Summary:
- A study finds that institutional trust is the strongest predictor of trust in high-risk AI, meaning people judge these systems politically as well as technologically.
- Three cities showed distinct trust profiles: Singapore paired high institutional trust with the highest perceived risk, while Seoul and Tokyo displayed lower government trust, requiring different governance strategies for each.
- Since trust in tech companies was linked to trust in high-risk AI, governments cannot rely on state-led governance alone to build public trust and should also engage trusted private-sector players, tailoring strategies to each society.
Imagine waking up with just 90 minutes to clear your name of murder – not to a jury of your peers, but to an icy, glitching artificial intelligence (AI) judge. While Hollywood turns automated sentencing and predictive policing into a thrilling action film like Mercy, society is already grappling with a cold reality: AI is shifting the foundations of how public institutions operate. As these systems increasingly drive high-risk decisions that determine who receives rights, resources, and opportunities, how do we regulate technology where a single line of code can permanently alter a human life?
Based on the European Union’s Artificial Intelligence Act, high-risk AI systems constitute those that pose risks to the fundamental rights, safety, and health of human beings. Such systems are used across diverse domains, such as biometric identification, facial recognition, recruitment, education, welfare, law enforcement, border control, and systems that monitor or manage transport or energy networks.
To counter these ethical risks, policy responses have focused on making AI systems more ‘trustworthy’ through technical and procedural safeguards. To date, policy frameworks and public debate on high-risk AI have focused on transparency, explainability, fairness, and robustness. These measures share a common assumption, that is, improve the technology, and public trust will follow.
However, building on evidence from Seoul, Singapore, and Tokyo, a recent study co-authored by Dean's Chair Associate Professor Araz Taeihagh of the Lee Kuan Yew School of Public Policy demonstrates that this hypothesis does not draw the complete picture.
“High-risk AI is not necessarily operated by its users,” Assoc Prof Taeihagh explains. “Governments and companies deploy it upon individuals within opaque institutional and regulatory structures, so people face a knowledge deficit.” Individuals thus assess high-risk AI through a political lens rather than a purely technological one.
The Tale of Three Cities
The study surveyed 3,600 respondents across Seoul, Singapore, and Tokyo. While all these cities are leaders in AI readiness and digital infrastructure, they differ in their institutional environments and patterns of public trust.
Key survey findings highlight how public sentiment differs across these digital powerhouses, reflecting the broader trust context.


In particular, Singapore is characterised by strong state capacity and relatively high levels of trust in the government. The authors note that Singapore’s highly centralised AI governance, which features coordinated agencies like GovTech, sits within a society that consistently ranks high in the Edelman Trust Barometer.
Despite being similarly technologically advanced, Seoul and Tokyo displayed lower levels of trust in the government. Seoul is part of a presidential democracy where civil society is actively involved in the discourse of AI governance. Meanwhile, Tokyo has a more fragmented governance landscape focused on industrial automation, and historically moderate public trust in political leadership.
The central finding of the study is consistent across all three cities: institutional trust is the strongest predictor of trust in high-risk AI systems. In the study’s statistical model, institutional trust comprises perceived appropriate use of high-risk AI, trust in government, and trust in technology companies.
Intriguingly, the study reveals a striking paradox in Singapore. While Singaporeans expressed the highest trust in institutions and technology companies, they also perceived the highest level of risk. A related pattern held in all three cities: people more inclined to trust technology also perceived greater risk, consistent with the ‘informed scepticism’ seen across cultures. In the study’s framework, risk is an emotional response driven by fear and dread, together with concerns about data quality, bias and job loss. For such citizens, the authors suggest, a deeper understanding of AI does not eliminate this fear; instead, it makes them highly discerning.
“Such a discerning public is likely to judge embedded systems by the capacities our respondents assessed: the ability of government to manage risks, protect data and monitor both corporations and the implementation of high-risk AI, and the trustworthiness of the companies involved,” Assoc Prof Taeihagh says.
Insight into AI Governance
The three countries have different, localised approaches to high-risk AI governance rather than a single, shared regional law like the EU AI Act. Singapore encourages companies to voluntarily test their systems using the government’s AI Verify toolkit; Japan’s AI Promotion Act uses non-binding guidelines and public ‘naming-and-shaming’ to keep developers in check without imposing fines; and South Korea enforces a legally binding AI Basic Act that mandates risk management and human oversight for ‘high-impact’ systems.
While all three governments actively push AI integration through high-profile initiatives like Singapore’s Smart Nation, Tokyo’s Smart Tokyo, and Seoul’s Smart City Master Plan, the study highlights the inherent limits of this top-down approach. State ambition and technical benchmarks alone cannot buy public trust. Without addressing the underlying institutional trust deficit, current strategies risk creating a wide gap between state policy and public acceptance. “Lack of trust can breed both misuse through overreliance and disuse through neglect or underuse,” Assoc Prof Taeihagh emphasises.
Because public trust profiles differ so vastly, strategies should differ by context, Assoc Prof Taeihagh says. In Tokyo and Seoul with low government trust and higher tech company trust, the immediate priority is addressing the trust deficit between citizens and the state – engaging the private sector in development and implementation offers an alternative route. In Singapore, where government trust and perceived risks are high, risk communication is paramount.
Crucially, the study also finds that trust in technology companies – not government – was positively associated with trust in high-risk AI in all three cities. This suggests that the role of state-led governance in fostering trust needs rethinking: trust in government alone does not carry over into trust in high-risk AI. Building confidence calls for governments to work hand-in-hand with trusted private sector partners, alongside regulation.
“Because disposition to trust is personality-based and cannot be manipulated by institutions, institutional trust is the channel through which governance earns trust,” Assoc Prof Taeihagh explains. “Cultivating it in low-trust societies can enhance trust in high-risk AI.”
Humanity Behind the Code
In other parts of the world, some governments are already trying such collaborative approaches. To illustrate, the Swiss public-private partnership framework leverages its collaborative technology ecosystem to drive responsible AI governance through multi-stakeholder initiatives like the Swiss Call for Trust and Transparency and the Swiss Digital Initiative. By uniting government agencies, leading academic institutions, civil society, and the private sector, Switzerland co-develops practical, interdisciplinary solutions to actively mitigate AI risks and foster global user trust.
We’ve been here before – nuclear energy and genetically modified crops were once hailed as engineering triumphs, only to stall for decades not because the science was flawed, but because the public did not trust the institutions deploying them. As the authors point out, technology design and science communication are not sufficient conditions to improve people’s trust when the perceived risks are deemed very high.
In Mercy, its AI judge actually failed when humans corrupted what they fed it. High-risk AI now stands at the same crossroads. The algorithm is only ever as trusted as the institution that deploys it, and that is a human relationship that must be earned in the real world.
Follow the Lee Kuan Yew School of Public Policy on our social channels to stay updated on the latest news, events, and announcements.