Long Way to Go: Attitudes,Knowledge,and Perception of Artificial Intelligence in Health Care,Among a Racially Diverse,Lower Income Population in Houston,New York,and Los Angeles

Artificial Intelligence (AI) and related technologies have the potential to revolutionize health care and approaches to patient management. By utilizing algorithms and data-driven feedback, health care professionals are able to access robust medical information and receive predictive assistance in assessing risks and outcomes, all while minimizing the risk for human error. ¹ Recent developments, such as Google Cloud Health Care Data Engine, have been described as having the potential to assist in diagnosis and management of patients with rare diseases, ² whereas Microsoft's Generative Pre-Trained Transformer (GPT-3) has been heralded as instrumental in promoting health care providers' efficiency. ³

These efforts have been bolstered with the release of Chat Generative Pretrained Transformer (ChatGPT), an AI chatbot trained to problem solve and generate intelligent, human-like responses based on user input. ⁴ However, skepticism toward ChatGPT and AI-related technologies highlights pre-existing inequities and distrust in technology among minoritized populations who may need it most. ⁵ Although prior studies have assessed attitudes regarding AI in health care, few have focused on low-income populations who may or may not have access to technology. The study described here examined knowledge, attitudes, and perceptions of AI and ChatGPT in health care among a low-income, racially and socioeconomically diverse US adult population.

Working with local community-based organizations, a paper survey was disseminated to community residents 18 years and older, in low-income communities in Houston (Third Ward, East End), New York (Bronx, Brooklyn, Queens), and Los Angeles (East Los Angeles, Hyde Park, Huntington Park), between April 2023 and August 2023. The survey assessed access to technology, internet, AI, and related technology knowledge and usage in the target population. The survey included previously published questions, drawing from the Pew Research Center, National Health Interview Survey, the Health Information National Trends Surveys, and other published studies on this topic. ⁶

Respondents spent on average 8 minutes to complete the survey. Responses on 4-item Likert scales (strongly disagree, somewhat disagree, somewhat agree, strongly agree) were dichotomized to form binary responses of disagree versus agree for each technology attribute that we examined. A total of 305 surveys were returned, of which 212 were complete (69.5% completion rate). Descriptive analyses employing frequencies and proportions were used to describe respondent characteristics.

Overall, 41% of survey respondents identified as Black, 33% as White, 22% as Hispanic, and <4% identified as Middle Eastern/Asian/Native American/Pacific Islander/Other. About half of the sample were aged 40–64 years, and ∼7 out of 10 (69%) reported a pretax annual household income of <$35,000. In total, 57% of respondents were female, and with regard to highest education completed, only 20% reported having a bachelor's degree or higher (Table 1).

Importantly, we found that only 15% of survey respondents reported being confident regarding their knowledge of AI technologies, such as ChatGPT. Knowledge of ChatGPT was greatest among 18–39-year olds, and those with a bachelor's degree or higher, as well as respondents who self-identified as Middle Eastern/Asian/Native American/Pacific Islander/Other.

One third of survey respondents (34%) agreed AI was beneficial for patient care, whereas only 33% of respondents wanted their personal medical treatment to be supported by AI. None of the demographic characteristics examined arose to significance in this study, with older adults similarly as likely as younger respondents in their view of the benefits of AI and related technologies. Likewise, additional education was not associated with a preference for respondents' medical treatments to be supported by AI technologies.

In total, 71% of respondents indicated they were “scared of the influence of AI on medical treatment,” and 74% believed ethical concerns affected their attitudes toward use of AI technologies in medical treatments. Fear of the influence of AI was greatest among White respondents, respondents making <$35,000 per year, and those with a high school education or less. Ethical concerns were also greatest among respondents making <$35,000 per year, and those with a high school education or less.

The percentages reported in this study indicate much lukewarmness toward AI and related technologies when compared with percentages observed in an earlier study. However, they generally align with previous patterns reported by the Pew Research Center, which found that only 18% of US adults were aware of ChatGPT ⁷ ; 38% believed that AI could improve patient outcomes ⁷ and 39% indicated that they would feel comfortable if their health care provider relied on AI for their medical care. ⁷

Although biological sex was not associated with AI confidence in this study, the marked effect among lower income and less educated groups aligns with earlier work suggesting that younger adults 18–29 years of age, those with more formal education, and those with higher household income were more confident in AI's ability to improve patient outcomes. ⁷

That 85% of respondents had low knowledge about AI suggests an untapped educational opportunity. The limited awareness and negative attitude toward AI and ChatGPT only further highlight the technological disparities that underserved populations face. With ChatGPT being many steps ahead in the process of “accelerated digitalization” ⁸ of health care, immediate implementation of the application may be ineffective and unable to serve low-income populations who are already at risk for suboptimal health outcomes.

With the proliferation of AI in health care, establishing trust and regulations must be a priority for policy makers and technology developers. This can be achieved by transparency regarding operational processes, algorithms, and how personal patient information is handled. ⁹ By taking accountability and responsibility, developers can demonstrate to users that AI can continue growing to be accurate, reliable, and adhere to ethical guidelines. ⁹ In a study conducted by Skjuve et al on early adapters of AI, notable characteristics of poor user experiences included technical issues and challenges to formulate requests. ¹⁰

However, paired with teaching digital media literacy skills to encourage usage, ¹¹ interventions to engender trust in ChatGPT may play a large role in shifting opinions toward AI and related technologies among minority populations. Early work by Mostafa et al revealed that compatibility, perceived ease of use, and social influence boosted user's initial trust in chatbots, enhancing intention and encouraging engagement. ¹² A model for assistive social robots for older adult users proposed by Heerink et al suggests a deeper relationship, with intention to use playing a mediating role in the relationship between trust and actual use. ¹³ Further, analysis by Choudhury et al supports these findings noting that trust in AI had a significant direct effect on intentions to use and actual use. ⁹

Given this interwoven relationship between intention and trust, community-level interventions are the most efficient way to increase knowledge and trust in AI technologies. Although creating digital health interventions may be easier for organizers, trust is built by working at the community level. ¹⁴ By starting at the community level, cultural disparities and local issues affecting vulnerable populations and their attitude toward technology can be considered. ¹⁴ When establishing trust in a topic as polarizing as AI, it is important for the intervention be comprehensive and patient centered.

Oftentimes, community-level interventions are one-time events, with technology experts visiting communities to have quick conversations about health equity with no follow-up. This approach misses opportunities to build trust. Previous high-engagement, community-level interventions such as the PEN America's Media Literacy Program have been successful in improving digital health literacy and information-investigating skills of community members. ¹¹ Models like this can be expanded to educate and increase trust in AI within minority communities.