Dataset Information

Reference Hallucination Score for Medical Artificial Intelligence Chatbots: Development and Usability Study.

ABSTRACT:

Background

Artificial intelligence (AI) chatbots have recently gained use in medical practice by health care practitioners. Interestingly, the output of these AI chatbots was found to have varying degrees of hallucination in content and references. Such hallucinations generate doubts about their output and their implementation.

Objective

The aim of our study was to propose a reference hallucination score (RHS) to evaluate the authenticity of AI chatbots' citations.

Methods

Six AI chatbots were challenged with the same 10 medical prompts, requesting 10 references per prompt. The RHS is composed of 6 bibliographic items and the reference's relevance to prompts' keywords. RHS was calculated for each reference, prompt, and type of prompt (basic vs complex). The average RHS was calculated for each AI chatbot and compared across the different types of prompts and AI chatbots.

Results

Bard failed to generate any references. ChatGPT 3.5 and Bing generated the highest RHS (score=11), while Elicit and SciSpace generated the lowest RHS (score=1), and Perplexity generated a middle RHS (score=7). The highest degree of hallucination was observed for reference relevancy to the prompt keywords (308/500, 61.6%), while the lowest was for reference titles (169/500, 33.8%). ChatGPT and Bing had comparable RHS (β coefficient=-0.069; P=.32), while Perplexity had significantly lower RHS than ChatGPT (β coefficient=-0.345; P<.001). AI chatbots generally had significantly higher RHS when prompted with scenarios or complex format prompts (β coefficient=0.486; P<.001).

Conclusions

The variation in RHS underscores the necessity for a robust reference evaluation tool to improve the authenticity of AI chatbots. Further, the variations highlight the importance of verifying their output and citations. Elicit and SciSpace had negligible hallucination, while ChatGPT and Bing had critical hallucination levels. The proposed AI chatbots' RHS could contribute to ongoing efforts to enhance AI's general reliability in medical research.

SUBMITTER: Aljamaan F

PROVIDER: S-EPMC11325115 | biostudies-literature | 2024 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Reference Hallucination Score for Medical Artificial Intelligence Chatbots: Development and Usability Study.

Aljamaan Fadi F Temsah Mohamad-Hani MH Altamimi Ibraheem I Al-Eyadhy Ayman A Jamal Amr A Alhasan Khalid K Mesallam Tamer A TA Farahat Mohamed M Malki Khalid H KH

JMIR medical informatics 20240731

<h4>Background</h4>Artificial intelligence (AI) chatbots have recently gained use in medical practice by health care practitioners. Interestingly, the output of these AI chatbots was found to have varying degrees of hallucination in content and references. Such hallucinations generate doubts about their output and their implementation.<h4>Objective</h4>The aim of our study was to propose a reference hallucination score (RHS) to evaluate the authenticity of AI chatbots' citations.<h4>Methods</h4> ...[more]

PMID: 39083799

Similar Datasets

Project description:BackgroundThe integration of artificial intelligence (AI) into health sciences students' education holds significant importance. The rapid advancement of AI has opened new horizons in scientific writing and has the potential to reshape human-technology interactions. AI in education may impact critical thinking, leading to unintended consequences that need to be addressed. Understanding the implications of AI adoption in education is essential for ensuring its responsible and effective use, empowering health sciences students to navigate AI-driven technologies' evolving field with essential knowledge and skills.ObjectiveThis study aims to provide details on the study protocol and the methods used to investigate the usability and efficacy of ChatGPT, a large language model. The primary focus is on assessing its role as a supplementary learning tool for improving learning processes and outcomes among undergraduate health sciences students, with a specific emphasis on chronic diseases.MethodsThis single-blinded, crossover, randomized, controlled trial is part of a broader mixed methods study, and the primary emphasis of this paper is on the quantitative component of the overall research. A total of 50 students will be recruited for this study. The alternative hypothesis posits that there will be a significant difference in learning outcomes and technology usability between students using ChatGPT (group A) and those using standard web-based tools (group B) to access resources and complete assignments. Participants will be allocated to sequence AB or BA in a 1:1 ratio using computer-generated randomization. Both arms include students' participation in a writing assignment intervention, with a washout period of 21 days between interventions. The primary outcome is the measure of the technology usability and effectiveness of ChatGPT, whereas the secondary outcome is the measure of students' perceptions and experiences with ChatGPT as a learning tool. Outcome data will be collected up to 24 hours after the interventions.ResultsThis study aims to understand the potential benefits and challenges of incorporating AI as an educational tool, particularly in the context of student learning. The findings are expected to identify critical areas that need attention and help educators develop a deeper understanding of AI's impact on the educational field. By exploring the differences in the usability and efficacy between ChatGPT and conventional web-based tools, this study seeks to inform educators and students on the responsible integration of AI into academic settings, with a specific focus on health sciences education.ConclusionsBy exploring the usability and efficacy of ChatGPT compared with conventional web-based tools, this study seeks to inform educators and students about the responsible integration of AI into academic settings.Trial registrationClinicalTrails.gov NCT05963802; https://clinicaltrials.gov/study/NCT05963802.International registered report identifier (irrid)PRR1-10.2196/51873.

