Truth, Tone, and Trust: Integrating an AI Exercise to Foster Responsible Science Communication in a Graduate Biomedical Science Course

As generative AI becomes a standard fixture in graduate education, the pedagogical challenge has shifted from detection to integration. This article describes a curriculum intervention developed for Anatomy of Scientific Error, a graduate-level course on scientific error analysis in interdisciplinary research practice and communication at the Johns Hopkins Bloomberg School of Public Health. To inform the curriculum design, we first administered a needs-assessment survey to the cohort. Results revealed a critical trust gap: while students utilized generative artificial intelligence extensively for efficiency, they expressed anxiety regarding hallucinations and a lack of verification skills. While the course integrated multiple AI-related tasks, this paper focuses on a specific assignment titled "Truth, Tone, and Trust." Situated in the sixth week of the curriculum, this assignment represented a deliberate design pivot away from standard text-based prompts toward a multimodal, high-engagement activity. Students were tasked with improving three flawed, AI-generated variations of a recent real-world press release regarding the R21 malaria vaccine, using track changes to correct errors, and recording a video defense of their editorial decisions. This approach shifts the focus from regulating AI to stewarding knowledge, ensuring that efficiency does not come at the cost of critical thinking.