{"id":1856,"date":"2023-06-20T17:46:18","date_gmt":"2023-06-20T21:46:18","guid":{"rendered":"https:\/\/geiselmed.dartmouth.edu\/medical-education\/?p=1856"},"modified":"2023-06-20T17:48:09","modified_gmt":"2023-06-20T21:48:09","slug":"chatgpt-utility-in-healthcare-education-research-and-practice-systematic-review-on-the-promising-perspectives-and-valid-concerns","status":"publish","type":"post","link":"https:\/\/geiselmed.dartmouth.edu\/medical-education\/2023\/06\/20\/chatgpt-utility-in-healthcare-education-research-and-practice-systematic-review-on-the-promising-perspectives-and-valid-concerns\/","title":{"rendered":"ChatGPT Utility in Healthcare Education, Research, and Practice: Systematic Review on the Promising Perspectives and Valid Concerns"},"content":{"rendered":"
\n
\"computer
Photo by BoliviaInteligente<\/a> on Unsplas<\/a><\/em>h<\/a><\/figcaption><\/figure>\n<\/div>\n\n\n

Promoting Equity or Hallucination?<\/h2>\n\n\n\n

Today\u2019s choice of article for the alcove is inspired by the Editorial that just came out<\/a> in Academic Medicine<\/em> setting a policy for AI tools like ChatGPT: \u201cThe use of artificial intelligence (AI) tools must be disclosed at the time of submission\u2026AI tools may not be listed or designated as authors. Authors are accountable for the quality and integrity of their scholarly work.\u201d They didn\u2019t come down 100% against AI, but they are making clear that they are not sentient enough to be \u201caccountable\u201d for authorship (at least yet\u2013read some speculative fiction<\/a> for some future possibilities!). <\/p>\n\n\n\n

This editorial got me wondering about what\u2019s out there so far on tools like ChatGPT in health professions education and, amazingly, there is already a review on it (Sallam, 2023\u2013see below for full citation).  Un<\/em>amazingly, this early review is published by the Multidisciplinary Digital Publishing Institute (MDPI), which meets the criteria for a predatory publisher<\/a> (one who publishes almost anything and makes money off of it). So, let\u2019s take the findings with a grain of salt\u2013thinking of it as having a learner or research assistant do an initial search for us, but knowing we\u2019ll have to do a more rigorous search and potentially some additional research later. <\/p>\n\n\n\n

First, some interesting history: AI as something scientists study (according to our \u201cgrain of salt\u201d article) originated at Dartmouth in the summer of 1956 (see then Assistant Professor John McCarthy\u2019s proposal along with three other scholars here<\/a>). In true speculative fiction fashion, they thought that \u201cevery aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it\u201d (McCarthy et al., 2006, p. 12). <\/p>\n\n\n\n

Fast forward to 2023 and we have generative pre-trained transformer (GPT) architecture, which \u201cutilizes a neural network to process natural language, thus generating responses based on the context of the input text\u201d (Sallam, p. 2). Sallam set out to probe the (minimal!) literature to determine the potential benefits and risks of this technology. He included in his search any<\/em> published research, whether peer reviewed or not, but excluded non-academic articles from things like newspapers and magazines. This is an example of just one of many choice points you have when you do a literature review (something you should definitely<\/em> bring one of our awesome librarians in on<\/a> sooner rather than later if you\u2019re looking to publish). <\/p>\n\n\n\n

Sallam ended up with a corpus of 60 articles that he extracted data<\/a> from. (A shameless plug: you don\u2019t need interviews to do qualitative research\u2013you can use many qualitative analytic techniques on articles\u2013it\u2019s a cool form of data extraction! Reach out to me if this intrigues you\u2026) Also distinctly unamazing, very few were peer reviewed research articles:<\/p>\n\n\n\n

\"graph
*Figure taken from Sallam, 2023, p. 11<\/em><\/figcaption><\/figure>\n\n\n\n

Yet these preprints and editorials give us a starting point for thinking about ChatGPT in education and research. Some of the benefits<\/strong> are:<\/p>\n\n\n\n