Although most Americans have heard of AI, they are less aware of the specific health applications of AI. Americans anticipate that medicine is most likely to benefit from advances in AI compared with other fields such as military science. The anticipated benefits vary depending on the type of application; however, they are less clear for applications that involve AI making health decisions for people.

A total of 2 surveys included questions related to familiarity and experience with AI generally. Most Americans have at least some familiarity with AI. A 2015 survey by the Monmouth University Polling Institute (n=1005) found that 70% of Americans had heard of the term “artificial intelligence” or “AI” [29]. For most, direct experience with AI comes through their cell phones and smart personal assistants, such as Alexa (Amazon) and Cortana (Microsoft Corporation), Siri (Apple), and Google Assistant (Google). A 2020 survey by the SCIMEP research group (n=2700) found that 83% of Americans used smartphones and 29% of Americans reported using smart personal assistants “at least once” to “a few times a week” [30].

However, poll data suggest that there is still ample room for Americans to learn more about AI. In the same 2015 Monmouth University survey (n=1005), only 12% of Americans reported having read or heard “a lot” about recent advancements in AI technology, whereas 88% of the Americans had read or heard “a little” or “nothing at all” [29]. As with many issues, people may report opinions on AI based on the information they hold at these early stages as public familiarity with AI increases.

A total of 3 surveys included questions related to the applications, risks, and benefits of AI in health care. Health care is a prominent area in which people anticipate the benefits of AI development and application. In a 2016 Vanity Fair/60 Minutes survey (n=1021), respondents were asked which fields were most likely to benefit from advances in AI [32]. Slightly fewer than half (44%) of the respondents answered “medicine,” topping the list. For comparison, 23% answered “military science” and 13% answered “automobile manufacturing” [32].

Despite the view that health care as a field will likely benefit from AI advancement, Americans are less aware of the specific health applications of AI (Figure S1 in Multimedia Appendix 2). The 2020 SCIMEP survey (n=2700) found that about half of Americans (48%) had either “little” or “no” awareness of using AI to diagnose diseases more efficiently, and only one-fifth of Americans (19%) were “quite a bit” or “a great deal” aware of use of AI in disease diagnosis [30].

When it comes to the likelihood of AI applications to improve individuals’ health, the 2020 SCIMEP survey (n=2700) found that Americans’ views were more divided, with 34% responding that AI is “likely,” “very likely,” or “certainly” going to improve health, 35% responding “somewhat likely,” and 31% responding “unlikely,” “very unlikely,” or “not at all likely” [30]. Consistent with trends of general support for AI [34], the SCIMEP survey found that the perceived likelihood of AI improving individuals’ health was positively associated with their education and income [30]. Among Americans with at least a 4-year college degree, for example, 43% said that AI will “likely” to “certainly” improve health. That is twice the number of respondents who said it is “unlikely” to “not at all likely” (21%) in that same group. In comparison, only 31% of those who had a high school diploma or less said that AI will “likely” to “certainly” improve health. Likewise, as household income increased, the perceived likelihood of AI improving individuals’ health also increased. Among Americans whose annual household income was over US $100,000, twice as many people were “likely” to “certainly” (45%) to believe that AI would improve health compared with “unlikely” to “not at all likely” (21%). Of those with a household income of US $30,000 to US $70,000, only 33% responded “likely” to “certainly” [30].

Although Americans generally have positive views about using AI technology to advance public health and medical fields, their attitudes become more cautious when it comes to allowing AI to make important personal health decisions. A 2021 survey by Pew Research Center (n=10,260) asked how “excited or concerned” respondents would be “if artificial intelligence computer programs could diagnose medical problems” [37]. Results were divided, with 40% of respondents either “very” or “somewhat” excited, 24% equally concerned and excited, and 35% either “somewhat” or “very” concerned. In another example, the 2016 Vanity Fair/60 Minutes survey (n=1021) found that only 8% of respondents felt comfortable letting a computer with AI decide on their end-of-life care [32]. Given the range of potential health applications of AI, additional research is required to explore how specific perceived risks and benefits vary across uses and contexts.

Specific application goals, such as disease prediction, diagnosis, and treatment, matter to Americans’ attitudes toward AI in health care. Secondary analyses showed clear demographic differences regarding acceptance of some of these applications, such as the use of AI for mental health monitoring and treatment. Americans also recognize the importance of future governance challenges related to the use of AI for disease diagnosis.

A total of 3 surveys included questions on disease prediction and diagnosis. The 2020 SCIMEP survey (n=2700) data indicated a salient difference between Black and White Americans in their concern about AI worsening discrimination against people based on health risks. Among Black Americans, 40% were “concerned” or “very concerned” about AI worsening discrimination on the basis of individuals’ health risks, compared with 33% of Black Americans who were “slightly” or “not at all” concerned [30]. In contrast, 30% of White Americans were “concerned” or “very concerned” about AI worsening discrimination against people based on health risks, whereas 44% of White Americans were “slightly” or “not at all” concerned about such worsening discrimination.

