Rare survey of AI experts exposes deep divide with public opinion.
Credit: piranka | E+
US experts who work in artificial intelligence fields seem to have a much rosier outlook on AI than the rest of us.
In a survey comparing views of a nationally representative sample (5,410) of the general public to a sample of 1,013 AI experts, the Pew Research Center found that "experts are far more positive and enthusiastic about AI than the public" and "far more likely than Americans overall to believe AI will have a very or somewhat positive impact on the United States over the next 20 years" (56 percent vs. 17 percent). And perhaps most glaringly, 76 percent of experts believe these technologies will benefit them personally rather than harm them (15 percent).
The public does not share this confidence. Only about 11 percent of the public says that "they are more excited than concerned about the increased use of AI in daily life." They're much more likely (51 percent) to say they're more concerned than excited, whereas only 15 percent of experts shared that pessimism. Unlike the majority of experts, just 24 percent of the public thinks AI will be good for them, whereas nearly half the public anticipates they will be personally harmed by AI.
Colleen McClain, a senior researcher for Pew Research Center, told Ars that a lack of studies examining "how the public's views lined up or did not line up with expert views" prompted the survey. Pew has spent the past four years surveying the public, observing that Americans have gradually grown more aware of AI and its potential.
But the new survey found that as awareness grows, "the US public has become more concerned over recent years." They're especially worried about deepfakes, misinformation, job displacement, and bias. It suggests that many Americans still feel very unsure about what AI is, what it can do, and how it might affect them. Pew expected that the experts' perspective was a missing piece of the puzzle when it came to parsing public opinion amid ongoing debates about how AI fits into society today.
To find US-based experts, Pew scoured AI conferences for "individuals who demonstrate expertise via their work or research in artificial intelligence or related fields" and created a list of authors and presenters. These conferences "covered topics including research and development, application, business, policy, social science, identity and ethics," featuring AI experts from industry and academia, as well as government and nonprofits.
While Pew could not ensure the sample was nationally representative of all experts in the AI field—which is very broad and hard to define—the survey represents a first step in gleaning how the people most committed to advancing AI view emerging technologies today.
Everyone agrees feds can’t be trusted to govern AI
Notably, Pew also found some common ground. Small percentages of each group expect AI will have a positive impact on news and elections, with most flagging concerns in these areas. And more than half of both sides agreed that they want more control over AI and do not trust the government to regulate AI—predicting that the US will be too lax. "They are also largely skeptical of industry efforts around responsible AI," Pew's survey said.
Last month, the White House fielded public comments for an AI Action Plan that will reveal to the public how the Trump administration intends to regulate AI. Among those who submitted comments was the nonprofit the Center for Democracy and Technology (CDT), which urged that more transparency and accountability are needed, as well as more public input. CDT recommended that the Trump administration seek public input and "evaluate and address risks to people’s privacy, civil rights, civil liberties, and safety."
In response to Pew's survey—which was conducted prior to Donald Trump's election win last year—CDT CEO Alexandra Reeve Givens told Ars that "AI’s widespread adoption is contingent on user trust. Just as traffic lanes and seat belts help people drive faster, well-tailored laws and norms will help people know what AI tools they can rely on in their daily lives. Without those safeguards, it’s no wonder the public is skeptical."
Alex Hanna, the director of research at the Distributed AI Research Institute, told Ars that the public and experts likely agree on regulation for a combination of reasons—because distrust in government broadly is a common sentiment and experts commonly expect the government to lack a sufficient understanding of technologies. And while "diversity is under attack" in government currently, it will be necessary for officials to incorporate diverse views on AI since "worldviews do get baked into" AI technologies, Hanna said, and that can affect people's lives.
Pew's survey found that experts and the public agreed that currently, "men’s views are better represented in AI design than the views of women." University of Washington AI professor Emily Bender suggested that to break that pattern, the US needs to be more genuine about including more perspectives in AI development at a time when big tech companies are ditching DEI initiatives.
"I think the through line is that these technologies are built to maintain the status quo and represent it as the norm," Bender told Ars. "What we really need is to move to a situation where the point of building this technology is for communities who are using it for their own purposes, have control over it, and decide when and how and where to use it."
Americans using AI more than they know, experts say
In a book due out this May, The AI Con, which provides guidance for policymaking and examines the human costs of profit-fueled corporate AI interests, Hanna and Bender work to help the public better understand AI's potential for good and bad. On their podcast Mystery AI Hype Theater 3000, they hope to further public scholarship, not to ensure the public is aligned with AI experts, but to help people form their own opinions about AI.