Project description:BackgroundIt is unlikely that applications of artificial intelligence (AI) will completely replace physicians. However, it is very likely that AI applications will acquire many of their roles and generate new tasks in medical care. To be ready for new roles and tasks, medical students and physicians will need to understand the fundamentals of AI and data science, mathematical concepts, and related ethical and medico-legal issues in addition with the standard medical principles. Nevertheless, there is no valid and reliable instrument available in the literature to measure medical AI readiness. In this study, we have described the development of a valid and reliable psychometric measurement tool for the assessment of the perceived readiness of medical students on AI technologies and its applications in medicine.MethodsTo define medical students' required competencies on AI, a diverse set of experts' opinions were obtained by a qualitative method and were used as a theoretical framework, while creating the item pool of the scale. Exploratory Factor Analysis (EFA) and Confirmatory Factor Analysis (CFA) were applied.ResultsA total of 568 medical students during the EFA phase and 329 medical students during the CFA phase, enrolled in two different public universities in Turkey participated in this study. The initial 27-items finalized with a 22-items scale in a four-factor structure (cognition, ability, vision, and ethics), which explains 50.9% cumulative variance that resulted from the EFA. Cronbach's alpha reliability coefficient was 0.87. CFA indicated appropriate fit of the four-factor model (χ2/df = 3.81, RMSEA = 0.094, SRMR = 0.057, CFI = 0.938, and NNFI (TLI) = 0.928). These values showed that the four-factor model has construct validity.ConclusionsThe newly developed Medical Artificial Intelligence Readiness Scale for Medical Students (MAIRS-MS) was found to be valid and reliable tool for evaluation and monitoring of perceived readiness levels of medical students on AI technologies and applications. Medical schools may follow 'a physician training perspective that is compatible with AI in medicine' to their curricula by using MAIRS-MS. This scale could be benefitted by medical and health science education institutions as a valuable curriculum development tool with its learner needs assessment and participants' end-course perceived readiness opportunities.

Project description:BackgroundArtificial intelligence (AI)-based chatbots can offer personalized, engaging, and on-demand health promotion interventions.ObjectiveThe aim of this systematic review was to evaluate the feasibility, efficacy, and intervention characteristics of AI chatbots for promoting health behavior change.MethodsA comprehensive search was conducted in 7 bibliographic databases (PubMed, IEEE Xplore, ACM Digital Library, PsycINFO, Web of Science, Embase, and JMIR publications) for empirical articles published from 1980 to 2022 that evaluated the feasibility or efficacy of AI chatbots for behavior change. The screening, extraction, and analysis of the identified articles were performed by following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines.ResultsOf the 15 included studies, several demonstrated the high efficacy of AI chatbots in promoting healthy lifestyles (n=6, 40%), smoking cessation (n=4, 27%), treatment or medication adherence (n=2, 13%), and reduction in substance misuse (n=1, 7%). However, there were mixed results regarding feasibility, acceptability, and usability. Selected behavior change theories and expert consultation were used to develop the behavior change strategies of AI chatbots, including goal setting, monitoring, real-time reinforcement or feedback, and on-demand support. Real-time user-chatbot interaction data, such as user preferences and behavioral performance, were collected on the chatbot platform to identify ways of providing personalized services. The AI chatbots demonstrated potential for scalability by deployment through accessible devices and platforms (eg, smartphones and Facebook Messenger). The participants also reported that AI chatbots offered a nonjudgmental space for communicating sensitive information. However, the reported results need to be interpreted with caution because of the moderate to high risk of internal validity, insufficient description of AI techniques, and limitation for generalizability.ConclusionsAI chatbots have demonstrated the efficacy of health behavior change interventions among large and diverse populations; however, future studies need to adopt robust randomized control trials to establish definitive conclusions.

