CSCI 1952A: Human‑AI Interaction

Professor Serena Booth • Fall 2025

Logistics

Class Time

TTh 1:00–2:20 PM

Location

UPDATED LOCATION: CIT 477 Lubrano

Office Hours

Prof. Booth: Tuesdays 2:30–3:30 PM in CIT 427

Devices

You should not use laptops or cell phones during class except for planned activities, or where accommodation is necessary.

Course Description

AI systems do not exist in a vacuum; they exist in a human world. This course will examine human‑AI interaction, spanning the gamut from how humans design AI systems, to how humans work with and alongside AI systems, to how AI systems affect human behaviors. Sometimes the AI systems under scrutiny will be embodied (robots!), sometimes not. The course will be both discussion and project based. As a final project, students will conduct a pilot study on human‑AI interaction. This course is aimed at graduate students or others with interest in conducting human‑AI interaction research. With perseverance and dedication, these course projects could be converted into conference papers at venues like AAAI, HRI, or NeurIPS.

Background Preparation

This is not an introductory course on AI. Our expectation is that you have taken a technical course on AI — like CSCI 0410 (Intro to AI), CSCI 1420 (Machine Learning), or CSCI 2470 (Deep Learning). You should have basic competency in building AI systems; for example, you should be able to train an MNIST classifier with only minimal support from the internet. The only exception to this background preparation is if you already have substantial research experience in another area of computer science — i.e., you have led the work of a technical paper. If you would like to discuss your preparation, please get in touch with the instructor.

Grading

This course is a seminar, and, as such, it requires substantial participation. The grade will be comprised as:

15%: participation. You must attend class and actively engage in class discussions.
15%: weekly comprehension and discussion questions. After reading each paper, you will be required to submit short responses and discussion questions in preparation for class.
30%: presentation of papers. Each student will be required to lead the discussion for two classes in small groups. These groups will consist of teams of 2 (occasionally 3) students.
40%: final project. Teams of 2–3 students will work together to propose a project, define their hypotheses and milestones, conduct the project, submit a 4‑page write‑up of this project in AAAI format, and present this project in a 10‑minute presentation at the end of term.

Late Submissions and Late Days

Since this is a seminar class, there are no late submissions or late days. We expect you to attend every class and to submit every assignment. If you cannot attend class or complete an assignment, please obtain a note from health services where appropriate. In other extenuating circumstances, reach out to Serena.

Presentation of Papers

For your assigned classes, you are expected to guide the discussion. The typical class has two assigned readings; in such cases, a typical discussion will consist of two 10‑minute presentations for each of the papers, alongside class discussion for each paper, with a 5‑minute break in between the two papers. You are free to choose another format. In either case, you should send the instructor a copy of your slides or planned activities 24 hours prior to the class; you should incorporate any instructor feedback into your presentation before class. You may propose a modification to the papers for that class (while maintaining the theme of that class). This modification proposal must be shared at least a week in advance of that class. And, the instructor reserves the right to overrule your proposed papers :)

Discussion Form

24 hours before each class, you should fill in this form to submit your weekly comprehension and discussion questions.

