AI sparks global divide: Who wins, who loses and what comes next

Clerical workers stand out as the only category disproportionately represented among self-perceived “losers” and underrepresented among “winners.” Manual workers and those in service roles are also more likely to see AI as a threat. Conversely, professional-managerial workers are among the most optimistic, as are men and respondents from higher income quintiles.

Artificial Intelligence is reshaping the future of work in Britain, but its impact is far from uniformly felt. While some workers expect AI to enhance their job prospects, many fear they will be displaced—regardless of what occupational data might suggest. These contrasting perceptions are now influencing political attitudes in ways that could reshape policy debates in the years ahead.

A new peer-reviewed study titled “Linking Artificial Intelligence Job Exposure to Expectations: Understanding AI Losers, Winners, and Their Political Preferences” was published in the April–June 2025 issue of Research and Politics. Authored by Jane Green, Zack Grant, Geoffrey Evans, and Gaetano Inglese, the paper uses original survey data from over 4,200 currently employed adults across Great Britain to examine how subjective expectations about AI align with expert measures of job exposure and political preferences.

How do workers perceive the impact of AI on their jobs?

The study reveals that over one-third of British workers anticipate a personal impact from AI in their professional lives. Specifically, 24% see themselves as AI “losers,” believing their job prospects will worsen, while just 10% consider themselves AI “winners.” The remaining 47% expect no significant change, and about 17–20% are uncertain.

This pessimism exceeds public concern around other economic shifts, including immigration, technological advancement in general, and import competition, and is second only to fears of a recession. Among those in jobs deemed highly exposed to AI, nearly 41% expect AI to either benefit or harm their prospects, compared to just 20% among those in low-exposure occupations.

Subjective expectations were linked, but not perfectly aligned, with expert-derived AI occupational exposure scores. Three main indices were used: the Felten et al. (2021) AIOE index, which maps AI capabilities to required job skills; the Pizzinelli (2023) model, which adds occupational context and complementarity; and the Gmyrek et al. (2023) index, which assesses automatability using GPT-4's interpretation of job tasks. These models overlap in classifying high-exposure roles but differ on nuances such as augmentation potential versus outright substitution.

Who sees themselves as AI winners or losers?

The study identifies that demographic variables significantly influence whether someone sees AI as a risk or opportunity. Younger people, university graduates, and full-time workers are more likely to believe AI will improve their job prospects. Interestingly, many of these same groups also show high levels of AI-related anxiety, suggesting intra-group polarization.

Clerical workers stand out as the only category disproportionately represented among self-perceived “losers” and underrepresented among “winners.” Manual workers and those in service roles are also more likely to see AI as a threat. Conversely, professional-managerial workers are among the most optimistic, as are men and respondents from higher income quintiles.

However, these perceptions persist even after accounting for actual AI exposure. That means people’s expectations are shaped not only by what their job entails but also by broader social and psychological factors, including digital literacy, trust in technology, and individual economic insecurity.

This divergence has critical implications. Workers in jobs technically less exposed to automation may still push for policy responses if they perceive themselves at risk. Similarly, those in high-exposure jobs may underestimate the threat, delaying reskilling efforts or political advocacy.

What do AI winners and losers want politically?

Perhaps the most consequential aspect of the study is how AI perceptions are beginning to polarize political preferences. Both AI winners and losers express stronger support for income redistribution than those who anticipate no impact. However, their views diverge significantly on two other fronts: education investment and immigration policy.

Self-declared AI winners overwhelmingly support increased government spending on adult education and retraining programs, as well as more liberal immigration controls. AI losers, on the other hand, are less supportive of immigration and somewhat less enthusiastic about investment in education, even though both groups show equal interest in economic redistribution.

These findings suggest AI may disrupt traditional political coalitions. While past waves of automation and globalization predominantly affected lower-skilled, working-class workers, AI’s reach extends into middle- and high-skill professions. As a result, subjective winners and losers are no longer confined to distinct socioeconomic groups. The highly educated, for example, are overrepresented among both optimists and pessimists.

This complexity makes AI harder to politicize along conventional lines. However, it also opens new opportunities for political entrepreneurs who can frame AI in ways that resonate with public anxieties or aspirations. Whether such narratives coalesce into electoral outcomes remains to be seen, but the data show the groundwork is being laid.