CMSC 848Y / Fall 2026

AI, Public Policy, and Society

Integrating AI with society with effective policy.

This course is for AI Ph.D. and M.S. students who want to understand how their research relates to public policy. The goal is for students to become policy-literate AI researchers who can advise decision makers.

Most of the course will focus on AI. We will also study other engineering fields which have had massive policy impacts to see how others have contended with national security, scientific self-governance, public participation, and international regulation.

The course will alternate between seminar discussions, policy exercises, writing workshops, and guest lectures. Students will use ideas from political science, international relations, and public policy to analyze technical research and its consequences.

Assignments.

Students will write four response papers during the semester. Papers may address any course topic and do not need to build on one another. We will discuss possible paper topics during class; no separate topic proposal is due.

Seminar preparation and participation

Due: Ongoing
Weight: 20%

Students are expected to complete the readings and participate in seminar discussions and policy exercises. When reading technical papers, students should be prepared to discuss five questions:

What was technically novel?
What capability, cost, or bottleneck changed?
What assumptions and evaluation conditions limit the result?
What changes when the method is deployed outside the paper?
Which policy claims follow from the paper, and which do not?

When reading policy papers, students should be prepared to discuss five corresponding questions:

How is the policy problem being defined?
What causal mechanism does the author claim?
Who has authority to act?
What instrument is proposed, and what constrains implementation?
Who benefits, who bears the costs, and what evidence could change the conclusion?

Response papers

Due: September 16, October 14, November 4, and December 9, 2026
Weight: 80%

Students will submit four response papers, each worth an equal share of the response-paper grade. Each paper may address any topic raised by the course. Students may focus on one reading, connect several readings from the same week, or develop an argument that draws on material from across the course. Response papers should go beyond summarizing the readings. A paper might critically assess an author's argument, explain how two or more readings relate to one another, develop an original causal claim, identify an important unanswered question, propose a way to study that question, or offer a policy recommendation. Each response paper should be no longer than four pages, approximately 1,000-1,200 words, double-spaced with 12-point font and 1-inch margins. Papers are due by midnight on the assigned class day.

Syllabus

August 31 and September 2, 2026

Unit 1: Becoming dual-fluent in technology and policy

From technical result to policy problem

Readings

P
Deborah A. StoneCausal Stories and the Formation of Policy AgendasPolitical Science Quarterly 104, no. 2 (1989)
T
Tom B. Brown et al.Language Models Are Few-Shot LearnersNeurIPS (2020), selected sections
B
Emily M. Bender et al.On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?FAccT (2021)

Slides

Course Intro Policy Basics

September 9, 2026

Unit 2: National security and the geopolitics of AI

Guest Lecture - Andrew Reddie, Research Assistant Professor, University of California, Berkeley

CalendarNo class September 7 (Labor Day).

Readings

T
David Silver et al.Mastering the Game of Go with Deep Neural Networks and Tree SearchNature 529 (2016)
P
Robert JervisCooperation Under the Security DilemmaWorld Politics 30, no. 2 (1978), selected sections
P
Michael C. HorowitzArtificial Intelligence, International Competition, and the Balance of PowerTexas National Security Review 1, no. 3 (2018)

Slides

Andrew's Slides (accessible to UMD logins only)

September 14 and 16, 2026

Unit 1: Becoming dual-fluent in technology and policy

Recombinant DNA and scientific self-governance

Readings

T
Stanley N. Cohen et al.Construction of Biologically Functional Bacterial Plasmids In VitroPNAS 70, no. 11 (1973)
B
Paul Berg et al.Potential Biohazards of Recombinant DNA MoleculesScience 185, no. 4148 (1974)
B
Paul Berg et al.Asilomar Conference on Recombinant DNA MoleculesScience 188, no. 4192 (1975)
P
J. Benjamin HurlbutTaking Responsibility: Asilomar and Its LegacyScience (2025)

AssignmentResponse Paper 1 due September 16

Slides

TBA

September 21 and 23, 2026

Unit 1: Becoming dual-fluent in technology and policy

Guest Lectures - Maria Molina, Assistant Professor, Atmospheric and Oceanic Science; Marissa Ramirez de Chanlatte, Principal Investor, Playground Global

