Public sees artificial intelligence as innovative yet risk-prone

Public attitudes toward artificial intelligence (AI) are often portrayed as either enthusiastic or fearful, but new research suggests the reality is more layered. The study finds that citizens distinguish sharply between different AI applications, weighing economic benefits, social risks, and cultural impacts in complex ways rather than adopting a single pro- or anti-AI stance.

Their findings are detailed in What Does the Public Think About Artificial Intelligence?—A Criticality Map to Understand Bias in the Public Perception of AI, published in Frontiers in Computer Science, which introduces a structured framework to map where public expectations and personal evaluations of AI align or diverge.

Public perception of AI: Optimism, fear, and domain differences

The research shows that public opinion on AI varies significantly depending on the domain of application. Respondents evaluated AI developments across six contexts: personal life, economic systems, industrial processes, societal effects, cultural impact, and healthcare.

In economic and industrial domains, AI was generally associated with innovation, productivity gains, and efficiency improvements. Many respondents considered it likely that AI would drive economic performance and automate unpleasant or repetitive tasks. These developments were typically rated as positive and probable, reflecting a pragmatic recognition of AI’s growing integration into business and industry.

Healthcare applications also received relatively favorable evaluations. AI systems that assist with diagnostics, treatment planning, and operational efficiency were perceived as both desirable and plausible. The perception of AI as a supportive tool rather than a replacement for human professionals appears to influence positive evaluations in this domain.

On the other hand, societal and cultural implications generated more critical responses. The most significant area of concern identified in the study was cybersecurity vulnerability. The prospect that AI systems could be hacked or exploited was rated as highly likely and strongly negative. This placed cybersecurity at the center of public anxiety regarding AI.

Other concerns included the concentration of AI development in the hands of elites or powerful actors, the potential erosion of communication quality, increased social division, and the risk that AI could eliminate more jobs than it creates. These developments were viewed as probable and undesirable, highlighting public apprehension about governance, equity, and social cohesion.

Interestingly, some developments were rated as highly desirable but unlikely. Participants expressed hope that AI could contribute to cultural enrichment, increase leisure time for all, or help solve complex global challenges. Yet they did not believe these positive transformations were probable. This gap between aspiration and expectation suggests a degree of skepticism about AI’s capacity to deliver broad societal benefits.

Likelihood and evaluation: A critical mismatch

According to the study, perceived likelihood and personal evaluation are not strongly correlated. In other words, people do not necessarily approve of developments they consider inevitable, nor do they believe that desirable outcomes will automatically materialize.

This mismatch creates what the authors conceptualize as zones of criticality. Developments that are considered likely and negative represent urgent areas of concern. Conversely, developments seen as positive but unlikely reflect unrealized potential that may require policy intervention or public engagement to become feasible.

The criticality map introduced in the study visualizes these tensions. It provides a framework for identifying where AI governance efforts should focus. For example, addressing cybersecurity vulnerabilities and concerns about elite control may reduce perceived risks. At the same time, promoting inclusive AI innovation could help align hopeful expectations with realistic pathways.

The research also examines how trust and distrust influence perception. Interestingly, individuals who expressed higher levels of distrust in AI tended to rate AI developments as slightly more positive overall but less likely to occur. In contrast, higher trust was associated with greater perceived likelihood but slightly less positivity.

This counterintuitive pattern suggests that trust shapes expectations differently from evaluations. Those who distrust AI may idealize its potential while doubting its realization. Those who trust AI may view its expansion as inevitable while remaining cautious about its social consequences.

The authors interpret these findings through cognitive bias frameworks, including the affect heuristic. For many citizens, AI remains an opaque or abstract concept. Media narratives, science fiction portrayals, and limited technical literacy can shape emotional responses. These influences may amplify certain fears or inflate certain hopes without a corresponding understanding of technical feasibility.

Implications for AI governance and public engagement

According to the authors, understanding public bias is essential for aligning AI development with societal values. The criticality map serves as a decision-support tool for multiple stakeholders.

For policymakers, it highlights domains where regulation or oversight may be necessary to address public concern, particularly in cybersecurity and power concentration. For developers, it underscores the importance of designing transparent, secure, and inclusive systems that address social anxieties. For educators, it reveals the need to strengthen AI literacy to reduce misconceptions and enable informed participation in policy debates.

The study also touches on labor market perceptions. While participants expressed concern about job displacement at the macro level, they did not strongly perceive personal vulnerability. This divergence suggests the presence of optimism bias or confidence in individual adaptability. It also reflects the complexity of public attitudes toward automation and employment.

The authors situate their findings within the broader context of the Collingridge dilemma, which holds that emerging technologies are easier to regulate early in their development but harder to change once entrenched. Regularly updating perception studies is therefore critical to democratic oversight of AI expansion.

The research calls for more nuanced public discourse about artificial intelligence. Binary narratives of technological salvation or catastrophe obscure the differentiated judgments citizens make across domains. Public perception is shaped not only by technical knowledge but also by values, trust, and social experience.

Importantly, the study acknowledges its limitations. The sample size was modest and geographically concentrated in Germany, which may limit generalizability. Future research across diverse cultural contexts is needed to capture global variation in AI perception. Nevertheless, the structured methodology and dual-dimension analysis provide a replicable model for further investigation.