased participant certainty, while Writing treatment decreased it (p = .006).
• Advantage over baselines:
  • Chatbot outperformed Control in reducing susceptibility to misinformation and showed greater inoculation effectiveness than Reading and Writing when controlling for baseline susceptibility.
• Experiments / evaluation:
  • Within-subject design with 61 participants, comparing four conditions (Chatbot, Reading, Writing, Control) across four topics.
  • Metrics included certainty score changes, intrinsic motivation inventory (IMI), and linguistic analysis of chatbot conversations.
• Limitations and future work:
  • Limited generalizability due to focus on health and nature topics.
  • Short interval between inoculation and misinformation attack.
  • Interactional friction and technical issues affecting chatbot engagement.
  • Future work should explore multi-agent systems, personalized strategies, and broader applications beyond attitudes.

Summary

This study introduces Conversational Inoculation as a novel method to combat misinformation through interactive dialogues with an LLM-powered chatbot. The MindFort system demonstrated effectiveness in enhancing resistance to misinformation, outperforming traditional methods when controlling for baseline susceptibility. Factors such as fostering independent thinking and building trust were identified as key contributors to inoculation success. While promising, the approach requires further refinement to address interactional friction and explore broader applications. Conversational Inoculation represents a scalable and engaging direction for misinformation resilience in HCI.