Navigating AI Comfort: A Holistic Examination of Public Attitudes

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Jan 26 2024

In a paper published in the journal Humanities and Social Sciences Communications, researchers conducted an omnibus survey with 1150 participants to investigate attitudes towards occupations based on their likelihood of automation. The findings revealed a general discomfort with artificial intelligence (AI) management across domains, with some deviations from the expected correlation between comfort levels and automation probability.

Demographic traits played a significant role, indicating that men and those with higher perceived technology competence were more comfortable with AI management, while individuals with lower internal locus of control were generally at ease with AI in various domains. Surprisingly, age, education, and employment did not influence comfort levels. This study offers a holistic perspective, incorporating demographic and self-efficacy variables, and highlights distinct perceptions of AI compared to other recent technological innovations.

Related Work

Past research has presented divergent views on the impact of AI, with optimists foreseeing societal improvement and pessimists anticipating negative consequences. The increasing adoption of AI in various sectors, encompassing decision-making in both private and governmental realms, has led to concerns about potential job displacement. Despite predictions that AI could replace almost half of American jobs, public opinion remains skeptical, with 50 percent of Americans expressing concerns about the negative societal impact of robot automation.

Survey on AI: Method and Analysis

The researchers conducted an omnibus survey with a sample size of 1150 participants using an online questionnaire through Qualtrics from April to June 2021. The survey focused on attitudes towards emerging technologies, including demographic and individual traits. The research team set quotas for gender, age, ethnicity, education, and income to align with the demographic distribution of the United States (U.S.) population. Although there were slight differences in age, education, and employment status, the sample closely mirrored the US population.

The dependent variable, comfort with AI, was measured using a Likert-type scale for various domains, categorized based on the likelihood of automation. Individual traits, such as locus of control, perceived technology competence, and innovativeness, were also assessed alongside demographic information.

The measurement section elaborates on the instruments used for data collection. Comfort with AI was gauged across different domains, aligning with automation likelihood. Researchers measured locus of control (LoC) using a modified 6-item scale, assessed perceived technology competence (PTC) with a scale from 2013, and measured innovativeness with a shortened version from 1977. Several demographic traits, including age, gender, income, education, race/ethnicity, and employment status, were included. The data analysis involved descriptive statistics, reliability tests, and ordinary least squares regression models to explore relationships between individual traits and participants' comfort levels with AI in different domains.

The study analyzed the results using the International Business Machines Corporation statistical package for the social sciences (IBM SPSS) statistics. Researchers presented descriptive statistics for all dependent variables, and regression models were employed to investigate the impact of individual traits on comfort levels with AI management across various domains. Variables were entered step-wise into the models, considering demographics and individual characteristics. This comprehensive methodology provides a foundation for understanding public attitudes towards AI, incorporating diverse variables to explore the nuanced factors influencing comfort levels in the face of increasing AI integration in different occupational domains.

AI Comfort and Predictors Summary

Participants exhibited a low level of comfort with AI management across various domains. Comfort levels somewhat aligned with the likelihood of automation probability, although certain domains deviated from this trend. Specifically, participants were least comfortable with AI managing lower automation probability tasks such as therapy, surgical teams, and air traffic control.

Conversely, they expressed slightly higher comfort with AI handling news desks and stock investing, falling within lower automation probability than with the automation-intensive domain of construction sites. Participants observed equivalence in comfort levels for tasks with higher automation probability, such as customer service, and functions with lower automation probability, like teaching.

Participants demonstrated the highest comfort with AI managing domains with higher automation probability, including supermarkets, sewage plants, and personal assistance. In human resources (HR) functions, participants were most at ease with AI managing employee work schedules. However, a decline in comfort was evident when it came to AI managing salary decisions and making hiring choices, with the slightest comfort observed in scenarios involving AI in firing decisions.

Researchers created indices to measure perceived comfort with AI in HR functions and occupations with high and low automation probability. Principal components analysis (PCA) validated these scales, revealing high reliability. A series of hierarchical linear regressions delved into the influence of individual traits on participants' comfort levels with AI across various domains.

Individual traits explained 11–20% of the variance in AI comfort across domains. Demographic characteristics, including gender, age, income, education, race/ethnicity, and employment status, explained most of the variance, ranging from 10 to 15%. Additional insights, constituting 5–7% of the variance, were provided by innovation and efficacy traits. Consistent predictors across domains included higher comfort levels for men, individuals with higher income, and those with greater perceived technology competence.

Conversely, individuals with a higher internal locus of control, indicating a greater sense of control over their lives, exhibited less comfort with AI management in every domain. Age did not influence participant comfort levels, while race and employment status showed limited impact. Education and innovativeness were not significant predictors in any of the models.

Conclusion

To sum up, the advent of generative AI tools has ignited widespread discussions on AI's potential and challenges. Marginalized groups, disproportionately affected by powerful algorithms, highlight the need to consider societal values carefully. While AI's computational capabilities have not eradicated social bias, they offer unique opportunities, especially in education, for historically marginalized individuals.

The study underscores the importance of regulatory actions aligning with public values. It emphasizes the next decade's critical collaboration among stakeholders to ensure sustainable and value-aligned AI implementation, focusing on empowering underserved communities for collective advancement.