Demand Side III: Public Opinion and the Information Environment

Session 6 · Mon, Oct 12

POLISCI 158 · AI & Politics · Stanford · Fall 2026

Roadmap

In this session
Questions we’ll tackle
Before class
After class

In this session

Americans are wary of AI. They are less trusting of their government to regulate it than people almost anywhere else, and they are far less optimistic than AI experts. We ask where these views come from: material interests beyond your job, values, beliefs about what others think, stories and the last tech revolution, and cues from political elites. In a short activity, you predict what explains Californians’ views of AI, then compare with real survey results. We then ask how opinion becomes pressure. Do AI harms raise demand for regulation? What happens when public comment can be manufactured? Can AI help or hurt democratic deliberation? Next Monday turns to persuasion, microtargeting and synthetic media.

Questions we’ll tackle

What do Americans think about AI now, and how do partisan patterns appear to be shifting?
Where does public opinion come from: preferences, attitudes or beliefs? How much is material interest, and how much is values, beliefs about others and elite cues?
What can a survey result tell us, and what would make it misleading?
Do AI-caused harms raise demand for regulation, and what evidence do we have so far?
Can AI strengthen democratic deliberation, or does it weaken the channels through which citizens are heard?

Before class

Read

Priority: The Politics of AI, Nicholas Bloom and Christos Makridis (2026), NBER Working Paper 34813. Read for whether the partisan gap in workplace AI use reflects ideology or where people work, and which gap does not go away. Note that it measures workers, not the general public.
Public Opinion, AI Index Report 2026, Stanford Institute for Human-Centered AI (2026). Survey data on AI optimism, nervousness and trust in government to regulate AI across countries. Note where the United States stands.

Listen or watch

Deliberative Polling: Changing the Tides of Democracy, Alice Siu, TEDxStanford (2023). An introduction to deliberative democracy from Stanford’s Deliberative Democracy Lab. Keep it in mind when we ask what AI adds to deliberation.

Explore

AI Incident Database, Responsible AI Collaborative. Read Incident 825, an AI-generated local news case, then browse a few other incidents. Which harms seem most likely to change public opinion?
AI Lawsuit Tracker (2026). A running list of lawsuits against AI companies. Find one case brought by creators or a civil society group.

After class

Key takeaways

Americans lean wary. In Pew’s 2026 survey, 52% were more concerned than excited about AI in daily life. In a separate Gallup question, 39% said AI does more harm than good. Only 31% of Americans trust their government to regulate AI, the lowest share in the AI Index comparison. AI experts are far more positive than the public: 56% versus 17% expect a positive impact on the U.S. over 20 years.
Partisan patterns are moving. Bloom and Makridis find Democrats report more frequent AI use at work than Republicans (30.1% versus 25% in early 2026), but that gap mostly reflects education and occupation. Perceived risk of losing one’s job is different: 19.7% of Democratic workers versus 15.5% of Republican workers in early 2026. In Pew’s data, Democrats are now more concerned about AI than Republicans (56% versus 49%), a reversal from 2021.
Opinion has several building blocks. Preferences are what you want, attitudes are how you feel, and beliefs are what you expect. Interests reach beyond your job, and values and second-order beliefs (what you think others think) matter too. Measuring opinion is hard: the options a survey offers can change the answer, and only a randomized design shows by how much.
Associations, not causes. In the PPIC November 2024 California survey, having heard more about AI and expecting one’s children to be better off were associated with a more favorable view of AI. Registered party was not, once the other factors were included.
Do harms raise demand for rules? The evidence is still thin.
- In Hoes and Gilardi’s experiments, existential-risk headlines raised concern but did not move support for AI policy.
- Grant, Green and Evans find that British workers exposed to AI support stronger regulation.
- We have not found a published experiment linking a real AI incident to support for regulation.
AI changes the channels between citizens and government.
- Public comment can be manufactured: an alleged AI-generated campaign that the company denies.
- In the Habermas Machine experiment, participants preferred AI-drafted group statements 56% of the time. The comparison “mediators” were randomly assigned participants, not professionals.
- Critics warn that treating deliberation’s problems as technical ones depoliticizes it.

Go further

Democrats Are Now More Worried Than Republicans About AI and Its Impact on Jobs, Anderson and Bishop, Pew Research Center (2026). The short read behind the partisan-reversal chart.
AI Can Help Humans Find Common Ground in Democratic Deliberation, Michael Henry Tessler, Michiel A. Bakker, Daniel Jarrett, Hannah Sheahan, et al. (2024), Science. The “Habermas Machine” experiment: an AI drafts group statements. Who were the human mediators it was compared with?
Why AI Technosolutionism Harms Democracy and Deliberation: The EU, Citizens’ Assemblies and the Habermas Machine, Oleart and Palomo Hernández (2025), Journal of Deliberative Democracy. A short, open-access critique of the Habermas Machine; read it alongside Tessler et al.
New Tech, New Threat? Occupational Exposure to Artificial Intelligence Increases Support for AI Regulation, Grant, Green and Evans (2026), Journal of European Public Policy. A British panel study of AI exposure and support for regulation.
How the U.S. Public and AI Experts View Artificial Intelligence, Colleen McClain, Brian Kennedy, Jeffrey Gottfried, Monica Anderson, and Giancarlo Pasquini, Pew Research Center (2025). The source of the public-versus-experts chart.
The Anxious Generation, Jonathan Haidt (2024), Penguin Press. The social media era as a reference point for how people may judge AI.
How the Chinese Government Fabricates Social Media Posts for Strategic Distraction, Not Engaged Argument, Gary King, Jennifer Pan, and Margaret E. Roberts (2017), American Political Science Review. The “50 cent party” study, a pre-AI benchmark for fabricated campaigns.
AI-Generated Comments Swayed California Air Decision, Hayley Smith, Los Angeles Times (2026), reprinted by GovTech. Reporting on public comments sent to Southern California air regulators. The use of AI is alleged, and the company maintains that AI was not used in this campaign.
Algorithmic Monocultures in Hiring, Rishi Bommasani, Sarah H. Bana, Kathleen A. Creel, et al. (2026), open version of their paper at the ACM Conference on Fairness, Accountability, and Transparency (FAccT). An independent audit of one hiring-algorithm vendor.
Authors Go to Washington: The Authors Guild Takes the AI Copyright Fight to Capitol Hill, Authors Guild (2026). A creators’ organization lobbying Congress alongside its litigation.
Impact of AI Search Summaries on Website Traffic: Evidence from Google AI Overviews and Wikipedia, Mehrzad Khosravi and Hema Yoganarasimhan (2026), working paper. How Google’s AI Overviews changed traffic to English Wikipedia compared with other language editions.
Google Users Are Less Likely to Click on Links When an AI Summary Appears in the Results, Athena Chapekis and Anna Lieb, Pew Research Center (2025). Users clicked a search result in 8% of visits with an AI summary versus 15% without one.

About this deck

Adapted from POLISCI linked public sources.

Curated by Simone Paci, Stanford University. Drafted with AI assistance (Claude) from the instructor’s course materials and reviewed by the instructor before publication; every external link is checked before each update.