Demand Side III: Public Opinion and the Information Environment

POLISCI 158 · AI & Politics · Session 6 · Mon, Oct 12, 2026

Simone Paci

Roadmap

  1. Part I: Public opinion
    • Explain where opinions about AI come from: preferences, attitudes, beliefs, interests and values
    • Read a survey trend or model result and name one limit
  2. Part II: Events, influence and deliberation
    • Read a study’s finding and name what it does and does not measure
    • Evaluate how AI changes the channels that shape opinion and deliberation: promise and risk

Do you trust AI more or less than most people do?

  • Hands up, or write it down

Part I: Public opinion

What do Americans think about AI now?

Pew Research Center stacked bars, 2021 to 2026: share of U.S. adults more concerned than excited about AI rose from 37% to 52%

Pew · more concerned than excited

Re-plotted bars: share saying AI does more harm than good, 40% in 2023, 31% in 2024, 31% in 2025, 39% in 2026

Gallup · more harm than good

Pew Research Center, Americans have grown more concerned about AI over time. U.S. adults, June 22–28, 2026.
Bentley-Gallup, Americans Cool Toward AI, Jul 28, 2026. Gallup Panel, May 2023–May 2026.
Different questions; not one series.

Do experts and the public see AI the same way?

Pew Research Center chart: 56% of AI experts versus 17% of U.S. adults expect a positive impact of AI on the U.S. over the next 20 years

Pew Research Center, How the U.S. Public and AI Experts View Artificial Intelligence, Apr 3, 2025. Public: Aug 12–18, 2024. Experts: Aug 14–Oct 31, 2024.

Who trusts their government to regulate AI?

AI Index 2026 Figure 9.3.1: trust in own government to regulate AI by country, 2025; Singapore 81% highest, global 54%, United States 31% lowest

Stanford HAI, AI Index 2026, Fig. 9.3.1. Data: Ipsos, 2025.

Is AI becoming partisan? Start with who uses it at work

Workers who use AI at work frequently, Q1 2026

  • Democrats 30.1% · Republicans 25%

Measures use at work, not opinion

Re-plotted bars: frequent AI use at work in Q1 2026, Democrats 30.1%, Republicans 25%

Bloom & Makridis, NBER Working Paper w34813, 2026. Gallup Panel workers, ages 25–65, 2024–2026.

Things are changing?

Pew Research Center panels by party, 2021 to 2026: Democrats more concerned than excited rose from 31% to 56%, Republicans from 45% to 49%

Pew Research Center, Anderson & Bishop, Democrats are now more worried than Republicans about AI and its impact on jobs, Sep 16, 2026. U.S. adults, June 22–28, 2026. Party includes leaners.

Where does public opinion come from?

Preferences → “what you want”

  • policies, candidates

Attitudes → “how you feel”

  • favorability, salience, values

Beliefs → “what you believe”

  • expectations

Why is measuring opinion so hard?

Rational ignorance

  • No one can know everything

Stated vs. revealed preference

  • Cheap talk vs. costly action

The menu of options

  • Change the alternative, change the answer?

Re-plotted stacked bars: 57% chose do not allow data centers against allow with no AI-specific regulation; 38% against allow with AI guardrails. Same 1,007 respondents answered both, in sequence.

AI Policy Institute, toplines and crosstabs, June 10–11, 2026. 1,007 U.S. likely voters, online opt-in panels. Not a randomized split ballot.

Does self-interest stop at your paycheck?

Teaching diagram: material stakes reach past the paycheck — consumer, parent or patient, resident, taxpayer

Do values matter as much as interests?

Sociotropic, others-regarding

  • = “happy if others are happy”

Moral, symbolic

  • justice, tradition, status = “happy for something to be right”

Teaching diagram contrasting material interests with values

What do you think others think?

First-order = “what I think”

Second-order = “what I think others think”

  • social desirability · bandwagon effects · pluralistic ignorance

Teaching diagram: first-order beliefs (what I believe) and second-order beliefs (what I think others believe), which move how I act

AI is new. Where do our beliefs come from: fiction?

