Fully 78% of Americans believe that at least one of these abuses of AI is “very” or “somewhat” likely to affect the outcome of the election. There are no notable partisan differences here. Those aligned with each party are equally likely to be concerned. Those most likely to express concerns include adults 65 and older, White people, those with at least some college education and those living in households earning more than $100,000. Those who say they are not sure of the impact of AI in these areas are more likely to be young adults ages 18-34, Black people and Hispanics, those in households earning under $50,000 and those with a high school education or less.

Fully 93% of Americans want to punish candidates who are caught maliciously and intentionally altering or faking photos, videos or audio files during the election. Some of the patterns in the data:
Prevention from holding office is particularly favored by women.
Prosecution is particularly favored by those in households earning more than $100,000 and those with college educations.
A serious fine garners more Republican support than from Democrats (17% vs. 8%).
No penalty has only single-digit support across different groups.

By a nearly 8-1 margin, more Americans feel the use of AI will hurt the coming election than will help it. Still, there are notable shares of adults who are not sure what the impact might be or feel that the use of AI will not have much impact on the election process. Republicans are more likely than Democrats that the use of AI will “greatly” or “somewhat” hurt the process (49% vs. 38%). Those most likely to say they are not sure of the impact are those with high school educations and Black Americans.

Americans are not particularly confident in their own abilities to detect altered or faked videos, photos and audio files. And they have much less confidence in others’ abilities to discern fakery. Democrats are more likely than Republicans to be confident in their own abilities to detect faked videos and audio files; men are more confident than women; and those ages 18-34 are more confident than older people. People’s skepticism of others’ abilities to detect fakes runs especially high among those ages 35 and older, those in households earning $50,000 or more, those with at least some college education and White people. When it comes to adults’ judgments about others’ abilities to discern fake material, there are not any notable partisan differences. Members of both parties are equally likely to worry about others’ capacities.

Republicans are more likely to worry about the current news and information environment. For instance, Republicans are much less confident than Democrats that they themselves will get accurate information during the election (39% vs. 17%). Republicans are also less confident than Democrats that most voters will get trustworthy political information. The gap is 68%-50%. Beyond partisan differences, men, those age 65 and older and those with college degrees are particularly likely to have confidence they can get accurate political news. On the opposite side, those with lower confidence that most voters will get accurate information include the college-educated, those in high-income households and White people.

Roughly half of Americans say they find it “very” or “somewhat” easy to get the political news and information they want. Still, it is noteworthy that about a fifth of Americans find it difficult. Some 68% of Democrats say they find it easy, compared with 47% of Republicans. Meanwhile, 27% of Republicans say they find it hard, compared with just 12% of Democrats. Others who say it is relatively easy for them include men, those ages 35 and older, and those with at least some college education.

Predictable differences by partisanship show up in the survey data across these issues. Strong majorities of Republicans support Trump and strong majorities of Democrats support Biden. Four questions on this survey are most relevant to AI and politics. Again, partisanship dominates people’s choices when it comes to their views about which candidate can do a better job dealing with AI, with social media, competing with other countries in AI, and leading the military in adopting AI. For example: 53% of Republicans think Trump would do a better job dealing with the development of AI, while 3% think Biden. And 55% of Democrats back Biden here, compared with 4% who pick Trump.

One major concern about bias in AI systems is whether large language models (LLMs) generate answers to user queries that are fair and unbiased against some groups and their points of view. In this survey, we were particularly interested in how people assessed the answers of LLMs about politics and policy issues. AI systems are so new that it is not surprising that most partisans say they are not sure if there is bias in AI systems’ answers. Still, there are some notable differences among partisans who do express opinions on questions related to bias. Republicans are generally more wary of AI models than Democrats are. For instance, Republicans are more likely than Democrats to think these systems are biased against Republicans (the margin is 36%-15%). At the same time, Republicans are also more likely than Democrats to think the systems are biased against Democrats. The gap is 23%-14%.

Again, the biggest insight among the findings on these questions is that most people are not sure if AI systems are mostly fair or mostly biased when it comes to the answers the systems generate about public policy questions. Still, among those who are sure of their views, men and women somewhat diverge on the question of AI fairness when it comes to the answers given by AI systems about politics and public policy issues. By a 20%-14% margin, men are more likely than women to think that AI systems’ answers are mostly fair. Women are more likely than men to say they are not sure about that issue. Men are also more likely than women (20% vs. 12%) to think AI systems’ answers are mostly fair to women.

There are no clear, broad patterns in the answers to these questions about AI fairness and bias by different racial and ethnic groups. Black people are more likely than others to believe that the answers of AI systems on political issues are mostly biased against Black people. Some 33% of Black Americans say that, compared with 21% of White Americans and 19% of Hispanics. Black Americans are also more likely than other groups to believe AI systems are biased against Hispanics. Some 29% of Black adults say that, compared with 18% of Hispanic adults and 19% of White adults. It is interesting to note that when LLM users (who make up 23% of adults) answer these questions about bias they are more likely than non-users to think the models are mostly fair.

On these questions, the sharpest partisan differences emerge on the issue of whether the vote-counting process is free and fair. Some 79% of Democrats think that describes the country “very” or “somewhat” well, while 56% of Republicans think it does not describe the country well. On another issue, Democrats are more likely than Republicans to think most voters cast well-informed votes. Both sides are evenly split on the question of whether people usually agree on basic facts. Partisans also share similar views when it comes to whether the registration process allows all interested citizens to register and vote.

There are strong partisan differences on this question. Some 83% of Democrats are confident that people’s votes will be accurately cast and counted in the election, while 60% of Republicans think the opposite — they are not confident this will be the case. These figures are comparable to findings in Gallup polls on election integrity in recent election cycles.