Assessing the Capability of ChatGPT in Answering First- and Second-Order Knowledge Questions on Microbiology as per Competency-Based Medical Education Curriculum.

Abstract

Background and objective ChatGPT is anartificial intelligence (AI) language model that has been trained to process and respond to questions across a wide range of topics. It is also capable of solving problems in medical educational topics. However, the capability of ChatGPT to accurately answer first- and second-order knowledge questions in the field of microbiology has not been explored so far. Hence, in this study, we aimed to analyze the capability of ChatGPT in answering first- and second-order questions on the subject of microbiology. Materials and methods Based on the competency-based medical education (CBME) curriculum of the subject of microbiology, we prepared a set of first-order and second-order questions. For the total of eight modules in the CBME curriculum for microbiology, we prepared six first-order and six second-order knowledge questions according to the National Medical Commission-recommended CBME curriculum, amounting to a total of (8 x 12) 96 questions. The questions were checked for content validity by three expert microbiologists. These questions were used to converse with ChatGPT by a single user and responses were recorded for further analysis. The answers were scored by three microbiologists on a rating scale of 0-5. The average of three scores was taken as the final score for analysis. As the data were not normally distributed, we used a non-parametric statistical test. The overall scores were tested by a one-sample median test withhypothetical values of 4 and 5. The scores of answers to first-order and second-order questions were compared by the Mann-Whitney U test. Module-wise responses were tested by the Kruskall-Wallis test followed by the post hoc test for pairwise comparisons. Results The overall score of 96 answers was 4.04 ±0.37 (median: 4.17, Q1-Q3: 3.88-4.33) with the mean score ofanswers to first-order knowledge questions being 4.07 ±0.32 (median: 4.17, Q1-Q3: 4-4.33) and that of answers to second-order knowledge questions being 3.99 ±0.43 (median: 4, Q1-Q3: 3.67-4.33) (Mann-Whitney p=0.4). The score was significantly below the score of 5 (one-sample median test p<0.0001) and similar to 4 (one-sample median test p=0.09). Overall, there was a variation in median scores obtained in eight categories of topics in microbiology, indicating inconsistent performance in different topics. Conclusion The results of the study indicate that ChatGPT is capable of answering both first- and second-order knowledge questions related to the subject of microbiology. The model achieved an accuracy of approximately 80% and there was no difference between the model's capability of answering first-order questions and second-order knowledge questions. The findings of this study suggest that ChatGPT has the potential to be an effective tool for automated question-answering in the field of microbiology. However, continued improvement in the training and development of language models is necessary to enhance their performance and make them suitable for academic use.