Student guide · Page 8 of 8
Where AI fits
What you may use a language model for in this course, what you may not, and the reasoning behind the line.
There is a line in this course, and it is drawn deliberately rather than defensively. AI is allowed to help you produce your work. It is not allowed to do the parts of the work that are the point.
Allowed
- Drafting and tightening a finding once you know what you want to say.
- Organizing and cross-referencing the accumulating record to find what your service already filed on a region.
- Explaining a mechanism, a term, or a structure you do not recognize.
- Image reconstruction and denoising applied to the CT itself, which the platform does for you.
Not allowed
- Asking a model to find the abnormality on the CT. Spotting the thing that should not be there is the skill.
- Having a model decide what is worth filing, or write a finding you did not make. Judging what matters is the skill.
- Generating observations you did not observe, in imaging or in the dissection.
Why the line sits there
Two different reasons, and it is worth separating them.
Reading images is a perceptual skill that only builds by doing it. If a model circles the lipoma before you have looked, you have not learned to see a lipoma. You have learned to read a circle. The cost of skipping it is paid later, on a patient.
A signed finding is a permanent claim about a real person. Once you sign it, the other seven or eight students on your donor act on it, faculty grade it, and it cannot be deleted, only appended to. A model does not know what happened at your table. It will produce something fluent and plausible, and fluency is what makes a wrong finding survive.
The honest version
Nobody is auditing your browser. This line holds because you enforce it, in a course where the consequence of not learning something is deferred by a few years and then arrives anyway. Use the tool where it saves you time. Do not use it where it saves you the effort that was the assignment.