Readings

T
Mario J. Molina and F. Sherwood RowlandStratospheric Sink for Chlorofluoromethanes: Chlorine Atom-Catalysed Destruction of OzoneNature 249 (1974)
T
J. C. Farman, B. G. Gardiner, and J. D. ShanklinLarge Losses of Total Ozone in Antarctica Reveal Seasonal ClOx/NOx InteractionNature 315 (1985)
P
Cass R. SunsteinOf Montreal and Kyoto: A Tale of Two ProtocolsHarvard Environmental Law Review 31 (2007), selected sections

Slides

TBA

September 28 and 30, 2026

Unit 1: Becoming dual-fluent in technology and policy

Comparing governance pathways: measurement, benchmarks, and risk

Readings

T
Rylan Schaeffer, Brando Miranda, and Sanmi KoyejoAre Emergent Abilities of Large Language Models a Mirage?NeurIPS (2023)
T
Inioluwa Deborah Raji et al.AI and the Everything in the Whole Wide World BenchmarkNeurIPS Datasets and Benchmarks (2021)
P
National Research CouncilUnderstanding Risk: Informing Decisions in a Democratic SocietyNational Academies Press (1996), summary and Chapter 1
S
National Institute of Standards and TechnologyAI Risk Management FrameworkCurrent version, selected Govern and Map sections

AssignmentComparative exercise in class

Slides

TBA

October 5 and 7, 2026

Unit 2: National security and the geopolitics of AI

Machine vision, targeting, and proliferation

Readings

T
Joseph Redmon et al.You Only Look Once: Unified, Real-Time Object DetectionCVPR (2016)
B
Ritwik Gupta et al.Open-Source Assessments of AI Capabilities: The Proliferation of AI Analysis Tools, Replicating Competitor Models, and the Zhousidun DatasetBerkeley Risk and Security Lab research preprint (2024)
P
Michael C. HorowitzWhen Speed Kills: Lethal Autonomous Weapon Systems, Deterrence and StabilityJournal of Strategic Studies 42, no. 6 (2019)

AssignmentResponse Paper 2 due October 14

Slides

TBA

October 14, 2026

Unit 1: Becoming dual-fluent in technology and policy

Guest Lecture - Brandon Cortino, Senior Associate, Institute for Security and Technology on AI-NC3

CalendarNo class October 12 (Fall Break).

Readings

T
Lisanne BainbridgeIronies of AutomationAutomatica 19, no. 6 (1983)
T
Gagan Bansal et al.Does the Whole Exceed Its Parts? The Effect of AI Explanations on Complementary Team PerformanceCHI (2021)
B
Edward Geist and Andrew J. LohnHow Might Artificial Intelligence Affect the Risk of Nuclear War?RAND Corporation (2018)
P
James M. ActonEscalation through Entanglement: How the Vulnerability of Command-and-Control Systems Raises the Risks of an Inadvertent Nuclear WarInternational Security 43, no. 1 (2018)

Slides

TBA

October 19 and 21, 2026

Unit 2: National security and the geopolitics of AI

Nuclear basics - to be changed

Readings

T
Lise Meitner and O. R. FrischDisintegration of Uranium by Neutrons: A New Type of Nuclear ReactionNature 143 (1939)
B
Otto Frisch and Rudolf PeierlsMemorandum on the Properties of a Radioactive Super-bombFrisch-Peierls memorandum (1940)
P
Scott D. SaganWhy Do States Build Nuclear Weapons? Three Models in Search of a BombInternational Security 21, no. 3 (1996/97)

AssignmentPaper topics discussed in class

Slides

TBA

October 26 and 28, 2026

Unit 2: National security and the geopolitics of AI

Compute as a governance proxy

Readings

T
Jordan Hoffmann et al.Training Compute-Optimal Large Language ModelsNeurIPS (2022)
B
Ben BuchananThe AI Triad and What It Means for National Security StrategyCenter for Security and Emerging Technology (2020)
B
Lennart Heim et al.Computing Power and the Governance of Artificial IntelligenceResearch paper (2024), selected sections
P
Ritwik Gupta and Andrew W. ReddieThe LLM Mirage: Economic Interests and the Subversion of Weaponization ControlsACM (2026)

