Ethics in AI Judgment: Bridging Human Insight and GPT's Prowess in Ethics Consultation Training and Assessment
Thursday, September 19, 2024
10:45 AM – 11:45 AM CT
Location: Regency Ballroom C (First Floor)
Abstract: Artificial Intelligence (AI), especially widely accessible public models like ChatGPT, is revolutionizing various facets of healthcare, including education, decision-making, and ethics consultation. Despite its rapid integration across these areas, a significant gap remains in understanding and harnessing AI's potential in health education assessment and clinical ethics consultation—a domain where nuanced decision-making intersects with moral values. This underexplored area presents both promising opportunities and substantial challenges as we seek to align AI capabilities with the intricate requirements of ethical consultation and trainee education in healthcare settings. To explore the alignment between a trained AI system (Chat GPT), using the Ethics Consult Quality Assessment Tool (ECQAT), and human graders in assessing the quality of ethics consultation we first trained GPT with ECQAT. We then had GPT and two humans each independently grade 18 ethics consultation notes and analyzed these using Cohen's Kappa for agreement on pass/fail categorization, agreement percentages on passing scores (≥3), and Bland-Altman plots for a visual representation of the grading alignment. In this presentation, we discuss the potential of AI in aligning with human expertise in grading the quality of ethics consults, its emerging role in ethics education and clinical practice assessment, and its limitations with qualitative feedback in this context. We then address opportunities for future exploration and potential uses, including benefits and harms related to implementation and application in diverse contexts. Finally, we consider the ways in which publicly available AI can serve as an adjunct to clinical ethicist education and training more broadly.
Learning Objectives:
After participating in this conference, attendees should be able to:
Evaluate the congruence between an ECQAT-trained GPT model and human graders in numerically grading ethics consults.
Describe the accuracy of the AI model in categorizing ethics consults as pass or fail, mirroring human grading standards.
Christian Vercler, MD – University of Michigan Medical School; Andrew Barnosky, DO, MPH – University of Michigan Medical School; Janice Firn, PhD, MSW, HEC-C – University of Michigan Medical School