Frankenstein 1931 film poster

R.U.R. Theatre Guild production photograph, 1928–29: costumed robots surround a man

Metropolis 1926 set photograph by Horst von Harbou: the Maschinenmensch

Frankenstein (1931) poster; R.U.R., Theatre Guild production, 1928–29; Metropolis (1926), set photograph by Horst von Harbou. All via Wikimedia Commons.

Lessons from the last tech revolution?

Cover of The Anxious Generation by Jonathan Haidt

Jonathan Haidt, The Anxious Generation, Penguin Press, 2024. Cover shown for commentary.

Activity: what predicts AI favorability?

PPIC Statewide Survey, Nov 2024 · 2,344 California adults

Outcome: Q46 · “Do you have a favorable or unfavorable opinion of AI or artificial intelligence?”

Predictors in the model

  • Exposure: heard about AI
  • Economic circumstances: income, employment, bills, health care, housing, expectations for your kids
  • Values and politics: ideology, registered party, views on the governor, workers, inequality, the country
  • Demographics: age, gender, education, race

PPIC codebook page 11 excerpt: definition of artificial intelligence, Q45 how much have you heard or read about AI, and Q46 favorable or unfavorable opinion of AI

Source: PPIC Statewide Survey; Mark Baldassare, Survey Director. PPIC bears no responsibility for interpretations or conclusions.

Pick a variable. Which way does it point?

  1. Alone (2 min): pick one variable from the list
  2. Pairs (4 min): predict its direction; write one reason
  3. Compare with your neighbor’s prediction

Written answers welcome

Reveal: what predicts favorability?

Coefficient plot of 22 predictors of a favorable view of AI, PPIC November 2024, with 95% intervals; significant: expects kids better off +0.162, Asian American +0.158, heard about AI +0.146, female -0.086, put off doctor for cost -0.103

Associations, not causes

PPIC Statewide Survey, Nov 2024; Mark Baldassare, Survey Director. PPIC bears no responsibility for interpretations or conclusions. Analysis: instructor’s weighted linear probability model, N = 2,121.

What would make this result misleading?

Confound

  • Is something else driving both?

Wording

  • What did respondents read first?

Sample

  • Who is, and isn’t, in the data?

Do elites tell us what to think about AI?

Elite cues → mass opinion

“…a minimally burdensome national standard — not 50 discordant State ones.”

  • Executive Order 14365, Dec 11, 2025

“We need a federal moratorium on AI data centers.”

  • Sen. Bernie Sanders, Mar 25, 2026

Teaching diagram of Zaller's Receive–Accept–Sample model

White House, Executive Order 14365, “Ensuring a National Policy Framework for Artificial Intelligence,” Dec 11, 2025; Sen. Sanders, press release, Mar 25, 2026.

Which AI story gets told, and how?

Agenda setting = which issues matter?

Framing = which side of the issue matters?

CNN Lite headline: Amazon is laying off 16,000 employees as AI battle intensifies, by Jordan Valinsky, January 28, 2026

Fortune headline: If AI is roiling the job market, the data isn't showing it, Yale Budget Lab report says, raising questions of AI-washing to justify mass layoffs; by Sasha Rogelberg, February 2, 2026

CNN, Jordan Valinsky, Jan 28, 2026 (CNN Lite edition shown); Fortune, Sasha Rogelberg, Feb 2, 2026.

Part II: Events, influence and deliberation

Do AI-caused harms raise demand for regulation?

When AI causes harm, do people want more rules?

  • Harm → concern → demand for regulation?
  • Which harms, and for whom?

Do alarming headlines raise concern, or demand for rules?

