CSCI 1952A: Human‑AI Interaction

Professor Serena Booth • Fall 2026

Logistics

Class Time

TTh 2:30–3:50 PM

Location

CIT 477 Lubrano

Instructor

Prof. Serena Booth

serena_booth@brown.edu

Sam Nelson <sam_nelson@brown.edu>

Office Hours

Thursdays, 1:30–2:30 PM in CIT 427

Course Description

AI systems do not exist in a vacuum; they exist in a human world. This course examines human–AI interactions of all forms: how humans design AI systems, how humans interact with and rely on AI systems, and how AI systems shape human behavior and society.

We will study both embodied systems (e.g., robots) and non‑embodied systems (e.g., language models and decision systems), with an emphasis on technical foundations, human‑centered evaluation, and societal implications. The course is structured as a discussion‑driven seminar with a research project component.

As a final project, students will conduct a substantive research study in human–AI interaction. Projects might propose and evaluate a new method, conduct a pilot user study to identify a failure or demonstrate a capability of an existing system, or combine human data collection with method development or evaluation. With sufficient development, these projects may form the basis of submissions to venues such as AAAI, RLC, HRI, or NeurIPS.

Learning Goals

By the end of this course, students will be able to:

Critically analyze human–AI interaction research across technical and socio‑technical dimensions
Understand and apply methods for reward design, preference learning, and interactive machine learning
Evaluate human–AI systems in terms of trust calibration, usability, overreliance, robustness, and societal impact
Design and execute a substantive research study in human–AI interaction
Communicate research findings clearly through written and oral formats

These goals align with both in‑class discussions and out‑of‑class assignments, culminating in the final research project.

Prerequisites

This is not an introductory AI course. Students are expected to have taken a technical AI/ML course such as:

CSCI 0410 (Foundations of AI & ML)
CSCI 1420 (Machine Learning)
CSCI 2470 (Deep Learning)
CSCI 2951‑F (Foundations of Sequential Decision Making)

…or have equivalent experience. Students should be comfortable implementing basic ML models (e.g., training a classifier) and have some familiarity with reinforcement learning methods. The primary method that this class studies is reinforcement learning.

Course Format

This course is a seminar. Each class typically includes:

Student‑led paper presentations
Structured discussion
Occasional instructor‑led mini‑lectures
Occasional guest lectures
Interactive activities

Students are expected to actively engage in discussion and come prepared having completed all readings.

Assessment

Component	Weight
Participation	25%
Weekly reading responses & discussion questions	25%
Paper presentations	25%
Final project (report + presentation)	25%

Participation (25%)

Includes attendance, contributions to discussion, and engagement with peers. Evaluation criteria include:

Demonstrating understanding of readings through contributions to class discussions
Asking thoughtful questions that reflect engagement with peers and with material
Contributing constructively to discussion

If you find it difficult to speak up in class, there are other ways to demonstrate engagement, such as participating in smaller‑group discussions or attending office hours.

Weekly Responses (25%)

Students will submit short written responses and discussion questions prior to each class. These should resemble academic peer reviews of research papers, and can take the form of bulleted lists and rough notes. These will be graded on a 0 to 2 scale, where 0 means non‑completion, 1 means completion, and 2 is reserved for especially thoughtful responses and will be given out very sparingly.

AI use in producing these weekly responses is not allowed for two reasons. First, these responses are primarily an opportunity for you to reflect on what you understood from the papers; since the grading is mostly for completion, this is not a strict assessment of your comprehension. Second, reading AI‑generated text is typically very boring, and the course staff doesn't have the patience for it.

Paper Presentations (25%)

Students will present papers in small groups (~2 students). Presentations should:

Clearly explain key ideas and methods and relationship to other works
Critically evaluate strengths and limitations
Facilitate interactive class discussion

Final Project (25%)

Teams of 2–3 students will:

Propose a substantive research question in human–AI interaction
Design and conduct a research study, which may involve method development, human data collection, or evaluation of an existing system
Analyze the results and articulate the research contribution
Submit a 4‑page AAAI‑format paper
Present findings in a ~10‑minute presentation [exact length TBD]

Course Policies

Attendance and Participation

Attendance is expected at all sessions. Because this is a seminar, missing class significantly impacts both your learning and others'. If you must miss class due to illness or extenuating circumstances, notify the instructor as soon as possible and provide a doctor's note where applicable.

Please note: missing class for job interviews is not an excused absence. The only exception to this is for exceptionally structured interviews for which the student has no control over the scheduled date (e.g., Schwarzman Scholars, Paul and Daisy Soros Fellowship).

Late Policy

There are no late days due to the discussion‑based structure of the course.

Academic Integrity

All students are expected to adhere to Brown's Academic Code. Collaboration is generally permitted (e.g., final project teams), but all submitted work must reflect your own understanding (e.g., you must submit your own weekly reading responses).

AI Policy

You may use LLMs in this class, except for the weekly reading responses. While you are not allowed to use LLMs to write these responses for you, we encourage you to use LLMs to help you develop your understanding of the research papers. For example, if there is a proof in a paper and you want to understand the nuance, working with an LLM to fill in all of the simplified steps can help you scaffold your learning such that you can reproduce the proof.

You also may use LLMs in the construction of your final project, and we expect you to do so. This means that it is no longer sufficient to just reimplement a paper; you must contribute some new analysis, and demonstrate comprehensive understanding of the problem you choose to address, or explain why you had to put substantial effort into reimplementing a paper even with the assistance of an LLM.

Accommodations and Support

Students seeking accommodations should contact Student Accessibility Services (SAS).
Students needing short‑term academic support should contact a Dean of the College.
For mental health support, students are encouraged to contact Counseling & Psychological Services (CAPS).

Devices Policy

Laptops, tablets, and phones should not be used during class except:

For structured activities
For leading discussions with slides
For accessibility accommodations

Habitual use of devices during class without the above reasons will result in a significantly lowered class participation grade.

Schedule

Week	Date	Topic	Readings / Notes
0	Thursday, Sept 10	Introduction	N/A
1	Tuesday, Sept 15	Guest Lecture: Skye Thompson (PhD student) + Sam Nelson (TA, Masters student) How to Read a Paper and Conceptualize Research	Impacts of AI for learning “AI Assistance Reduces Persistence” — Liu et al.
1	Thursday, Sept 17	Reward specification	“The Perils of Trial-and-Error Reward Design” — Booth et al. “Towards Improving Reward Design in RL: A Reward Alignment Metric for RL Practitioners” — Muslimani et al.
2	Tuesday, Sept 22	Human feedback	TAMER — Knox et al. COACH — MacGlashan et al.
2	Thursday, Sept 24	Learning from demonstrations	DAgger — Ross et al. “Maximum Entropy Inverse Reinforcement Learning” — Ziebart et al.
3	Tuesday, Sept 29	Learning from corrections	“Learning from physical human corrections, one feature at a time” — Bajcsy et al. “Fixing Model Bugs with Natural Language Patches” — Murty et al.
3	Thursday, Oct 1	Proxy rewards	“Inverse Reward Design” — Hadfield-Menell et al.
4	Tuesday, Oct 6	Preference learning	“Deep RL from Human Preferences” — Christiano et al. “Models of Human Preference for Learning Reward Functions” — Knox et al.
4	Thursday, Oct 8	Unifying reward learning	“Reward-rational (implicit) choice” — Jeon et al.
5	Tuesday, Oct 13	Guest Lecture: Dr. Eura Nofshin	N/A
5	Thursday, Oct 15	RLHF for LLMs	“Training LMs to Follow Human Feedback” — Ouyang et al. “Constitutional AI” — Bai et al.
6	Tuesday, Oct 20	LLMs for reward design	“Eureka” — Ma et al.
6	Thursday, Oct 22	Guest Lecture: Dr. Isaac Sheidlower	N/A Note: project proposals due
7	Tuesday, Oct 27	Trust	“Trust in Automation: Designing for Appropriate Reliance” — Lee & See “Planning with Trust for Human-Robot Collaboration” — Chen et al.
7	Thursday, Oct 29	Designing AI interventions	“To Trust or to Think: Cognitive Forcing Functions Can Reduce Overreliance on AI in AI-Assisted Decision-Making” — Buçinca et al. “Improving Human Performance with Value-Aware Interventions: A Case Study in Chess” — Narayanan et al.
8	Tuesday, Nov 3	NO CLASS: Election Day	NO CLASS: Election Day
8	Thursday, Nov 5	Explanations	“Highlights: Summarizing Agent Behavior to People” — Amir & Amir “Plan explanations as model reconciliation: Moving beyond explanation as soliloquy” — Chakraborti et al.
9	Tuesday, Nov 10	Human–AI teams, shared autonomy	“Beyond Accuracy: The Role of Mental Models in Human-AI Team Performance” — Bansal et al. “Shared Autonomy with Learned Latent Actions” — Jeon et al.
9	Thursday, Nov 12	Model updates	“Updates in Human-AI Teams” — Bansal et al. “Autonomous Driving Systems: A Preliminary Naturalistic Study of the Tesla Model S” — Endsley
10	Tuesday, Nov 17	NO CLASS — work on final projects	NO CLASS
10	Thursday, Nov 19	NO CLASS — work on final projects	NO CLASS
11	Tuesday, Nov 24	Attendance optional	Serena will talk about research from the GIRAFFE lab :)
11	Thursday, Nov 26	NO CLASS: Thanksgiving	NO CLASS: Thanksgiving
12	Tuesday, Dec 1	Guest Lecture: Dr. Katie Collins	N/A
12	Thursday, Dec 3	AI & society	“Inherent Trade-offs in the Fair Determination of Risk Scores” — Kleinberg et al. (Sections 1 and 2 ONLY) “Algorithmic Monocultures and Social Welfare” — Kleinberg & Raghavan
13	Tuesday, Dec 8	Final project presentations	N/A
13	Thursday, Dec 10	Final project presentations	N/A

Deadlines

Oct 22: Final project proposal (1 paragraph + team)
Nov 20: Final project progress report due (1 page: proposed work, analysis, progress, and a timeline)
To be scheduled: in‑person check‑in around November 20th
Friday, Dec 11: Final paper (4 pages, AAAI format)

Workload Expectations

This course follows Brown's guideline of ~12 hours/week (180 hours total):

In‑class time: ~3 hours/week
Reading papers: ~4 hours/week
Weekly responses: ~2 hours/week
Project work: ~3 hours/week (increasing toward end of semester)

Materials

All readings will be provided electronically (no required textbook; $0 cost).

Process for Final Project

To ensure steady progress and feedback:

Instructor approval required for project topic
Structured milestones (proposal, progress report)
Ongoing feedback during project planning sessions
Final presentation + paper submission

Acknowledgements

This course draws on materials from:

Matthew Taylor (University of Alberta)
Scott Niekum (UMass Amherst)
Anca Dragan (UC Berkeley)
Elena Glassman (Harvard University)

Fall 2025 materials are archived here.