Microcredential komex Computational Social Science with Artificial Intelligence

A hands-on, five-day in-person course on doing social science with and about AI — designing experiments on language models, auditing their bias and behavior, simulating AI societies, and building reproducible pipelines for LLM, agent, and audit studies.

What Is This Course About?

AI systems are quickly becoming both a part of the social world we study and a tool for studying it, and the next generation of social scientists will need to do both. This hands-on, five-day in-person course teaches you to do social science with and about AI — using language models as experimental subjects, agents, and simulated populations, and auditing their behavior, bias, and social impact. Across five days you will move from persona-conditioned experiments and bias measurement to multi-agent simulations, collective decision-making, and the economic and democratic consequences of AI. You will leave able to design and run your own reproducible study on AI systems, whatever your substantive field.

Course repository (living syllabus, updated before the course):
https://github.com/carinahausladen/komex-computational-social-science-ai

Learning Goals

After this course you will be able to:
- Design and run experiments that treat AI models as study subjects, and
 judge when a language model can stand in for a human participant and
 when it cannot.
- Measure bias and behavior in AI systems with established
 social-science tools — embeddings, WEAT-style probes, intersectional
 regression — and say what those numbers do and do not show.
- Build multi-agent simulations, including LLM agents acting on real
 social networks, and check them against real-world data rather than
 trusting them at face value.
- Use social-choice and collective-decision theory to think about how AI
 systems aggregate people's preferences, and what alignment means.
- Set up a reproducible pipeline — code, prompts, model versions,
 results — so that your AI study can be rerun by others and by you.

Recommended Readings for the Course

- Hausladen, C. I., Knott, M., Camerer, C. F., & Perona, P. (2025). Social Perception of Faces in a Vision-Language Model. Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (FAccT), 639–659. arXiv:2408.14435. [instructor's own work]

- Hausladen, C. I., Schubert, M. H., & Engel, C. (2026). Identifying Latent Intentions via Inverse Reinforcement Learning in Repeated Linear Public Good Games. arXiv:2601.08803. [instructor's own work]

- Park, J. S., O'Brien, J. C., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023). Generative Agents: Interactive Simulacra of Human Behavior. Proceedings of UIST 2023. arXiv:2304.03442.

Assignments for the Course

- Daily hands-on lab exercises in the session (coding, not graded).

- Short daily homework: a mini-audit, a small coding exercise, or a quiz, plus the assigned reading for the next day.

- Final take-home project: participants design and prototype a small AI-based study relevant to their own research.

- The final project is submitted for pass/fail feedback.

Schedule

Standardized KOMEX in-person teaching times, the same each day (4

contact hours):

- 09:00–10:30  Course

- 10:30–11:00  Break

- 11:00–12:30  Course

- 12:30–13:30  Lunch break

- 13:30–14:30  Course

- Afterwards: daily homework and reading in preparation for the next

  day.

Indicative topic and hands-on lab per day:

- Day 1 — AI as experimental subjects. Homo silicus, persona

  conditioning, prompt-based experimental design, behavioral

  calibration; re-running classic behavioral-economics paradigms on LLMs

  and the validity questions this raises. Lab: persona-conditioned LLM

  in a social dilemma (structure vs. framing).

- Day 2 — Measuring behavior and bias. From decades of social-science

  bias measurement to LLM/VLM audits: embeddings, WEAT-style probes,

  masked-token and pseudo-log-likelihood metrics, generated-text

  measures, intersectional regression. Lab: bias-metric toolkit plus a

  vision-language face-perception audit.

- Day 3 — AI agents and societies on a network. Cognitive architecture

  (memory, planning, reflection), generative agents, and populating a

  real social network with LLM agents; emergent conventions, collective

  bias, and convergence diagnostics. Lab: an LLM agent-based model of

  microfinance diffusion over real Indian village networks

  (Banerjee–Chandrasekhar–Duflo–Jackson), validated against who actually

  joined; plus naming-game emergence.

- Day 4 — Cooperation, learning, and institution design. Reinforcement

  learning in social games (prisoner's dilemma, Werewolf, Diplomacy), RL

  as mechanism/institution designer, and social choice for AI alignment

  and preference aggregation. Lab: Q-learning in a social game plus

  preference-aggregation rules.

- Day 5 — Social impact, democracy, and your own study. Economic

  exposure of AI, political audits and bias in deployment, augmented

  democracy and deliberation, agentic misalignment; building

  reproducible pipelines. Lab: community-notes / political-audit plus

  capstone project clinic.

Who Is Your Instructor?
LinkedIn: https://www.linkedin.com/in/carina-hausladen/