Beyond Disclosure: Adaptive Explanation Systems to Counteract Demographic Interaction Biases in AI Writing Evaluation

Building directly on Cheong et al.’s finding that both humans and LLMs penalize AI disclosure, but only LLMs’ punitive effects interact significantly with author demographics, this research proposes introducing adaptive, personalized AI assistance explanations that vary in transparency, rationale, and framing depending on the demographic context. The study would systematically vary how AI assistance is disclosed—from plain declarations to context-rich rationales or narrative explanations—and adapt the explanation's tone, content, or emphasis based on the author’s randomized demographic attributes or rater sensitivity profiles. The goal is to measure whether this 'explanation-as-intervention' reduces or reverses the demographic interaction effect seen in LLMs and shifts human raters’ fairness judgments. This approach moves beyond mere observation to proactively engineer communication ('adaptive transparency') to address why certain groups are penalized differently by algorithmic evaluators, potentially reframing transparency standards in AI-assisted writing toward more equitable communication. The broader impact includes informing best practices for academic publishers, employers, and platforms seeking transparency and fairness where author identity and AI assistance intersect, and contributing to actionable, systems-level bias mitigation strategies in ethical AI deployment.

References:

1. Penalizing Transparency? How AI Disclosure and Author Demographics Shape Human and AI Judgments About Writing. Inyoung Cheong, Alicia Guo, Mina Lee, Zhehui Liao, Kowe Kadoma, Dongyoung Go, Joseph Chee Chang, Peter Henderson, Mor Naaman, Amy X. Zhang (2025). arXiv.org.