Assessing Chat GPT's ethical proficiency by testing it's performance at the Situational Judgement Test

Objectives: This study examines the proficiency of Chat GPT, an AI large language model, in answering Situational Judgement Test (SJT) questions to gauge its ethical limitations in healthcare. The SJT is a widely used assessment tool for evaluating the fundamental competencies of medical graduates in the UK. Methods: The Oxford Assess and Progress: Situational Judgement Test book matched the inclusion criteria for providing a convenience sample of 252 SJT questions (82 multiple-choice and 170 ranking questions) for this research to ensure a fair representation of the competencies to be tested. The responses generated by the AI were compared with the answers provided in the book. Results: Despite the unavailability of population statistics and the scoring system of the SJT, it can be reasoned that Chat GPT performed fairly on SJT questions with a mean accuracy of 77.67% and a standard error of 1.09%. Moreover, it was consistent across the five tested domains and the two question types, indicating its potential as an AI decision-making tool to assist junior doctors in ethical dilemmas. Conclusion: Apart from demonstrating Chat GPT’s accuracy in situational judgement, this study highlights the need for further research and development of AI models to go beyond knowledge comprehension to enhance their ethical capabilities for better implementation in healthcare settings.