Project description:ImportanceMultimodal artificial intelligence (AI) chatbots can process complex medical image and text-based information that may improve their accuracy as a clinical diagnostic and management tool compared with unimodal, text-only AI chatbots. However, the difference in medical accuracy of multimodal and text-only chatbots in addressing questions about clinical oncology cases remains to be tested.ObjectiveTo evaluate the utility of prompt engineering (zero-shot chain-of-thought) and compare the competency of multimodal and unimodal AI chatbots to generate medically accurate responses to questions about clinical oncology cases.Design, setting, and participantsThis cross-sectional study benchmarked the medical accuracy of multiple-choice and free-text responses generated by AI chatbots in response to 79 questions about clinical oncology cases with images.ExposuresA unique set of 79 clinical oncology cases from JAMA Network Learning accessed on April 2, 2024, was posed to 10 AI chatbots.Main outcomes and measuresThe primary outcome was medical accuracy evaluated by the number of correct responses by each AI chatbot. Multiple-choice responses were marked as correct based on the ground-truth, correct answer. Free-text responses were rated by a team of oncology specialists in duplicate and marked as correct based on consensus or resolved by a review of a third oncology specialist.ResultsThis study evaluated 10 chatbots, including 3 multimodal and 7 unimodal chatbots. On the multiple-choice evaluation, the top-performing chatbot was chatbot 10 (57 of 79 [72.15%]), followed by the multimodal chatbot 2 (56 of 79 [70.89%]) and chatbot 5 (54 of 79 [68.35%]). On the free-text evaluation, the top-performing chatbots were chatbot 5, chatbot 7, and the multimodal chatbot 2 (30 of 79 [37.97%]), followed by chatbot 10 (29 of 79 [36.71%]) and chatbot 8 and the multimodal chatbot 3 (25 of 79 [31.65%]). The accuracy of multimodal chatbots decreased when tested on cases with multiple images compared with questions with single images. Nine out of 10 chatbots, including all 3 multimodal chatbots, demonstrated decreased accuracy of their free-text responses compared with multiple-choice responses to questions about cancer cases.Conclusions and relevanceIn this cross-sectional study of chatbot accuracy tested on clinical oncology cases, multimodal chatbots were not consistently more accurate than unimodal chatbots. These results suggest that further research is required to optimize multimodal chatbots to make more use of information from images to improve oncology-specific medical accuracy and reliability.

Project description:BackgroundThe use of patient health and treatment information captured in structured and unstructured formats in computerized electronic health record (EHR) repositories could potentially augment the detection of safety signals for drug products regulated by the US Food and Drug Administration (FDA). Natural language processing and other artificial intelligence (AI) techniques provide novel methodologies that could be leveraged to extract clinically useful information from EHR resources.ObjectiveOur aim is to develop a novel AI-enabled software prototype to identify adverse drug event (ADE) safety signals from free-text discharge summaries in EHRs to enhance opioid drug safety and research activities at the FDA.MethodsWe developed a prototype for web-based software that leverages keyword and trigger-phrase searching with rule-based algorithms and deep learning to extract candidate ADEs for specific opioid drugs from discharge summaries in the Medical Information Mart for Intensive Care III (MIMIC III) database. The prototype uses MedSpacy components to identify relevant sections of discharge summaries and a pretrained natural language processing (NLP) model, Spark NLP for Healthcare, for named entity recognition. Fifteen FDA staff members provided feedback on the prototype's features and functionalities.ResultsUsing the prototype, we were able to identify known, labeled, opioid-related adverse drug reactions from text in EHRs. The AI-enabled model achieved accuracy, recall, precision, and F1-scores of 0.66, 0.69, 0.64, and 0.67, respectively. FDA participants assessed the prototype as highly desirable in user satisfaction, visualizations, and in the potential to support drug safety signal detection for opioid drugs from EHR data while saving time and manual effort. Actionable design recommendations included (1) enlarging the tabs and visualizations; (2) enabling more flexibility and customizations to fit end users' individual needs; (3) providing additional instructional resources; (4) adding multiple graph export functionality; and (5) adding project summaries.ConclusionsThe novel prototype uses innovative AI-based techniques to automate searching for, extracting, and analyzing clinically useful information captured in unstructured text in EHRs. It increases efficiency in harnessing real-world data for opioid drug safety and increases the usability of the data to support regulatory review while decreasing the manual research burden.

Dataset Information

Reference Hallucination Score for Medical Artificial Intelligence Chatbots: Development and Usability Study.

Background

Objective

Methods

Results

Conclusions

Publications

Reference Hallucination Score for Medical Artificial Intelligence Chatbots: Development and Usability Study.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets