Fall 2026 · College of William & Mary
Artificial Intelligence for Social Impact
CSCI 100 · Section 01 · CRN 13164 · Critical Questions in Computer Science
AI gives us powerful tools for analysing vast amounts of data, finding hidden patterns, and informing decisions in public health, agriculture, energy sustainability, and climate change. But that power is not self-directing. Deployed without domain expertise or a human-centred approach, AI can do serious harm: entrenching bias, violating privacy, and undermining trust. This course asks a single question: what does it take to use AI responsibly, for the greater societal good?
You will hear directly from nonprofits, government bodies, and researchers deploying AI against real problems around the world. You will think critically about what AI can and cannot do, engage with questions of fairness, privacy, and security, and complete a project applying AI ideas to a social challenge you care about. No prior programming experience is required: every programming skill the course needs is taught in class.
- Instructor
- Ayan Mukhopadhyay · contact
- Lectures
- Tuesdays & Thursdays, 3:30–4:50 PM
- Location
- Integrated Science Center 3348
- Office hours
- To be discussed in class
- Credits
- 4 credit hours across 3 contact hours per week
- Prerequisites
- None
COLL 100: big ideas and communication
This course is part of William & Mary’s COLL 100 curriculum, which organises courses around big ideas: the significant questions, concepts, and discoveries that have shaped how we understand the world. The big idea here is that AI is among the fastest-growing and most consequential technologies in human history, and that using it well for societal good demands both technical understanding and genuine engagement with the human, ethical, and domain context it lands in. We draw on computer science, data science, philosophy of technology, and public policy, and you are encouraged to bring your own disciplinary perspective to bear.
COLL 100 also emphasises communicating in different forms. Written communication is developed through the talk reflections and the policy brief; oral and visual communication through the mid-term and final presentations, where you have to make technical and ethical arguments land with a non-specialist audience using slides, diagrams, and live discussion.
What you will be able to do
- Explain how supervised, unsupervised, and large-language-model systems work conceptually, including what they optimise for and where they fail.
- Evaluate real-world AI deployments across social-impact domains (wildlife conservation, financial inclusion, maternal health, rural development, food security, and the interpretability of high-stakes systems), with attention to fairness, privacy, and accountability.
- Use Colab-based Python notebooks to modify, run, and interpret basic machine-learning and optimisation code.
- Design and critically test a prompt-based AI tool for a social-good use case, including edge-case and adversarial evaluation.
- Communicate technical and ethical arguments about AI to a non-specialist audience, in writing, in slides, and live.
How the course is structured, and why it is not one-to-one.
The course alternates between technical foundation weeks, where we build up a specific method, and applied weeks, where invited speakers describe real deployments and we work through case studies together. These do not map onto each other one-to-one, and they are not meant to. Real systems almost never use a single technique in isolation: an anti-poaching platform combines supervised learning with game-theoretic optimisation and multi-agent planning; a maternal-health chatbot combines language models with information retrieval and human-in-the-loop feedback; a food-distribution dispatcher combines forecasting, optimisation, and matching.
So sometimes we preview a method before it is formally introduced (you will see the machinery in Week 10; for now, notice what the system is doing, and for whom, and sometimes we revisit a technique through a new application. The goal is for AI to become an integrated toolkit for social impact rather than a collection of disconnected methods.
Schedule
Fifteen weeks, two sessions each. Exact dates and all deadlines are on the course management system.
| Week | Sessions |
|---|---|
| Week 1 | Introduction: what is AI? Course overview; COLL 100 framing; AI experience survey. |
| Week 2 | Algorithms, data, and decisions Google Colab and computational thinking Conceptual foundations, then a hands-on Colab introduction. |
| Week 3 | Supervised learning I: learning from labeled examples Invited talk: Lily Xu, Columbia University In-class deliberation. |
| Week 4 | Supervised learning II: overfitting, generalization, and evaluation Invited talk: Fundación CapitalA1 out |
| Week 5 | Invited talk: Amulya Yadav, Penn State Information literacy In-class deliberation, then the librarian session with Swem Library. |
| Week 6 | Unsupervised learning I: finding structure without labels Unsupervised learning II: applications and limitsA2 out Assignment 1 due. |
| Week 7 | Project workshop Fall break: no class Project proposal and team declaration due. |
| Week 8 | AI in agriculture, rural development, and sustainability Invited talk: Purnima Venkat, Goonj In-class deliberation. Assignment 2 due. |
| Week 9 | Optimization and resource allocation Invited talk: Ryan Shi, University of PittsburghA3 out Linear programming and allocation trade-offs; in-class deliberation. |
| Week 10 | Fairness in AI: metrics, trade-offs, and the COMPAS case Privacy, surveillance, and data rights |
| Week 11 | Election Day: no class Invited talk: Meiyi Ma, Vanderbilt University In-class deliberation. Assignment 3 due. |
| Week 12 | Large language models I: how they work Large language models II: prompting, evaluation, and limitsA4 out |
| Week 13 | Mid-term lightning talks Synthesis: responsible AI deployment Slides due before class. Assignment 4 due. |
| Week 14 | Peer review workshop (online) Thanksgiving: no class Draft policy brief due before the workshop. |
| Week 15 | Final project presentations, part 1 Final project presentations, part 2 Final policy brief due at the end of the week. |
Invited talks
A central feature of the course is a series of talks from practitioners and researchers deploying AI for social good around the world. Each session runs about 90 minutes: 30 minutes of presentation, 15 minutes of live Q&A, and the remainder given over to in-class deliberation on what we heard: what worked, what did not, and what we would want to know before deploying a system like it.
That deliberation is a structured discussion, and you are expected to bring questions and observations formed during the talk itself. It is the moment where the course’s ethical disposition gets exercised on live material rather than a case study from a book. I will circulate a paper or two by each speaker beforehand; reading at least the introduction will let you engage with them meaningfully.
| Week | Speaker | Expertise |
|---|---|---|
| Week 3 | Lily XuColumbia University | AI for wildlife conservation |
| Week 4 | AngelicaFundación Capital | Micro-enterprise and financial inclusion |
| Week 5 | Amulya YadavPennsylvania State University | AI for maternal health |
| Week 8 | Purnima VenkatGoonj | Disaster relief and rural development |
| Week 9 | Ryan ShiUniversity of Pittsburgh | AI for resource allocation |
| Week 11 | Meiyi MaVanderbilt University | Designing explainable AI |
All speakers are tentative until confirmed before their session. Talk titles and framing questions are circulated in class the week before. If a speaker cannot attend, the session is covered by the teaching assistant with related material, or rescheduled.
Fundación Capital is a social enterprise improving the financial lives of people living in poverty, working with governments and financial institutions to create economic opportunity. Goonj is an Indian NGO that turns discarded and underused urban material into a resource for rural development and disaster relief.
Assignments and grading
- Four topical assignments40%
- Invited talk reflections + writing assignment25%
- Project25%
- Participation, pop quizzes, and discussion10%
Participation (10%)
Case-study discussion, pop quizzes, and peer feedback. Quality of engagement matters more than quantity. Because so much of the intellectual work happens in the room, sustained absence will hurt here. There will be no more than eight pop quizzes, and the best five count.
Talk reflections and writing assignment (25%)
At least four structured reflections of one to two pages on invited-speaker sessions. Submit more and the best four count. Each addresses:
- What problem was the speaker solving, and why was AI the right tool?
- Who might benefit, and who might be harmed?
- What domain knowledge would be needed to deploy such a system responsibly?
- What data would you collect, and how would you keep it from violating user privacy?
- What would you change? (Arguing that you would change nothing is a valid answer.)
Reflections are due at 11:59 PM on the Friday of the week following the talk. These must be your own writing; no LLM assistance.
Alongside them, one writing assignment developed with the librarian, whose session falls in Week 5. It is due the Friday after fall break; note that this collides with the Assignment 2 deadline.
Four topical assignments (40%)
One Colab notebook per major technical topic, 10% each. Notebooks are scaffolded: you modify parameters, run code, interpret results, and write a short reflection. No prior programming experience needed.
-
Supervised learning: loan approval fairness audit
Released Week 4 · due end of Week 6
Train a classifier on loan applications and audit it across demographic groups. Compute group-level accuracy and false-positive/negative rates, shift the decision threshold, and watch the disparities move. Is this model deployable? What would you demand to see first?
-
Unsupervised learning: neighborhood clustering for health interventions
Released Week 6 · due end of Week 8
Run k-means over census-tract features (income, health outcomes, air quality, park and food access) for one US city, at several values of k. Interpret the centroids in plain English, and reflect on what it means to define “types of neighborhood” algorithmically.
-
Optimization: resource allocation under constraints
Released Week 9 · due end of Week 11
Adjust budgets, demand weights, and equity constraints in a linear program allocating scarce resources across regions. Compare solutions that optimise for coverage, for equity, and for cost-efficiency, and write about what “optimal” means once you have to say what you are optimising for.
-
LLMs: design and evaluation for a vulnerable population
Released Week 12 · due end of Week 13
Design a prompt-based tool for a social-good use case serving a vulnerable population. Submit two iterations of the system prompt with your reasoning; test on five typical and five adversarial inputs; score every response for helpfulness, harmlessness, and hallucination; then argue whether it should be deployed, and to whom.
Project (25%)
Identify a real social problem, research how AI could address it, and produce a 3–4 page policy brief. Work alone or in a pair; both are equally valid and neither is graded more strictly. A curated menu of pre-scoped problem areas comes with suggested datasets and readings; your own topic is welcome with approval. The brief covers the problem and its context with cited sources, your proposed approach and its trade-offs, fairness, privacy and security considerations, and an honest assessment of what AI can and cannot do here. A working prototype is not required, though you are welcome to build one and bring it to office hours.
- Proposal (2 pages) and team declarationWeek 7
- Mid-term lightning-talk slidesWeek 13
- Three-minute lightning talks, in classWeek 13
- Draft policy brief, for peer reviewWeek 14
- Peer review workshop, onlineWeek 14
- Final presentation slidesWeek 15
- Five-minute final presentationsWeek 15
- Final policy brief (3–4 pages)End of term
Exact deadlines are on the course management system.
One more thing.
I will work with one project team to submit a short paper to the AAAI 2027 student track, and we will sponsor your travel to Montreal to present the poster.
Readings
You do not need to buy anything. Readings are shared through the course management system, drawn from:
- The Ethical Algorithm, Kearns and Roth
- AI and Society, Mukhopadhyay and Vorobeychik (eds.)
- AI for Social Impact, Tambe, Fang, and Wilder (eds.)
- The Smart Enough City, Ben Green
- Weapons of Math Destruction, Cathy O’Neil
- Artificial Intelligence, Melanie Mitchell
- Robot Ethics, Mark Coeckelbergh
- Thinking in Bets, Annie Duke
- The Data Detective, Tim Harford
Plus articles and policy briefs through the semester. Everything is chosen to be readable without a technical background. Readings usually go out at the end of Thursday’s class for the following week, so you have the weekend with them.
Information literacy
Researching AI and its social implications means navigating peer-reviewed papers, policy briefs, journalism, preprints, and gray literature, each with different standards of evidence and different relationships to current events. Locating, evaluating, and citing sources across all of them is a core skill this course builds. A librarian from Swem Library leads a session in Week 5, timed to land before the proposal and the policy brief. You are encouraged to keep working with your Swem liaison as your project develops.
Academic integrity and LLM policy
You are encouraged to use LLMs for project ideation, coding help, and improving your writing. Three limits:
- Talk reflections must be your own writing; no LLM assistance at all.
- The final policy brief must be your own analysis. LLMs may help with structure and grammar, not substance.
- All LLM usage must be documented in a brief appendix saying what you asked for and how you changed the output.
All other written submissions are individual work. Violations are handled under William & Mary’s Honor Code.
A note on slides.
I do not post lecture slides publicly. If you are teaching a similar course and would find them useful, get in touch with a sentence about your course and I will happily share them. Enrolled students get everything through the course management system.