Assessing the performance of Bing Chat artificial intelligence: Dental exams, clinical guidelines, and patients’ frequent questions

Abstract

ObjectivesBing Chat is a large language model artificial intelligence (AI) with online search and text generating capabilities. This study assessed its performance within the scope of dentistry in: (a) tackling exam questions for dental students, (ii) providing guidelines for dental practitioners, and (iii) answering patients’ frequently asked questions. We discuss the potential of clinical tutoring, common patient communication and impact on academia. MethodsWith the aim of assessing AI's performance in dental exams, Bing Chat was presented with 532 multiple-choice questions and awarded scores based on its answers. In evaluating guidelines for clinicians, a further set of 15 questions, each with 2 follow-up questions on clinical protocols, was presented to the AI. The answers were assessed by 4 reviewers using electronic visual analog scale. In evaluating answers to patients’ frequently asked questions, another list of 15 common questions was included in the session, with respective outputs assessed. ResultsBing Chat correctly answered 383 out of 532 multiple-choice questions in dental exam part, achieving a score of 71.99 %. As for outlining clinical protocols for practitioners, the overall assessment score was 81.05 %. In answering patients’ frequently asked questions, Bing Chat achieved an overall mean score of 83.8 %. The assessments demonstrated low inter-rater reliability. CONCLUSIONSThe overall performance of Bing Chat was above the regularly adopted passing scores, particularly in answering patient's frequently asked questions. The generated content may have biased sources. These results suggest the importance of raising clinicians’ awareness of AI's benefits and risks, as well as timely adaptations of dental education curricula, and safeguarding its use in dentistry and healthcare in general. Clinical significanceBing Chat AI performed above the passing threshold in three categories, and thus demonstrated potential for educational assistance, clinical tutoring, and answering patients’ questions. We recommend popularizing its benefits and risks among students and clinicians, while maintaining awareness of possible false information.