Culturally Adaptive Explainable LLM Assessment for Multilingual Information Disorder: A Human-in-the-Loop Approach

LLMs struggle with culturally-specific misinformation: The paper's preliminary evidence from the multilingual Information Disorder corpus shows that current large language models "struggle to explain manipulated news consistently across communities," revealing that AI systems trained primarily on English data may miss culturally-specific manipulation tactics that matter to local audiences.

Cultural context is essential for detecting misinformation: The research demonstrates that "judgments about manipulation depend on cultural and linguistic context," meaning a one-size-fits-all approach to identifying false or misleading information is fundamentally inadequate—what counts as manipulation in one community may be framed differently in another.

Human expertise should guide AI assessment of misinformation: The paper proposes a "human-in-the-loop" framework that grounds AI decisions in "human-written rationales from native-speaking annotators," suggesting that responsible AI for detecting information disorder requires ongoing collaboration with people who understand local contexts rather than relying solely on automated systems.

Transparency matters, but must be culturally appropriate: The study evaluates "the quality and cultural appropriateness of generated rationales," indicating that explainable AI isn't just about showing your work—explanations must be meaningful and relevant to the specific cultural and linguistic communities affected by misinformation.

One-size-fits-all AI design is insufficient for global challenges: By testing adaptive prompting strategies across different languages (Farsi and Italian), the research suggests that responsible AI development requires customization for different communities rather than deploying identical systems worldwide.