Schedule

Week	Day	Date	Theme	Sub-topic	Papers
0	Thursday	Sept 4	Introduction to Human-AI Interaction		N/A
1	Tuesday	Sept 9	Societal importance of interaction	On the consequences of our decisions	Paper ID 100) “Do artifacts have politics?” — Langdon Winner Paper ID 101) “An investigation of the Therac-25 accidents” — Leveson and Turner Paper ID 102) “Uber insiders describe infighting and questionable decisions before its self-driving car killed a pedestrian” — Business Insider
1	Thursday	Sept 11		Emergent interaction vs. designed interaction	Paper ID 103) "ELIZA—a computer program for the study of natural language communication between man and machine” — Joseph Weizenbaum Paper ID 104) “Principles of mixed-initiative user interfaces” — Eric Horvitz
2	Tuesday	Sept 16	Designing AI	Specifying reward functions	Paper ID 105) “Where do rewards come from?” — Satinder Singh et al. Paper ID 106) “The perils of trial-and-error reward design: misdesign through overfitting and invalid task specifications” — Booth et al. Paper ID 107) “Reward is enough” — David Silver et al.
2	Thursday	Sept 18		Learning reward functions from feedback	Paper ID 108) “Interactively shaping agents via human reinforcement: The TAMER framework” — Knox et al. Paper ID 109) “Interactive Learning from Policy-Dependent Human Feedback” — MacGlashan et al.
3	Tuesday	Sept 23	Designing AI	Shared autonomy	Paper ID 110) “Controlling assistive robots with learned latent actions” — Losey et al. Paper ID 111) “Shared autonomy via deep reinforcement learning” — Reddy et al.
3	Thursday	Sept 25		Guest Lecture by Dr. Isaac Sheildlower. “Providing people with control in interaction.”	N/A
4	Tuesday	Sept 30	Designing AI	Imitating human behavior	Paper ID 112) “Apprenticeship Learning via Inverse Reinforcement Learning” — Ng et al. Paper ID 113) “A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning (DAgger)” — Ross et al.
4	Thursday	Oct 2		Learning from observations and corrections	Paper ID 114) “Inverse Reward Design” — Hadfield-Menell et al. Paper ID 115) “Fixing Model Bugs with Natural Language Patches” — Shikhar Murty et al.
5	Tuesday	Oct 7	Designing AI	Learning reward functions 2	Paper ID 116) “Deep reinforcement learning from human preferences” — Christiano et al. Paper ID 117) “Models of human preference for learning reward functions” — Knox et al.
5	Thursday	Oct 9		Reward models for LLMs	Paper ID 118) “Training language models to follow instructions with human feedback” Paper ID 119) “RLAIF: Scaling Reinforcement Learning from Human Feedback with AI Feedback”
6	Tuesday	Oct 14	Using AI	Trust and automation bias	Paper 120) “Trust in Automation: Designing for Appropriate Reliance” — Lee and See Paper 121) “Humans and Automation: Use, Misuse, Disuse, Abuse” — Parasuraman and Riley
6	Thursday	Oct 16	N/A	Final Project Planning	N/A
7	Tuesday	Oct 21	Guest lectures	Prof. Brad Knox Guest Lecture	N/A
7	Thursday	Oct 23		Trust and automation in practice	Paper 122) “Hello AI: uncovering the onboarding needs of medical practitioners for human-AI collaborative decision-making” — Carrie Cai et al. Paper 123) “Planning with trust for human-robot collaboration”
8	Tuesday	Oct 28	Using AI	AI explanations	Paper 124) “Towards a rigorous science of interpretable machine learning” — Finale Doshi-Velez & Been Kim Paper 125) “The Mythos of Model Explainability” — Zachary Lipton
8	Thursday	Oct 30		Mental models, overreliance	Paper 126) “To trust or to think: cognitive forcing functions can reduce overreliance on AI in AI-assisted decision-making” Paper 127) “Beyond accuracy: The role of mental models in human-AI team performance”
9	Tuesday	Nov 4	Using AI	Explanations and policy summaries	Paper 128) “Why should I trust you?” Explaining the predictions of any classifier Paper 129) “Highlights: Summarizing agent behavior to people”
9	Thursday	Nov 6		Failures of explanations	Paper 130) “Sanity checks for saliency maps” Paper 131) “Do feature attribution methods correctly attribute features?”
10	Tuesday	Nov 11	Using AI	Human-AI Teams	Paper 132) “Updates in human-AI teams: Understanding and addressing the performance/compatibility tradeoff” — Gagan Bansal et al. Paper 133) “Autonomous driving systems: A preliminary naturalistic study of the Tesla Model S”
10	Thursday	Nov 13		Fairness and emergent behaviors	Paper 134) “Inherent trade-offs in the fair determination of risk scores” — Kleinberg et al. Paper 135) “Discrimination Exposed? On the Reliability of Explanations for Discrimination Detection”
11	Tuesday	Nov 18	Guest lectures	“Human-AI Interaction”: Prof. Matt Taylor Guest Lecture	N/A
11	Thursday	Nov 20		“Mechanistic Interpretability”: Dr. Yilun Zhou Guest Lecture	N/A
12	Tuesday	Nov 25	Societal impacts of AI	AI Safety & Societal Impact	Paper 136) “Concrete problems in AI Safety” Paper 137) “Algorithmic monoculture and social welfare” — Jon Kleinberg and Manish Raghavan
12	Thursday	Nov 27		—	THANKSGIVING HOLIDAY
13	Tuesday	Dec 2	No class	Work on final projects	N/A
13	Thursday	Dec 4		AI Policy	Paper 138) “Blueprint for an AI Bill of Rights” — White House OSTP Paper 139) “Position: Strong Consumer Protection is an Inalienable Defense for AI Safety in the United States” — Serena Booth
14	Tuesday	Dec 9	Final project presentations	Final Project Presentations	Final Project Presentations
14	Thursday	Dec 11		Final Project Presentations	Final Project Presentations

Deadlines

Oct 23: Final Project 1-paragraph proposal and team due
Nov 11: Final Project 1-page progress report due
Nov 25: Final Project updated 1–2 page progress report due
Dec 11: Final Project 4-page double-column AAAI format report due

Acknowledgements

This class draws on the syllabi of others. These are Interactive Machine Learning taught by Prof. Matthew Taylor at the University of Alberta; Human‑Centric Machine Learning taught by Prof. Scott Niekum at the University of Massachusetts at Amherst; Algorithmic Human‑Robot Interaction taught by Prof. Anca Dragan at the University of California at Berkeley; and Human‑AI Interaction taught by Prof. Elena Glassman at Harvard University.