Foundations of Artificial Intelligence in Health

Course information

Artificial intelligence is often used interchangeably with machine learning — but ML is a subfield of AI concerned with inductive reasoning, and a great deal of what biomedical informatics actually deploys rests on deductive reasoning instead. This course covers those founding principles: knowledge representation, logic, ontologies, search, rule-based expert systems, and probabilistic inference. Machine learning shows up in the last few weeks, for contrast.

The course is about how biomedical data can be organized, represented, interpreted, searched, and applied in order to derive knowledge, make decisions, and ultimately make predictions while mitigating bias. BMIN 5200 is the first of three AI courses in a sequence: it is followed by BMIN 5210 (Machine Learning) and BMIN 5220 (Natural Language Processing).

Objectives

• Describe the challenges of representing human knowledge in silico
• Explain the difference between deduction and induction, and the role of both in AI
• Discuss past and current obstacles to implementing AI in biomedical domains
• Show how common AI approaches address real biomedical problems
• Implement core AI methods in Python — not only describe them

Prerequisites

Familiarity with basic biomedical concepts and terminology, statistics, and computer programming. BMIN 5010, BMIN 5030, and a programming course are suggested but not required. All programming is in Python. No previous exposure to AI is assumed.

Example lecture itinerary

0–30 min

Homework review

~60 min

Lecture

~25 min

In-class coding exercise

10 min

Break

~25 min

Journal club

Grading

1.0 cu. Everything is due 11:59pm on a Wednesday, the night before the class that reviews it. Late work receives a zero — answers are worked through in class the next afternoon, so a late submission cannot be accepted. Attendance is expected rather than credited: there are no points for showing up, but each unexcused absence deducts 5% from the final grade, separately from the table above. An absence is excused if you email Joe about it — what makes one unexcused is not getting in touch, so email as soon as something comes up.

Recording and attendance

Class sessions are recorded over Zoom and posted to Canvas, so you can review a lecture later or catch up on one you missed. Recordings are for enrolled students only — do not redistribute them. There is no option to attend virtually. The recordings are for review, not a substitute for being in the room: the course runs on discussion, journal club, and in-class work, none of which happen watching a video afterward.

Journal club

From Week 3 onward we spend roughly 25 minutes on one paper per week. A pair of students presents each paper — one covers background and methods, the other results and critique, and both lead the discussion. With eleven papers, everyone presents once. Presenters must also run the paper past an LLM chatbot and report which of its criticisms hold up and which are wrong.

The presentation is worth 10% of the final grade — the same as one homework assignment. Discussion on the other ten weeks is worth a further 4%: everyone is expected to have read the paper and to take part, not only the pair presenting it.

Final project

A research project (doctoral students) or an implementation project (master's and professional students), addressing a biomedical problem of your choice. Deductive inference must be central — a project built primarily on machine learning is not acceptable. Both paths produce an 8–10 page report typeset in the AAAI LaTeX template, plus a 10-minute recorded presentation. Full specification in the final project instructions.

A one-page proposal is due Wed Oct 28, stating the problem, the deductive component, the data or resources you will use, and which of the two project types you are doing. It is graded, and it is how your topic gets cleared before you build anything. It falls the week after CLIPS is taught, so implementation students propose a tool having already used it.

Generative AI policy

Although this is a course on AI, the four homework assignments and the final project — proposal, code, report, and presentation — must be completed without generative AI assistance. Violations are treated as plagiarism under Penn's Code of Academic Integrity. Two exceptions: the LLM-driven critique in journal club, where its use is required, and the in-class exercises, which earn credit for engagement rather than correctness.

Reusing course material

Everything in the public course repository — the notebooks, this site, and the structure of the course — is released under CC BY-NC-SA 4.0. Share and adapt it with attribution, non-commercially, under the same licence. Instructors elsewhere are welcome to it — the full terms are in NOTICE.md.

Anything behind a PennKey login is a different matter. Slides, journal club papers, assignment files, and class recordings on Canvas or Box are for enrolled students only and must not be redistributed — much of that material is copyrighted by publishers, and the course does not hold a licence to pass it on. Rule of thumb: if PennKey was required to get it, it is not yours to share.

Disability accommodations

Penn provides reasonable accommodations to students with disabilities approved by Student Disabilities Services. Please make an appointment with Joe as early as possible to discuss your accommodations. SDS services are free and confidential; appointments can be made at (215) 573-9235.

Office hours

Office hours will be held on Mondays, from 3:00-4:00 PM, in 3600 Civic Center Blvd., Room 05-011A. This is the same building as where class is held, but one floor down. If you cannot make that time, email Anh or Joe to set up an appointment.

Resources

Documents are hosted on Penn Box and require PennKey sign-in. Canvas is used for submitting assignments, viewing grades, and watching the class recordings.

Schedule

Twelve meetings. Lecture slides and journal club papers are on Box (PennKey required); the in-class notebooks open directly in Colab. Each journal club paper covers the previous week's lecture topic, not that day's — so the material has always been taught before anyone has to present on it. A due item is listed on the class meeting it is due for; the deadline itself is 11:59pm the night before, on the Wednesday (submitted through Canvas).