Page 14 - AIH-2-1

Artificial Intelligence in Health AI in higher medical education

risk that a student using ChatGTP or a student generating, examinations, the United States Medical Licensing Exam
for example, text for his or her work will not be able to (USMLE), and certain university tests. In all cases, ChatGPT
5
find irregularities that may be important. Chio et al. put came out sufficiently well to be deemed able to set and mark
83
forward the same argument, raising ChatGPT’s lack of tests, although Giannos and Delardas found that it had
89
critical reflection in the case of nurse education. ChatGTP a poor knowledge of science and mathematics. Indeed, in
does not act in such a way as to assess the credibility of many areas of medical education, this tool does not offer
sources; when asked to provide literature, in many cases, specialized knowledge, as in the case of pediatric cardiology
it creates references that do not exist. This all amounts education. In a further study, Danesh et al. (2024) tested
90
to a substantial ethical risk. Simply put, the output from both the free and the premium versions of ChatGTP
GenText cannot be completely relied upon and needs in terms of its ability to pass professional examinations
human moderation. In a study conducted by Tsegay et al. (excluding questions containing imaging data). ChatGTP
84
on writing in an undergraduate medical degree course in was able to answer 50% of the questions correctly in the
Darussalam, they found substantial inclusion of ChatGPT- free version and 70% of the questions in the paid version.
generated text and citations to non-existent references. As Similar examination results were obtained in the fields of
a result, they propose that educators should be more aware orthopedics and health professional exams. Finally, Sevgi
91
92
of Gen Text detection tools. Another ethical issue arising et al. proposed an evaluation of ChatGTP in the field of
93
here concerns the fact that ChatGPT can give answers neurosurgery by asking it to create questions at the level of
to students very quickly: The normal study processes a neurosurgery board exam. The question format was to be
of critical thinking (assessing information, making multiple choices and the answers were also to be generated.
connections, and drawing conclusions) are thus bypassed. Next, it was asked to devise artificial neurosurgical cases
Another ethical issue is the possibility of students writing with examinations and treatment histories. The final stage
potentially fraudulent assignments. This indicates the need involved an evaluation of the tool’s ability to create articles
for teaching how ChatGPT can be used appropriately in in this area. It turned out that the proposed cases did serve
the learning situation. Apart from discursive documents to help neurosurgery students develop their knowledge.
and reports, there are many other types of text that can be However, it transpired that a correct assessment of the
generated. These include personalized learning plans and solutions proposed by ChatGTP was only possible under
treatment plans. If these are relied on without scrutiny the supervision of a person with appropriate medical
or moderation, adverse results could occur, such as knowledge, in this case, an experienced neurosurgeon.
misdiagnosis and bad treatment (of oneself and others).
This needs to be regulated on a high level and accounted 5. Conclusion
for in local ethical policies and educational practices, and The application of AI allows the efficient analysis of huge
85
calls are emerging for the development of new educational amounts of data in a finite time. It can be considered a
governance of AI in higher medical education. 86 powerful computational tool for solving complex problems
The existing system of text types and uses provides related to pattern recognition, classification, grouping,
a benchmark against which to assess ChatGPT, but the behavior prediction, or, more generally, approximation of
components of that benchmark are not 100% accurate, functions and processes. Consequently, AI is becoming a
acceptable, correct, or free from contestable interpretation. highly precise tool in medicine. It is worth stressing that
Indeed, the production of error (however defined) and compared to statistical methods; it is more susceptible to
the principle of falsifiability are a necessary condition for various types of threats in comparison, which is a result
the advancement of knowledge. In that case, if ChatGPT of its complexity, data dependence, and susceptibility to
87
is to be criticized as falling short, as it does, to what extent adversarial attacks. Although AI offers many benefits in
can it serve as a useful tool? The immediate answer lies medical education, ethical concerns about its accessibility,
in a case-by-case detailed evaluation and benchmarking validity, use, and implementation raise many questions.
process where each instance is allocated a point on a scale AI can be implemented in medical education in a variety
of usefulness and risk. Specific cases include that discussed of beneficial and relatively uncontroversial ways. These
by Abdelhady and Davis, who reported that ChatGPT was include the rapid analysis of large-scale simulated datasets
88
able to record operative notes extremely quickly and to a (thus obviating requirements of real-life patient data
high level of accuracy compared with manual procedures regulation), pattern recognition and diagnostics (as in
and was deemed acceptable by surgeons and patients alike. radiology), general educational support in the design of
Furthermore, several research studies have been conducted personalized learning programs (at least on a basic level),
where ChatGPT was required to take a variety of medical models built in VR as teaching aids, and uses of GenText
tests, such as the UK BMAT, TMUA, LNAT, and TSA such as rapid assembly of post-operative notes.

Volume 2 Issue 1 (2025) 8 doi: 10.36922/aih.3276