Cross-National Variations in Public Attitudes Toward Artificial Intelligence: A Latent Class Analysis

• In Event: AI, Labor Markets, and Economic Impacts

Abstract

Governments and organizations are rapidly integrating artificial intelligence (AI) into decision-making, public services, and economic systems. Understanding public attitudes toward AI has become increasingly important as these views shape societal acceptance and the legitimacy of governance frameworks (Cave et al., 2019; Fast & Horvitz, 2017; Araujo et al., 2020). Existing research often treats these attitudes as uniform or examines them along a single dimension, overlooking the diverse combinations of beliefs people may hold about AI. Our study addresses this gap by examining how public attitudes toward AI vary across individuals and national contexts. Using cross-national survey data from the University of Toronto’s Global Public Opinion on Artificial Intelligence (GPO-AI) dataset, we apply latent class analysis (LCA) to identify clusters of respondents with similar attitudes toward AI. We then link these profiles to country-level indicators such as GDP per capita, Gini coefficients, and internet freedom to assess how national contexts shape public opinion. The LCA identifies seven distinct classes based on eight questionnaire items measuring the global valence, concerns about catastrophes and existential risks, and opinion on the pace of AI development. Excited accelerationists (11.2%) are highly positive and excited about AI, relatively unconcerned about risks, and supportive of rapid development. Concerned enthusiasts (17.0%) are also highly positive, but express serious concerns and prefer a more cautious pace. Moderate supporters (18.4%) are generally positive with limited concern yet favor slower development. Dismissive skeptics (12.9%) hold mostly moderate or neutral views, are not excited about AI, and are skeptical of catastrophic or existential threats. Concerned moderates (16.3%) are similarly moderate in overall outlook but express serious concerns and strongly prefer caution. Staunch opponents (11.2%) are deeply concerned about AI-related risks and strongly opposed to AI. Disengaged/undecided respondents (13.1%) are neutral or unsure and show no clear preferences. Preliminary analysis reveals substantial cross-national variation. Excited accelerationists are more common in emerging economies such as Brazil, China, and India. Dismissive skeptics are more prevalent in continental Europe (e.g., France, Germany, and Spain). Staunch opponents are most common in affluent Western countries such as the US, Canada, UK, and Australia. The proportion of “disengaged/undecided” is lower in developing countries such as Kenya, South Africa, India, and Indonesia but exceedingly high in Japan. Our next step is to employ multilevel modeling to examine how these patterns are associated with economic development, modernization, and income inequality (Inglehart & Baker, 2000; Sturgis & Allum, 2004; Acemoglu & Restrepo, 2020), while considering a potential legacy effect by evaluating whether government responses to the COVID-19 pandemic have in any way shaped public attitudes toward AI (You et al., 2024). This study contributes to the literature by demonstrating that public attitudes toward AI are structured and context-dependent rather than simply positive or negative. By integrating latent class modeling with cross-national analysis, it provides a more nuanced understanding of global public opinion on AI and offers insights for policymakers seeking to design equitable and context-sensitive AI governance strategies.

Authors

• Mahzabin Fatema Kabir, University of North Carolina, Charlotte
  Presenting Author

• Yang Cao, University of North Carolina, Charlotte
  Non-Presenting Co-Author