Views of Risks, Opportunities and Regulation of AI

Pew Research Center's comprehensive April 2025 survey comparing AI experts' and the general public's views on AI risks, opportunities, regulation, and personal impact.

Overview

Major comparative survey examining where AI experts and the American public diverge and align on AI's challenges and opportunities. Key findings: experts are far more enthusiastic while the public is increasingly concerned, but both groups worry about insufficient regulation and lack confidence in government and industry to address AI responsibly.

Survey timing: Summer and fall 2024 (before change in presidential administration)

Key Findings

Concern vs. Excitement: The Great Divide

AI experts:

47% more excited than concerned
15% more concerned than excited
38% equally concerned and excited

U.S. public:

11% more excited than concerned
51% more concerned than excited
38% equally concerned and excited

Trend: Public concern increasing

2021-2022: ~40% more concerned than excited
2023: ~50% more concerned than excited (current)

Gender Gaps on Excitement

Among AI experts (wider gap):

Men: 53% more excited than concerned
Women: 30% more excited than concerned
Men: 11% more concerned than excited
Women: 24% more concerned than excited

Among U.S. public:

Men: 15% more excited than concerned
Women: 7% more excited than concerned
Men: 46% more concerned than excited
Women: 55% more concerned than excited

Pattern: Men consistently more excited in both groups, but gap is wider among experts.

Specific Concerns

Top Concerns (Expert vs. Public)

Inaccurate information:

Experts: 70% extremely/very concerned
Public: 66% extremely/very concerned
Closest alignment

Job elimination:

Experts: 25% extremely/very concerned
Public: 56% extremely/very concerned
Biggest divergence (31 percentage points)

Loss of human connection:

Experts: 37% extremely/very concerned
Public: 57% extremely/very concerned

AI bias:

Experts: 55% extremely/very concerned
Public: 55% extremely/very concerned
Perfect alignment

Data misuse:

Experts: 60% extremely/very concerned
Public: Majority extremely/very concerned

AI impersonation (deepfakes):

Experts: ~67% extremely/very concerned
Public: Majority extremely/very concerned

Gender Differences on Specific Concerns

Biggest gaps (among experts):

Data misuse
AI bias
Inaccurate information

Minimal gender differences:

Impersonation
Job loss

Loss of human connection (both groups):

Among experts: Women 45% vs. Men 35% highly worried
Among public: Women 63% vs. Men 52% highly worried

Expert Quotes

Reasons for Excitement

"I think broadly some of the things that excite me are things like applications that can save people a lot of time from repetitive and mundane tasks. So I think automating some of those workflows."

"I've seen that the AI can improve a lot the accuracy of the diagnosis of different diseases. Also, it can boost the development of different medicines for different treatments. Like for instance, for breast cancer classification, it can improve a lot. It can decrease the false positive rates and false negative rates. Most excited about the positive impact that it could have in the health industry."

Reasons for Concern

"I do think about how that [airport biometrics] technology is used, especially from a privacy and security standpoint. … Where's that data going? How is it being housed? Where is it being used for? Where is my consent? Can I really, truly say no, I don't want my picture taken, but what is the consequence of me saying that and still trying to make it to my flight at home?"

"Misinformation has always been an issue with technology. … But I think the main issue with AI and misinformation is that you can now do misinformation at scale, at a way larger scale."

Personal Impact Perception

Experts (optimistic):

76% say AI more likely to benefit them
15% say more likely to harm them
9% unsure

Public (pessimistic/uncertain):

24% say AI more likely to benefit them
43% say more likely to harm them
33% unsure

Gender Gap on Personal Benefit

Among U.S. adults:

Men: 31% foresee personal benefit
Women: 18% foresee personal benefit

Among AI experts:

Men: 81% foresee personal benefit
Women: 64% foresee personal benefit

Pattern: Men in both groups more likely to think they'll personally benefit.

Representation in AI Design

Racial and Ethnic Groups

Experts' views on how well perspectives are represented:

White adults: 73% say well-represented
Asian adults: 50% say well-represented
Black adults: 27% say well-represented
Hispanic adults: 25% say well-represented

Public's views (with more uncertainty, ~40%+ unsure):

White adults: 40% say well-represented
Asian adults: 25% say well-represented
Black adults: 19% say well-represented
Hispanic adults: 17% say well-represented

Pattern: White adults' views seen as far better represented than other groups in both expert and public perception.

Gender Representation

Experts:

Men's views: 75% say well-represented
Women's views: 44% say well-represented

Public (~40% unsure):

Men's views: 42% say well-represented
Women's views: 27% say well-represented

Women's own views on women's representation:

Among experts: 50% of men vs. 27% of women say women's views well-represented
Among public: 33% of men vs. 22% of women say women's views well-represented

Pattern: Women themselves less optimistic than men about representation of their own group.

