Intelligent Agents: Computational Game Solving
CSCE 631 — Summer 2026 — Texas A&M University
Online Asynchronous • Section 700 • 3 Credit Hours • May 26 – June 29
A rigorous exposition of game theory—Nash equilibria, regret minimization, CFR, and game abstraction—compressed into five intensive weeks, with a final module connecting the theory to modern autonomous agent design.
5-Week Intensive
Online Async
2 Programming Assignments
Research Project
No Exams
Why This Course
Game theory is the mathematical foundation for reasoning about strategic interaction. This course covers the core theory—from Nash equilibria through CFR and game abstraction—and connects it to the design of autonomous agents.
Format
Intensive & Flexible
Five weeks, fully online asynchronous. Pre-recorded lectures, weekly deadlines, and active discussion boards. Designed for graduate students with summer schedules.
Assessment
Reproduce, Don’t Memorize
No exams. Two programming assignments grounded in published work, plus an individual research project worth 50%. Understanding is the bottleneck, not implementation.
Depth
Theory With Applications
Four weeks of classical game theory, one week of modern applications. The final week connects game-theoretic foundations to the design and evaluation of autonomous agents.
Research
Course → Lab
Strong projects can become research collaborations. The instructor’s group works on submodular optimization, game-theoretic monitoring, and action abstraction.
Weekly Schedule
Five weeks, roughly two modules of classical game theory per week, with a final module on autonomous agents and game theory.
| Week |
Dates |
Topics |
Content |
| 1 |
May 26–30 |
Foundations of Game Theory |
Normal-form games, Nash equilibrium, mixed strategies, maxmin, correlated equilibrium, dominated strategies. Notes |
| 2 |
Jun 2–6 |
Computing Equilibria & Regret Minimization |
Support enumeration, LP for zero-sum, external/internal/swap regret, MWU, RM, RM+, no-regret → CCE convergence. Notes |
| 3 |
Jun 9–13 |
Extensive-Form Games & CFR |
Game trees, information sets, subgame perfection, sequence form. Vanilla CFR, CFR+, Monte Carlo CFR. Notes |
| 4 |
Jun 16–20 |
Game Abstraction & Poker |
Information/action abstraction, safe subgame solving, Libratus & Pluribus. Notes |
| 5 |
Jun 23–29 |
Advanced Topics & Autonomous Agents New |
Deep CFR, agent architectures (ReAct, ToT, LATS), multi-agent debate as a game, red-teaming as zero-sum. Notes |
The narrative arc: Define games → Solve games → Learn in games → Handle sequential games at scale → Apply the theory to autonomous agents.
Course Materials
Lecture slides, module notes, and programming assignments are linked below. Video recordings are hosted separately.
General
Module Notes
- Module 1 — Foundations of Game Theory (Lectures 1–4)
- Module 2 — Computing Equilibria & Regret Minimization (Lectures 5–9)
- Module 3 — Extensive-Form Games & CFR (Lectures 10–14)
- Module 4 — Game Abstraction & Poker Case Studies (Lectures 15–17)
- Module 5 — Advanced Topics & Autonomous Agents (Lectures 18, 5.1–5.3)
Programming Assignments
- PA1 — Normal-Form Games: Utilities, Best Responses, and Equilibria
- PA2 — Multi-Agent Debate with LLM Agents via the TAMU API
- Course Project — Topic list, rubric, and submission guidelines
Programming Assignments
Two PAs grounded in specific published work. The implementation is the ticket to entry; the analysis is the assessment.
Normal-Form Games: Utilities, Best Responses, and Equilibria
Support enumeration & Nash equilibrium verification
Implement core algorithms for 2-player normal-form games: expected utility, best response, Nash equilibrium verification, strictly dominated strategy elimination, and correlated equilibrium verification. Write a methods note reflecting on your design choices. Extra credit: compute a CE via linear programming.
Download notebook.
Multi-Agent Debate with LLMs
Du et al., “Improving Factuality and Reasoning in Language Models through Multiagent Debate” (2023)
Using the TAMU API, implement a multi-agent debate protocol. Run experiments varying agent count, round count, and model choice. Analyze through a game-theoretic lens: does debate converge, what are the failure modes, and how do information sets and equilibrium concepts help interpret behavior? $5/day budget.