"We're trying to help people learn how to ask specific questions and understand automation in its context and sort of say, well, what's being automated and why?" Bender told Ars.
Bender and Hanna suggested that the goal shouldn't be seeking alignment and total agreement but fielding a diversity of opinions from all sides of society that would ensure that AI makers truly understand how to plug AI into various industries and communities.
"I think that there are folks out there, including us, who are trying to help people understand that everyone's expertise in their own field of work and personal relationships is really valuable," Bender said. "And the lens through which we should be evaluating any technology that somebody's trying to sell us is 'how does this actually work for me, for my community, and if I'm using it, or it's being used on me.'"
Although it's too late for the public to weigh in on the AI Action Plan, Pew plans to continue monitoring public opinion of AI to help "everyday Americans' voices" be included in these broader debates, McClain told Ars. Without more public awareness, AI experts threaten to dominate debates, potentially pushing views on policymakers that do not reflect the greater public's feelings or readiness for AI adoption.
That could quickly become a problem for many people who do not see the AI writing on the wall, Pew's survey suggested, since experts surveyed believe that Americans are already using AI more often than they think they are. Nearly 80 percent of experts responded that people likely use "AI almost constantly or several times a day," where only 27 percent of the public "think they interact with AI at this rate."
"New developments and tools evolve at a rapid pace, it's going to be important to continue tracking public feelings about these and tracking public awareness," McClain told Ars.
Americans do not expect AI to make them happy
Among both the general public and AI experts, women were more likely to be wary of AI than men, Pew's survey found, with the gender divide between the random expert sample even wider than the public sample.
The expert sample is not representative of the AI field, as Bender points out to Ars that AI is not a "thing" or even "a coherent set of technologies," and an expert in one area doesn't necessarily understand other areas. But still, Pew noted the key takeaway and conducted in-depth interviews to find out more about why men and women are or are not excited about AI.
"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," one male expert respondent said, describing his excitement about automating workflows.
A female respondent expressed concerns about biometrics collected at airports, noting, "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?”
But the "starkest" difference the survey found was in how AI experts and the public expect AI to impact jobs and the economy.
The public is "more anxious than experts about job loss," Pew's survey said. Where 73 percent of experts said that "AI will have a very or somewhat positive impact on how people do their jobs over the next 20 years," that share dropped to 23 percent among US adults. And while 69 percent of experts think the economy will benefit from AI, just 21 percent of US adults predict the same. What's more, "few in the public think AI will outperform humans on any of the tasks" the Pew survey explored, including parole decisions, medical diagnoses, hiring decisions, driving, providing customer service, or writing a song.
"Even as medical care is the one area in which the public is most optimistic about AI’s impact, experts are 40 percentage points more likely than the general population to believe it will positively affect medical care" (84 percent vs. 44 percent), Pew found.
Bender told Ars that pretending AI experts know best is likely problematic because "AI experts don't know very much about how people who work in other fields do their jobs, and the people who do those jobs are the ones who know what that's about."
Further, because the survey lumps together experts who build the technology and experts who study its societal impacts, it may be "obscuring some very different takes on the technology in a way that is also gendered" since "women tend to be clustered in the critical technology studies areas." Hanna guessed the woman describing concerns about biometrics above is likely an example.
Hanna said the survey is still helpful because it recruits a body of expertise to begin analyzing how views differ within the field and with the public. It perhaps helps push back against extreme narratives from tech leaders making "ridiculous" statements about a future that amounts to "fully automated communism," Hanna said, or predicting AI doomsdays.
For Americans, the future of AI apparently looks bleak, not because it possibly spells the end of the world but because 83 percent don't think it will make them more productive and 94 percent believe it won't make them any happier. Only 13 percent of Americans think they'll ever get to a point where they trust AI to make a decision for them.
Hanna told Ars that in some sectors, AI is already being used as an excuse not to hire more workers, and "even though that work is still there, it is being shoddily done, and it's possible that those people are being hired back at a fraction of their wage to look, and their work looks much more like gig work rather than more stable careers."
By contrast, one female AI expert quoted in Pew's survey expressed excitement about AI despite how it may impact her job.
“I’m excited about further automation of code, even though a lot of my job is software engineering, so that’s in competition with my job. I am excited about making the process even simpler than it is right now," she said. "In general, I think of AI as helping people along jobs. So I think of the biggest outcome is automation of processes that feel very slow and feel like they don't necessarily require full brain power, being automated by AI.”