Expert Quotes on Representation and Bias

"I identify as a woman, female, and I belong to a minority community. And I definitely think we need more representation, not just in terms of gender, but also in terms of ethnicities, places where people come from." – Asian expert

"I wanted [to use AI] to create an image of a Black family sitting around the table … Of course, there's many different blends of African American, and very different blends of Black people, but I thought it was interesting that it generated a series of images where everyone was biracial. If you don't know or if you can't relate to what a Black family looks like, it's only going to be amplified in the model that you're creating, right? … Now, the interesting thing is it's only a reflection of the biases that are in the people who are creating the information … we all have unconscious biases." – Black expert

"Most of the data that these large-scale AI systems are trained on is scraped from the internet. And [that] data … is inherently biased, not only according to those social categories that you mentioned, gender and race and so on, but … generally it comes from wealthy Western countries, wealthy White countries, and in particular along the coastline of those countries. … So it's a misconception that there is a malicious computer scientist embedding racism and sexism into the model. It's completely unintentional. It's essentially a product of ignorance of the data sources and the biases that are inherent in where we web-scrape the data from." – White expert

AI vs. Humans on Specific Tasks

Experts more optimistic AI will outperform humans:

Driving: 51% of experts vs. 19% of public say AI better
Customer service: Experts more optimistic (public often unsure - 16% not sure)
Medical diagnoses: 26% of public think AI better (still minority)
Loan decisions: Experts more optimistic

Both groups doubt AI superiority on:

Writing songs: 16% of experts, 14% of public think AI better
Parole decisions: 10% of public think AI better (lowest on list)
Hiring decisions: Low confidence in both groups

Pattern: Public has low confidence AI will outperform humans on any task surveyed (10-26% range). Experts more optimistic but views still mixed on many tasks.

Regulation and Responsible Use

Concern About Regulation Extent

Both groups worry regulation will be too lax:

More concerned regulation won't go far enough:

U.S. adults: Majority
AI experts: Majority

Uncertain:

Experts: 21%
Public: 16%

By Political Party (U.S. Adults)

Concerned regulation won't go far enough:

Democrats: 64%
Republicans: 55%

Pattern: Bipartisan concern, but Democrats more worried about insufficient regulation.

Confidence in Government and Industry

Lack confidence in U.S. government to regulate AI effectively:

Public: 62% have not too much/no confidence
AI experts: 53% have not too much/no confidence

Lack confidence in U.S. companies to develop/use AI responsibly:

Public: 59% have not too much/no confidence
AI experts: 55% have not too much/no confidence

Pattern: Widespread skepticism about both government and industry capacity.

Political Divides on Government Confidence

Republicans:

70% lack confidence in government to regulate AI effectively

Democrats:

54% lack confidence in government to regulate AI effectively

On company responsibility: No partisan difference

Republicans: 60% lack confidence
Democrats: 59% lack confidence

Expert Views by Job Sector

Confidence in companies to develop/use AI responsibly:

Those at colleges/universities:

60% have not too much/no confidence

Those at private companies/businesses:

39% have not too much/no confidence

Pattern: Academic experts more skeptical of companies than those working in private sector.

On government regulation: Similar views across sectors.

Expert Quotes on Regulation

"So I'm on both sides of this. I think we need to have limited regulations so that we can innovate and we can compete, because if we regulate too much, we're going to be left behind … [but] for us not to have guardrails around [AI], to me, is wild." – Expert working at a nonprofit

"[Companies] need to be transparent, and they need to include ways for people to either opt out or to correct anything that does not represent a person in their product." – Expert working in government

Methodology Notes

Survey of AI experts (methodology details in full report)
Survey of U.S. adults (representative sample)
Fielded: Summer and fall 2024
Before change in presidential administration (pre-2025)
Includes qualitative in-depth interviews with experts

Significance

Why it matters:

Rare comparative data - Expert vs. public opinion on same questions
Gender gaps documented - Excitement, concern, and personal impact perceptions differ significantly by gender in both groups
Representation concerns - Data on perceived bias in AI design by race/ethnicity and gender
Regulatory skepticism - Both groups want more regulation but doubt government/industry capacity
Job displacement anxiety - Biggest divergence between experts and public (31 percentage points)
Baseline measurement - Pre-2025 administration data for tracking future changes

Key tensions:

Want regulation ↔ Don't trust regulators or companies
Experts optimistic ↔ Public pessimistic
Concern about bias ↔ Low representation of minority groups in AI design
Innovation concerns ↔ Safety concerns ("we need guardrails" but "regulate too much, we're left behind")

dario-amodei-adolescence-of-technology - Expert view on AI risks (Anthropic CEO)
agent-psychosis - Critical view on AI coding quality concerns
Public opinion research on AI
AI bias and fairness issues
AI regulation debates

Context

Timing significance: April 2025 publication

After several high-profile AI incidents
During intense AI regulation debates
Public concern rising from ~40% (2021-2022) to ~50% (2023+)
Before some major 2025-2026 developments

What it captures: Snapshot of expert-public divide at moment when AI capabilities rapidly advancing but regulation lagging.

By

Pew Research Center