Download notebook.
Course Project
Apply a game-theoretic concept from this course to a multi-agent or human-agent interaction problem. Individual project, 50% of final grade. See the full project description for the 10 curated topics and rubric.
Example Topics
- Build and evaluate a multi-agent debate system using regret minimization.
- Implement a monitoring protocol for a coding agent and analyze its game-theoretic properties.
- Design an auction mechanism for AI agents and measure equilibrium behavior.
- Red-team an AI agent and formalize the attack as an exploitation strategy.
- Compare CFR self-play to debate-style self-play on a structured task.
Deliverables
| Proposal (1 page) | Jun 6 (Wk 2) |
| Code + Report (4 pages) | Jun 29 (Wk 5) |
Reading List
Core papers and textbook chapters organized by week. Papers are provided as PDFs on Canvas.
Week 1: NF Games + Equilibria
- Shoham & Leyton-Brown, Multiagent Systems, Ch. 3–4
- Guo et al., “Game Theory Meets Large Language Models: A Systematic Survey” (IJCAI 2025) Framing
Week 2: Regret Minimization
- Cesa-Bianchi & Lugosi, Prediction, Learning, and Games, Ch. 4
Week 3: Extensive-Form Games + CFR
- Shoham & Leyton-Brown, Ch. 5
- Zinkevich et al., “Regret Minimization in Games with Incomplete Information” (NeurIPS 2007)
- Lanctot et al., “Monte Carlo Sampling for Regret Minimization” (NeurIPS 2009)
Week 4: Abstraction + Poker
- Sandholm, “Abstraction for Solving Large Incomplete-Information Games” (AAAI 2015)
- Brown & Sandholm, “Superhuman AI for Multiplayer Poker” (Science, 2019)
Week 5: Autonomous Agents + Game Theory
- Yao et al., “ReAct: Synergizing Reasoning and Acting in Language Models” (ICLR 2023) PA2
- Du et al., “Multiagent Debate” (2023) PA2
- Anthropic, “Building Effective Agents” (blog, Dec 2024)
- Greenblatt et al., “AI Control: Improving Safety Despite Intentional Subversion” (ICML 2024)
- Shinn et al., “Reflexion” (NeurIPS 2023)
Logistics
Format
Online asynchronous. Pre-recorded lecture videos released at the start of each week. Weekly deadlines for assignments and discussion posts. Office hours via Zoom.
Grading
| Component | Weight |
|---|
| PA1 (Normal-Form Games) | 20% |
| PA2 (Multi-Agent Debate) | 30% |
| Course Project | 50% |
No midterm. No final exam.
Late policy: 3 slip days total, usable in 24-hour increments (max 2 per assignment). Assignments submitted after slip days are exhausted will not be accepted.
All Deadlines
| Item | Due |
|---|
| PA1 (Normal-Form Games) | Fri Jun 6 |
| Project Proposal (1 page) | Fri Jun 6 |
| PA2 (Multi-Agent Debate) | Sat Jun 27 |
| Project Report + Code | Mon Jun 29 |
Prerequisites
CSCE 420 or CSCE 625, or instructor approval. Mathematical maturity (proofs, probability, optimization) expected.
Resources
- Syllabus: PDF
- Textbook: Shoham & Leyton-Brown, Multiagent Systems (free online)
- Supplementary: Cesa-Bianchi & Lugosi, Prediction, Learning, and Games
- API: TAMU API ($5/day; setup guide provided with PA2)
- Computing: Python 3.x, Jupyter. No GPU required.
AI Tool Policy
You may use AI tools for writing code. All written analysis must be your own work. The analysis is the assessment—outsourcing it defeats the purpose.
Instructor
Alan Kuhnle
Assistant Professor, Computer Science & Engineering
Email: kuhnle at tamu.edu
Office Hours: By appointment via Zoom
Zoom: tamu.zoom.us/my/kuhnle
Research: submodular optimization, algorithmic game theory, autonomous agent design.