Comprehensive Summary
This article goes over a cross-sectional observational study comparing the ability of both Deepseek and GPT4.0 to develop informed consent forms. This was achieved by asking a set of standardized questions based on 10 commonly performed ophthalmologic procedures, asking Deepseek and GPT4.0 to generate informed consent forms. Both AI models achieved a high level of correctness with 9/10 procedures, with no significant statistical difference. In terms of completeness, Deepseek scored higher, with 60% of consents being complete complete compared to 4/10 for GPT4.0. Both models, however, were not able to produce a viable consent form fit for Indian standards. The authors were able to conclude that currently, both chatbots fall short in reliably producing consent forms, indicating that they need to further refine their knowledge base to improve their abilities for the future, increasing efficiency in healthcare.
Outcomes and Implications
Preparing a consent form can be a time consuming task, taking away from valuable time with a patient for a doctor or other healthcare worker. For now, they still require human oversight to be refined for practical use. As the technology develops, this oversight can be reduced, but ethical considerations would have it that it is never truly gone. Even with this oversight, it still can increase efficiency, creating a human-AI collaboration model that can take the burden of clinicians. Overall, these chatbots can be used as a crutch, but not be completely independent as of current models.