AI governance

Today

• Today asks how governments should make rules for AI. This topic is called AI governance.
• Congress and many states are writing new laws on AI. A project at Brown tracks these bills.
• You learn to weigh policy alternatives against each other and to recommend one.
• Three pairs each present one debate about how to govern AI.
• The class puts the three presentations side by side in one table.

Assigned readings

• CNTR AISLE, “How It Works”. It explains how the project describes each AI bill.
• One core reading selected by your pair for its presentation. Bring the citation and source. The AI Index policy chapter and the existing reading pool are references.

Example: a Maryland law on election deepfakes

• A deepfake is a fake image, video, or audio recording made with AI. It shows a real person saying or doing something they never said or did.
• In 2026 Maryland passed a law on deepfakes in elections, Senate Bill 141 (MD SB141). It does two things:
  • It prohibits creating or spreading a deepfake to produce materially false information, knowingly or with reckless disregard.
  • It requires the State Administrator of Elections to publish correct information after a credible report of election misinformation.
• Many states and Congress are writing laws like this one.

The CNTR AISLE Bill Library

• Brown’s Center for Technological Responsibility, Reimagination and Redesign (CNTR) tracks these bills. Its Bill Library listed 6,921 federal and state bills in November 2026.
• You can filter the bills by state, status, and date. Each bill has a page with a summary and a link to the full text.
• Analysts describe each bill with yes-or-no questions of the form “Does the bill …?”. The questions fall into seven areas:
  • accountability and transparency,
  • data protection,
  • bias and discrimination,
  • labor force,
  • institutions,
  • education,
  • synthetic content, such as deepfakes.
• Each bill page shows the share of questions answered “yes” in each area. For the Maryland law, the share is 32% for synthetic content and 8% for accountability and transparency. It is 0% in the other areas.
• The questions work like your Week 5 codebook. Each answer is a code assigned by an analyst.
• A share of zero does not always mean the bill has no rules in that area. The “How It Works” page lists the reasons.
• Use the Bill Library to find a concrete case for your presentation. Open the full text of the bill and check the summary against it.

Choosing a topic

• People disagree about how AI should be governed. Each pair presents one of these debates.
• We encourage you to choose your own topic. Any debate about how to govern AI works.
• A topic close to your group project is a good choice. So is a debate you follow in the news.
• The six areas below are examples. Each lists a few readings to start from.
• Open-source AI models, from Week 9, is another example.
• Your pair chooses its own core reading. You may take one listed below or find one with the agent.

Frontier-model risk

• Example readings:
  • International AI Safety Report.
  • "Backfiring Effect of Weak AI Safety Regulation," PNAS (2026).

Jurisdiction and regimes

• Example readings:
  • Bradford, Digital Empires (2023), Introduction. It compares how the United States, China and the European Union regulate technology.
  • Nelson, Science (2026).
  • Matt Sheehan, "China's AI Regulations and How They Get Made," Carnegie (2023), on how the rules are made.
  • Migliorini, "China's Interim Measures on Generative AI" (2024), on what the rules say and how they compare with the EU AI Act.
• Example case: the December 2025 executive order on preemption.

Geopolitics and chips

• Example reading: a brief on U.S. export controls on AI chips.

Power and concentration

• Example reading: Kak and West.

Information ecosystem

• Example reading: Kostka and Bogs, "The Expanding Digital Reach of the Chinese State" (2026).
• Examples:
  • Health codes repurposed against protesters.
  • Facial recognition used on Uyghurs and on schoolchildren.
• In each example, a system built for one purpose is used for another. This is called function creep.
• Rules you could compare:
  • China's 2025 labeling rules.
  • Article 50 of the EU AI Act.
  • State deepfake-election laws in the United States.

Labor and economy

• Example readings:
  • Brynjolfsson, Mitchell and Rock (2018), available in the reading reference pool.
  • Bradford, "False Choice" (2024).
  • A Draghi excerpt.
  • Jia, Jin and Wagman (2021).

Trade-offs between policy alternatives

• States have taken two approaches to AI in election campaigns:
  • Maryland bans deceptive deepfakes (MD SB141, above).
  • Louisiana requires campaigns to disclose the use of AI in telephone campaign messages (LA HB639, 2026).
• Suppose a state legislature is choosing between a ban and a disclosure requirement.
• Suppose the evidence supports the entries in this table.

• The ban and the disclosure requirement are courses of action the legislature could take. As in Week 9, each one is a policy alternative.
• Protection from deception and room for political speech are the standards used to judge them. These are the criteria.
• The ban gives up some political speech to gain protection from deception. The disclosure requirement does the reverse.
• Neither alternative is best on both criteria. To gain on one criterion, the legislature gives up something on another. This is called a trade-off.
• State a trade-off in that form: what each alternative gains, and what it gives up.
• Sometimes one alternative is at least as good as another on every criterion. Then those two have no trade-off. Drop the weaker one.
• Confronting the trade-offs is Step 6 of the eightfold path.

Policy decisions and recommendations

• After weighing the trade-offs, you choose one alternative. You also say who should adopt it.
• In the example, the state legislature has the power to adopt the ban or the disclosure requirement. The person or body with that power is the decision-maker.
• Other examples of decision-makers:
  • a governor,
  • a federal agency,
  • the European Commission.
• A recommendation has three parts:
  • the decision-maker,
  • the alternative you recommend,
  • the main reason, with its evidence.
• Name the trade-off that matters most for that choice. A decision-maker who weighs that trade-off differently would choose differently.
• The best choice depends on the decision-maker. The European Commission and a U.S. governor can weigh the same criteria differently.
• If you cannot choose, go back to an earlier step. Look for one of these:
  • more evidence,
  • a new alternative,
  • a clearer criterion.
• Deciding is Step 7 of the eightfold path.
• Bardach and Patashnik describe both steps in Part I of A Practical Guide for Policy Analysis.

Contents of a presentation

• Each pair presents one debate in AI governance.
• Students present in three pairs. Each pair speaks for 12 to 15 minutes.
• Each presentation covers six points:
  • what the dispute is about,
  • who is on each side,
  • one concrete case, with its source and date,
  • the core reading,
  • a comparison of at least two policy alternatives, with the criteria and the evidence behind each,
  • a recommendation for a named decision-maker, with the trade-off that matters most for that choice.
• Use the agent to find and organize sources.
• Check every claim in the presentation against its source.

Class comparison table

• The three presentations cover different topics. One table puts them side by side.
• After the presentations, the class builds the table together. Each pair fills one row.

Alternatives, trade-offs and decisions

• What does each alternative improve? What costs or risks does it introduce?
• Would a different criterion change the recommendation?
• How did AI help with your pair’s evidence, alternatives, or comparison?

Weekly memo · due Dec 3 before class. Write 1–2 observations from your project or the classroom case: the policy-analysis step you attempted, how AI helped or fell short, and what you would do next. Aim for 150–200 words total. Describe what you actually tried and observed in your own words. Do not use AI to write the memo. Begin in class and submit to Canvas Discussions. See memo grading.

Checkpoint 3: project website · 5%

• Put your project website online with GitHub Pages.
• See Week 8 for what the website covers.
• Add the website address to the README of your repository.
• In Week 11 a classmate reads your website and your repository.

Due Dec 3 before class.

Next class

Class resumes Dec 3 after the Nov 26 break. Week 11 is “Collaboration and replication.” You will work in a classmate’s project.