Surveys that focus on the use of AI for mental health monitoring and treatment, however, find public opinion to be divided somewhat differently among Black and White Americans, with White Americans being more concerned about such uses (Figure S2 in Multimedia Appendix 2). The 2019 Pew Research Center poll (n=4272) asked whether social media companies should monitor their users’ posts for signs of depression to identify people who are at risk of self-harm and connect them to counsel [27]. Overall, 45% of Americans found this use of personal data to be unacceptable compared with 27% who found it acceptable. Among those, half of the White Americans (50%) found it unacceptable, whereas only a quarter (24%) found it acceptable. In contrast, Black Americans showed slightly higher acceptance (41%) than unacceptance (34%) [27].

The 2020 SCIMEP survey (n=2700) asked a similar question regarding companies using AI and personal information to make mental health predictions, although in this case, insurance providers use AI to predict whether a person will develop depression in the future [30]. Consistent with the 2019 Pew Research Center data, White Americans showed higher concern (51% “concerned” or “very concerned”) about this use than Black Americans (43% “concerned” or “very concerned”) and Hispanic Americans (39% “concerned” or “very concerned”). Education levels mattered as well across all races. More than half of Americans with a 4-year college degree or higher education (55%) were “concerned” or “very concerned” about the application of predicting depression by insurance companies, without substantial differences across racial groups [30].

Americans express a desire for transparency in the management of AI disease-diagnosis technology. The 2019 Zhang and Dafoe survey (n=2000) found that nearly half of Americans (44%) agreed that accuracy and transparency in AI used for disease diagnosis would likely be an AI governance challenge that will impact large numbers of people in the United States in the next 10 years [34]. Among those, 10% thought this governance challenge would be “very likely,” 13% “likely,” and 21% “somewhat likely.” More than three-quarters of Americans also viewed AI applications in disease diagnosis as very important (56%) or somewhat important (22%) for technology companies and governments to manage carefully [34]. The perceived importance of carefully managing the challenge of AI disease diagnosis is comparable with the perceived importance of managing challenges in AI applications in surveillance (59% “very important” and 19% “important”) and a bit lower than that of data privacy (64% “very important” and 18% “important”).

In addition to health monitoring and diagnosis tools supported by AI, 3 surveys included questions about AI providing treatment in the literal sense of AI robots as caregivers (Figure S3 in Multimedia Appendix 2). Americans seem to be skeptical of the value of robots and AI in caregiving and care-related decision-making, at least as replacements for human caregivers. A 2014 Pew Research Center survey (n=1001) found that the majority (65%) of respondents thought it would be a change for the worse, if lifelike robots became the primary caregivers for the older adults and people in poor health, whereas 28% of respondents thought it would be a change for the better [31]. There were no substantial gaps in attitudes toward robotic caregivers across racial and gender groups.

Americans hold similarly unfavorable views regarding the use of AI robots in hospital-level decision-making and resource allocation. The 2018 Ghafur et al [33] survey (n=1114) found that only 38% of respondents were happy to receive AI-assisted health advice. In the 2015 Monmouth University survey (n=1005), 65% of Americans said it was a “bad idea” to use robotic nurses with AI to diagnose situations and decide when to administer medicine for bed-ridden patients, whereas only about 3 in 10 thought it was a “good idea” [29]. Broken down by demographics, Hispanic (49%) and Asian Americans (44%) had the most favorable views of this application of AI, followed by Black (33%) and White Americans (27%). A gender gap exists here as well, with 38% of men thought it was a good idea, whereas only 24% of women agreed [29].

Data sharing and privacy associated with health-related data are key concerns among Americans. Most Americans reported wanting control over their personal health data. Willingness to share their personal health information largely depends on which organization or institutional actor collects the data, the intended purpose of data sharing, and the specific groups being tracked. Americans report lower trust in commercially oriented organizations, such as technology companies, compared with doctors or pharmacists. Americans also express a desire to control their health information, especially concerning how others access and share it. Demographics such as race and education shape attitudes regarding data sharing and privacy.

A total of 3 surveys included questions on control over personal health data. In a 2019 survey by the Rock Health and Stanford Center for Digital Health (n=4000) on consumer adoption of digital health technologies, most respondents (82%) expressed a desire to control those who had access to their health data. Similarly, 81% said they “agree” or “strongly agree” that “I should be told what data had been collected about me” [35].

Americans also want control over the life span of their personal health data (Figure S4 in Multimedia Appendix 2). In a 2019 Pew Research Center survey (n=4272), 71% of respondents agreed that all Americans should have the right to have their personal medical data collected by a health provider permanently deleted [27]. White Americans (74%) were more likely to agree that they should have the right to remove personal medical data compared with Black (60%) and Hispanic (66%) Americans. There was a prominent education-based divide within the Black American community. Black Americans who had obtained at least a college degree were more likely to agree that they should have the right to delete their medical data (79%) than those who only had a high school diploma or less (46%) [27]. In this survey, education did not similarly predict attitudes for White Americans.

