Design GenAI to
Disseminate Expert Knowledge
For years, building AI for specialized domains such as medicine and law was constrained by limited labeled training data. The emergence of pre-trained AI models has changed that: Even off-the-shelf LLMs can make sense of the specialized knowledge embedded in clinical trial reports and legal documents, opening up a wealth of design opportunities. This project explores the design opportunities LLMs open for making specialized knowledge accessible and useful to laypeople.
We have built various LLM applications that extract clinical knowledge from biomedical literature to serve clinicians from other specialties, as well as patients and families.
- Harnessing Biomedical Literature to Calibrate Clinicians’ Trust in AI Decision Support Systems (Yang et al., CHI’23)
- Med-EASi: Finely Annotated Dataset and Models for Controllable Simplification of Medical Texts (Basu et al., AAAI’23)
- An Exploration of Knowledge‐Organizing Technologies to Advance Transdisciplinary Back Pain Research Lotz et al., the Journal of Orthopaedic Research (JOR) Spine, 2023
- Leveraging Generative AI for Clinical Evidence Synthesis Needs to Ensure Trustworthiness (Zhang et al., Journal of Biomedical Informatics, 2024)
More recently, we have been designing LLM applications to help citizens with low policy literacy participate in federal policymaking. We ask: How might we design LLM applications to empower lay users to share lived experiences persuasively with policymakers without biasing their policy views?
We thank Schmidt Futures Foundation, Google, and Cornell’s Digital and AI Literacy Initiative for supporting this project. Yeonju Jang is the student lead on the policy front. Alumni Michael Hedderich and Chandrayee Basu were student leads on the medical front.