Fear, Optimism and Institutions: What Really Shapes Public Trust in AI

Artificial intelligence (AI) is presented as a technical challenge: make systems safer, more transparent, more explainable and more accurate, and public confidence should follow. However, the study "Beyond the Machine: Risk, Fear, Optimism and the Foundations of Public Trust in AI," by Steven David Pickering, Martin Ejnar Hansen and Yosuke Sunahara, published in AI & SOCIETY, argues that trust in AI is shaped just as much by the institutions, emotions and social expectations surrounding the technology as by AI itself.

Using nationally representative survey data from 3,235 respondents in Japan and the United Kingdom, the researchers find that trust in government, university scientists and other people consistently predicts trust in AI across both countries. Optimism raises trust, fear weakens it, and confidence in one's own ability to understand AI also matters. Yet the study also reveals sharp national differences, especially around employment: in Japan, believing AI may replace one's job is positively associated with trust in the technology, while no comparable relationship emerges in the UK.

Trust in AI Spills Over From Trust in Institutions

According to the study, citizens do not evaluate AI in isolation. People who trust their national government, university scientists and other members of society are significantly more likely to trust AI, even after demographic and attitudinal differences are taken into account. The relationship appears in both Japan and the UK and remains stable across multiple statistical models.

This supports what the researchers describe as a "trust spill-over" dynamic. When people lack detailed knowledge of how an AI system operates, they often rely on judgments about the institutions developing, regulating or legitimising it. In effect, confidence in political and scientific actors becomes a shortcut for deciding whether an unfamiliar technology deserves trust.

This makes AI governance partly an institutional credibility problem. Governments may publish technical standards, ethical principles and algorithmic transparency rules, but those measures operate within a wider political environment. If citizens distrust public institutions or scientific authorities, technically strong AI safeguards may still struggle to produce confidence.

The finding also complicates the idea that AI trust can be engineered principally through better interfaces or model performance. The researchers argue that public confidence depends on a broader ecosystem of institutional legitimacy and social trust. This means trust in AI may rise or fall alongside confidence in the actors seen as responsible for its deployment.

For policymakers, that shifts the focus from merely making AI systems trustworthy to making the institutions around them demonstrably trustworthy. Accountability, credible oversight and visible responsibility become central not just to ethical governance, but to public acceptance itself.

Optimism Builds Trust, but Fear Can Override Technical Confidence

Institutional trust is only one part of the picture. The study also shows that emotional responses to AI strongly influence whether people are willing to trust it. Respondents who expected AI to generate future benefits were significantly more trusting in both countries, while those who reported that AI scared them expressed lower levels of trust.

The contrast is especially striking in the UK. British respondents reported lower average trust in AI than Japanese respondents, 2.48 compared with 3.41 on a seven-point scale, and also recorded higher levels of fear. Japanese respondents, meanwhile, were more optimistic about AI and slightly more confident in their ability to understand how it could help them.

Public trust cannot be reduced to technical literacy alone. Citizens may understand AI reasonably well and still distrust it if they associate it with surveillance, discrimination, job loss or loss of control. Conversely, people may tolerate uncertainty about the technology when they expect substantial future benefits.

The study also finds that AI self-efficacy, the belief that one understands how AI can be useful, is positively associated with trust. In Japan, this relationship remains significant even after other factors are considered. In the UK, much of the relationship appears to operate indirectly through technological optimism: feeling capable of understanding AI seems to make people more hopeful about its potential, which in turn strengthens trust.

This gives policymakers a more nuanced reason to invest in AI literacy. Education should not simply teach citizens how systems work. It should also give people enough understanding and control to assess benefits and risks for themselves. The objective is informed confidence, not automatic enthusiasm.

Japan and the UK Reveal Two Different Politics of Automation

The most unexpected result concerns employment. Conventional assumptions would suggest that people who believe AI threatens their jobs should trust it less. The study finds that this relationship does not hold uniformly.

In Japan, respondents who believed AI would replace their own job were actually more likely to express trust in AI, even after accounting for other factors. In the UK, by contrast, the relationship was unstable and generally not statistically significant.

The authors suggest that national context may help explain the difference. Japan faces demographic ageing and persistent labour shortages, conditions under which automation may be interpreted less as an economic threat and more as a practical response to missing workers. AI can therefore be seen simultaneously as disruptive and useful.

The UK presents a different pattern. Fear plays a stronger negative role in trust, and job displacement does not generate the same positive association. The researchers interpret this as evidence that labour-market narratives and wider social expectations shape how economic vulnerability is understood.

For development policy, this is an important warning against universal AI communication strategies. The same technology can acquire different political meanings depending on labour shortages, unemployment, welfare protection, institutional trust and national debates over automation.

The lesson is particularly relevant for developing economies. In labour-abundant countries, AI-related job displacement may be interpreted very differently from countries struggling with ageing populations or worker shortages. Governments therefore need governance and communication strategies that reflect local economic realities rather than assuming that public attitudes toward AI will travel neatly across borders.

Trustworthy AI Requires More Than Transparency Rules

The researchers argue that governments should treat AI policy as part of a broader institutional ecosystem rather than as a technical regulatory exercise. Where trust in government, science and social institutions is weak, confidence in AI is unlikely to grow simply because systems are well designed.

Public education, workplace training and civic participation can strengthen people's sense of competence when dealing with AI. At the same time, risk communication needs to address real fears rather than dismiss them. In settings where anxiety is especially high, explainability, transparency and visible ethical safeguards may be necessary to reduce perceptions that AI is intrusive or uncontrollable.

Employment policy also becomes part of AI governance. The study argues that innovation strategies should be paired with safety nets, reskilling opportunities and credible plans for economic transition. Public trust is more likely to be durable when citizens believe institutions are capable of managing the costs of automation as well as promoting its benefits.

The research nevertheless has important limitations. It measures attitudes toward AI in broad terms rather than separating different applications, and its observational design identifies associations rather than proving causality. The authors therefore caution against assuming that raising institutional trust or AI self-efficacy will automatically cause higher AI trust.

There is also a deeper governance caution. Higher trust is not necessarily desirable in every circumstance. The authors explicitly warn that misplaced trust can magnify harm when oversight is weak or risks are poorly communicated. The objective should therefore be calibrated trust, confidence where institutions and technologies deserve it, and scepticism where safeguards are inadequate.