Moreover, the intended purpose of health data sharing is also a key influence on attitudes toward sharing health data. In the same 2019 Pew Research Center survey (n=4272), respondents were asked whether they thought “DNA testing companies sharing their customers’ genetic data with law enforcement agencies in order to help solve crimes” was an acceptable or unacceptable use of their data [27]. About half of the White (51%) and Hispanic (50%) Americans thought it was acceptable. Black Americans (46%) showed slightly less acceptance compared with Americans of other races. In addition, Black Americans with a college degree showed lower acceptance (38%) than their Black counterparts who had attended college (45%) or had a high school diploma or lower (50%) [27]. The 2020 SCIMEP survey (n=2700) similarly found that Black Americans had greater concern (46% “concerned” or “very concerned”) about DNA testing companies sharing their customers’ genetic data with law enforcement agencies to help solve crimes than White (40% “concerned” or “very concerned”) or Hispanic Americans (31% “concerned” or “very concerned”) [30].

In contrast, the 2019 Pew Research Center survey (n=4272) found that when the purpose of data sharing was “makers of a fitness tracking app sharing their users’ data with medical researchers seeking to better understand the link between exercise and heart disease,” White Americans had lower acceptance (40%) than their Black and Hispanic counterparts (47% and 49%, respectively) [27]. Across all racial groups, Hispanic Americans with higher levels of education (some college, 58%; college degree or more, 60%) or an income of US $30,000 to US $74,999 (64%) had the highest levels of acceptance.

Three surveys included questions related to trust in controlling personal health data. Surveys indicate that the willingness to share personal health data largely depends on who the data-receiving body is (Figure S5 in Multimedia Appendix 2). A 2018 survey by Ghafur et al [33] (n=1114) indicated that Americans were most willing to share their anonymized health information with personal doctors (61%), followed by one’s family (41%), academic or medical research institutions (26%), and pharmacists (25%). They were least willing to share with technology companies, such as Google, Amazon (4%), or any other commercial companies (3%). In addition, 15% of respondents were unwilling to share their anonymized health information with anyone [33].

A 2020 survey by Rock Health and the Stanford Center for Digital Health (n=7980) found that consumers were most willing to share their health information with their physicians (72%), followed by their families (52%), health insurers (53%), pharmacies (46%), research institutions (35%), health technology companies (25%), pharmaceutical companies (22%), their employers (15%), government organizations (12%), and technology companies (11%) [36]. The same question was asked in a 2019 survey by Rock Health (n=4000) and the findings were consistent [35].

The 2020 Rock Health survey (n=7980), which was conducted during the COVID-19 pandemic, also asked respondents about their willingness to share their COVID-19 results with these groups [36]. One-third (32%) of the respondents were willing to share their COVID-19 results with their employers, and approximately one-fourth (23%) were willing to share them with government organizations. This is twice the number of respondents willing to share their general, broader health information with these entities (15% and 12%, respectively) [36].

However, overall, these findings suggest that respondents’ willingness to share personal health information decreases as the data-receiving institution becomes more commercially oriented.

One survey included questions related to the location tracking of health-related data. The results of this survey suggest that views on whether such tracking is useful or acceptable are mixed. A survey by the Pew Research Center conducted in April 2020 (n=4917) asked respondents, “if the government tracked people’s locations through their cell phones during the coronavirus outbreak, do you think this would help a lot, a little, or not make a much difference in limiting the spread of the virus?” [28]. Most US adults (60%) said that such actions would not make much of a difference in limiting the spread of the virus.

Regarding acceptability, people have nuanced views depending on who is being tracked and for what purpose. In the scenario of governments using people’s cell phones to track their location, Americans were relatively tolerant of the use for those who have “tested positive for the coronavirus in order to understand how the virus may be spreading,” with just over half (52%) of respondents reporting it was either “very” or “somewhat” acceptable (24% and 28%, respectively) [28]. This decreased slightly for those who may have “had contact with someone who tested positive for the coronavirus,” with 45% responding that it was “very” or “somewhat” acceptable (19% and 26%, respectively). The lowest acceptance was for location data use to “ensure [people] are complying with experts’ advice on limiting social contact during the coronavirus outbreak,” with 37% responding “very” or “somewhat” acceptable (14% and 23%, respectively) [28].

White Americans had the lowest acceptance for all 3 location-tracking strategies among racial groups. Nearly half of White Americans (47%) reported it was either “very” or “somewhat” acceptable (19% and 28%, respectively) for the government to track people who tested positive for coronavirus [28]. In comparison, 56% of Black Americans said that it was “very” or “somewhat” acceptable (34% and 22%, respectively) and 66% of Hispanic Americans said it was “very” or “somewhat” acceptable (33% and 32%, respectively). Similarly, only about a third (31%) of White Americans thought the use of location tracking was “very” or “somewhat” acceptable (12% and 19%, respectively) to ensure limiting social contact. This was much lower than the 45% of Black Americans who said it was “very” or “somewhat” acceptable (24% and 21%, respectively) and 55% of Hispanic Americans who said it was “very” or “somewhat” acceptable (23% and 32%, respectively) [28].

These distinctions raise interesting and important questions for communication and engagement on how differences in experience with the pandemic, governments, technologies, and other factors vary depending on one’s race to shape views of AI use and acceptability.