ChatGPT in medical imaging higher education

Area of Science:

• Medical Radiation Science
• Higher Education
• Scientific Writing

Background:

• Advances in Artificial Intelligence (AI), specifically chatbots like ChatGPT (GPT-3.5), challenge academic integrity in higher education and scientific writing.
• The capabilities of AI in generating human-like responses necessitate an objective evaluation of its impact on academic standards.

Purpose of the Study:

• To evaluate the performance of ChatGPT against undergraduate medical radiation science students in exams and written assignments.
• To assess the risks and benefits of AI tools in academic and scientific writing within this field.

Main Methods:

• ChatGPT was tested on six subjects across three undergraduate years, completing three written assignments and six exams.
• Submissions were marked against standardized rubrics, compared to student cohorts, and evaluated for similarity and AI detection using Turnitin.

Main Results:

• ChatGPT performed below average student performance in all written tasks, with performance decreasing in advanced subjects.
• The AI tool outperformed average students in foundational exams but lacked the depth for discipline-specific subjects.
• Turnitin software readily identified ChatGPT submissions as AI-generated, with instances of errors and fabricated information.

Conclusions:

• ChatGPT presents both risks to academic integrity and potential as a learning tool, primarily impacting lower-order learning outcomes.
• Higher-order cognitive tasks and professional communication remain areas where AI's utility and risks are constrained.
• Current AI models have limited capacity for advanced academic work and professional application in medical radiation science.