Slides

TBA

November 2 and 4, 2026

Unit 2: National security and the geopolitics of AI

Compute governance: chokepoints, efficiency, and circumvention

Readings

T
Samyam Rajbhandari, Jeff Rasley, Olatunji Ruwase, and Yuxiong HeZeRO: Memory Optimizations Toward Training Trillion Parameter ModelsSC (2020)
B
Lennart Heim et al.Governing Through the Cloud: The Intermediary Role of Compute Providers in AI RegulationOxford Martin School (2024)
P
Henry Farrell and Abraham L. NewmanWeaponized InterdependenceInternational Security 44, no. 1 (2019)
B
Ritwik Gupta and Andrew W. ReddieAccelerating the Evolution of AI Export ControlsTech Policy Press (2023)

AssignmentResponse Paper 3 due November 4

Slides

TBA

November 9 and 11, 2026

Unit 3: Domestic governance, climate, and impacts

Civil rights, discrimination, and auditing + guest lecture by Gustavo Flores-Macias, Dean of the UMD School of Public Policy

Readings

T
Moritz Hardt, Eric Price, and Nathan SrebroEquality of Opportunity in Supervised LearningNeurIPS (2016)
T
Joy Buolamwini and Timnit GebruGender ShadesFAT* (2018)
P
Solon Barocas and Andrew D. SelbstBig Data's Disparate ImpactCalifornia Law Review 104 (2016), selected sections

Slides

TBA

November 16 and 18, 2026

Unit 3: Domestic governance, climate, and impacts

Automated decisions, public administration, and due process

Readings

T
Alexandra Chouldechova, Diana Benavides-Prado, Oleksandr Fialko, and Rhema VaithianathanA Case Study of Algorithm-Assisted Decision Making in Child Maltreatment Hotline Screening DecisionsFAT* (2018)
P
Danielle Keats CitronTechnological Due ProcessWashington University Law Review 85, no. 6 (2008), selected sections
S
U.S. Government Accountability OfficeArtificial Intelligence: An Accountability Framework for Federal Agencies and Other EntitiesGAO-21-519SP (2021), selected sections

Slides

TBA

November 23, 2026

Unit 3: Domestic governance, climate, and impacts

Labor, markets, and productivity

CalendarNo class November 25 (Thanksgiving Recess).

Readings

T
Long Ouyang et al.Training Language Models to Follow Instructions with Human FeedbackNeurIPS (2022)
P
David H. AutorWhy Are There Still So Many Jobs? The History and Future of Workplace AutomationJournal of Economic Perspectives 29, no. 3 (2015)
B
Shakked Noy and Whitney ZhangExperimental Evidence on the Productivity Effects of Generative Artificial IntelligenceScience 381, no. 6654 (2023)

Slides

TBA

November 30 and December 2, 2026

Unit 3: Domestic governance, climate, and impacts

Datacenters, energy, and environmental effects

Readings

T
Emma Strubell, Ananya Ganesh, and Andrew McCallumEnergy and Policy Considerations for Deep Learning in NLPACL (2019)
T
Alexandra Sasha Luccioni, Yacine Jernite, and Emma StrubellPower Hungry Processing: Watts Driving the Cost of AI Deployment?FAccT (2024)
P
International Energy AgencyKey Questions on Energy and AIIEA (2026), executive summary and selected sections
P
Eugénie DugouaInduced Innovation and International Environmental Agreements: Evidence from the Ozone RegimeReview of Economics and Statistics 107, no. 6 (2025)

Slides

TBA

December 7 and 9, 2026

Unit 3: Domestic governance, climate, and impacts

Synthetic media, evidence, and democratic legitimacy

Readings

T
Ye Jia et al.Transfer Learning from Speaker Verification to Multispeaker Text-to-Speech SynthesisNeurIPS (2018)
P
Robert Chesney and Danielle CitronDeep Fakes: A Looming Challenge for Privacy, Democracy, and National SecurityCalifornia Law Review 107 (2019), selected sections
B
Sarah Barrington, Emily A. Cooper, and Hany FaridPeople Are Poorly Equipped to Detect AI-Powered Voice ClonesScientific Reports 15 (2025)

AssignmentResponse Paper 4 due December 9

Slides

TBA