Three preregistered experiments, N = 10,800

  • Treatment: headlines about existential AI risk
  • Concern rose
  • Support for AI policy did not move (Study 2)
Headline
↗
Concern ✓
↘
Policy support ✗

Hoes & Gilardi, “Existential risk narratives about AI do not distract from its immediate harms,” PNAS 122 (16), e2419055122, 2025. doi:10.1073/pnas.2419055122.

Does feeling exposed raise support for regulation?

UK panel: the same people, over time

  • More AI exposure or perceived job risk → more support for stronger regulation
  • Optimism does not offset it
  • Observational, not experimental

Exposure / perceived job risk → Regulation support

Grant, Green & Evans, “New tech, new threat? Occupational exposure to artificial intelligence increases support for AI regulation,” Journal of European Public Policy, online Aug 4, 2026. doi:10.1080/13501763.2026.2710714.

What we still don’t know

Headline → concern ✓ · policy support ✗ (experiment)

Job exposure → regulation support ✓ (panel)

A real incident → regulation support?

  • We have not found a published experiment that tests this

Who is really speaking? Alleged AI astroturfing

Civil society = the space between the state and markets

South Coast AQMD, June 2025

  • 20,000+ comments opposing clean-air rules, reportedly generated by one firm, CiviClick
  • Board rejected the rules 7–5
  • AI use alleged; the company denies it

GovTech reprint of Los Angeles Times story, February 24, 2026, by Hayley Smith, headlined AI-Generated Comments Swayed California Air Decision

Hayley Smith, Los Angeles Times, Feb 17, 2026; GovTech reprint, Feb 24, 2026. Company response: Los Angeles Times, Aug 28, 2026.

What is deliberative democracy?

Representation vs. deliberation

Public debate as a central tenet of democracy

  • Legitimacy = reasoning > votes

America in One Room logo, Stanford Deliberative Democracy Lab

America in One Room logo © Stanford University, Deliberative Democracy Lab.

Can AI help deliberation?

The Habermas Machine

  • AI drafts group statements from participants’ views
  • Preferred 56% of the time vs. 44% for human mediators
  • “Mediators” = randomly assigned participants, not professionals

Re-plotted bars: AI-drafted statements preferred 56% of the time versus 44% for human mediators

Tessler et al., “AI can help humans find common ground in democratic deliberation,” Science 386 (6719), eadq2852, Oct 18, 2024.

Could AI hurt deliberation?

Technological solutionism = “societal problems can be ‘solved’ through technology”

  • For deliberation: technology as an “efficient” fix for the “problems” of minipublics
  • Case: the Habermas Machine

First page of Oleart and Palomo Hernández, Why AI Technosolutionism Harms Democracy and Deliberation, Journal of Deliberative Democracy, 2025: title, authors and abstract

Oleart & Palomo Hernández, “Why AI Technosolutionism Harms Democracy and Deliberation: The EU, Citizens’ Assemblies and the Habermas Machine,” Journal of Deliberative Democracy 21 (1): 1–6, 2025. CC BY 4.0.

Debate: does AI-assisted deliberation strengthen or weaken legitimacy?

Pairs, assigned sides

  • Side A: it strengthens legitimacy
  • Side B: it weakens legitimacy

2 minutes to build your best case · write or speak

Put it together: from beliefs to pressure to deliberation

Beliefs

Where do views come from?

→

Pressure

How do they reach government?

→

Deliberation

Can AI change how we decide together?

Exit card (optional): one testable hypothesis to sharpen your project

See you next class

Backup slides

Can litigation change AI rules?

Strategic litigation

  • Bartz v. Anthropic · final approval of a $1.5 billion class settlement, Jul 20, 2026
  • Authors Guild v. OpenAI · no ruling yet

Authors Guild, Jul 21, 2026; Authors Guild, Sep 21, 2026.

Does AI search divert traffic from knowledge institutions?

Pew Research Center chart: users clicked a link 8% of the time when an AI summary appeared versus 15% without one; 1% clicked a link inside the summary

Pew Research Center, Chapekis & Lieb, Jul 22, 2025. 68,879 searches by 900 adults, March 2025.