OHSU Introduction to Biomedical & Health Informatics
Course Policy for Use of ChatGPT and Generative AI
William
Hersh, MD
Professor
Department of Medical Informatics & Clinical Epidemiology
School of Medicine
Oregon Health & Science University
Last updated: September 30, 2024
This page reflects course policy for the Oregon Health & Science
University (OHSU) course, Introduction to Biomedical &
Health Informatics. There are versions of this course in
several OHSU programs, including:
ChatGPT and generative AI systems based on large language models
(LLMs) can be a useful tool for learning all kinds of topics,
including in biomedical and health informatics. These tools should
not, however, be used to substitute one’s own knowledge. Students
can “converse” with ChatGPT or generative AI systems to get ideas
for answers to questions, but the final responses to discussion
forums, quiz and test questions, and the term paper, should reflect
their own thinking, judgment, and language.
I recently published a peer-reviewed
paper showing that ChatGPT and other LLMs can "pass" the
knowledge-assessment portions of this course, which was summarized
in an OHSU
news release. This policy is based in part on the results of
this study.
It is critically important that students not "shortchange" their
learning by being overly reliant on generative AI systems. While
most scientific fields have long surpassed the amount of knowledge
that can be maintained in a human brain, it is important to have a
fundamental core of knowledge and understanding in memory to be
able to apply critical thinking to problems and analyses. In
addition, just as students must attribute use of papers, books,
and other sources in their work, they must also attribute use of
generative AI when it is used in discussion forums or assignments.
This policy is derived from the overall
OHSU policy for academic integrity, including the use of AI.
The OHSU
Biomedical Informatics Graduate Program is developing a
general policy for use of generative AI in courses, but in the
meantime, I have adopted the following guidelines for course
activities:
- Discussion forums - the purpose of the discussion
forums is for students to discuss issues that elaborate on unit
course materials. Individual forum postings are not graded,
although a component of the course grade is based on
participation in the forums, comparable to what used to be
participation in live classrooms. While students can “converse”
with generative AI to get ideas for responses to forum
questions, what is actually posted in the forum by students
should represent their own ideas, language, and thought
processes.
- Homework self-assessment - students can ask generative
AI about topics mentioned in the multiple-choice questions but
are expected to answer the questions based on their own
knowledge of materials covered in the lectures and not use
generative AI with the questions themselves until after they
have submitted their answers to the questions.
- Term paper/project - students can ask generative AI for
help in brainstorming about their term paper/project. Generative
AI systems do not write long papers, and their output tends to
focus on generalities and may be prone to confabulation,
especially in generating references. The 10-15 term
paper/project should have a focus on a specific topic, and delve
into it with coherent discussion and ample references, including
recent ones, as outlined in the course syllabus.
- Final exam - students must not access generative AI
during the final exam, just as they may not consult other humans
during the open-book exams that is given.
If you are a student and have a question on whether use of
generative AI is appropriate, please reach out directly to me (email
is best for initial contact).
As a guiding principle, we expect and require that all work
submitted be the student's own, original work. When considering
using such a generative AI tool, students should ask themselves:
Will the tool’s output be something I will be turning in directly?
In general, students may use such tools as a source of information,
but not to produce output that they intend to turn in or as a
replacement for a traditional cited reference.
Most ethical and conduct policies in our informatics educational
programs, and in the work we subsequently do as professionals, are
enforced through an honor code. We recognize we cannot
police all inappropriate use of AI or other activities. We hope that
students will find ways to use LLMs to enhance their learning but
not substitute for or become dependent on it.