How Do We Build Trust in AI Applications? Human Centric Design

The UK government is creating extraordinary opportunities to accelerate the adoption of artificial intelligence (AI) across public services. As the current Government’s AI action plan states, “AI should become core to how we think about delivering services, transforming citizens’ experiences, and improving productivity”. Fully realising this potential, however, requires revisiting a long-standing debate: the role of quantitative versus qualitative research in understanding human experiences. Accordingly, this blog summarises a review of government policy to explain the importance of qualitative research for the following three areas of human-centric AI:

• Conducting qualitative research about human experiences

• Qualitative Approaches to Evaluating AI Systems

• Qualitative Insights into Public Trust of AI Systems

Why Human-Centric AI Requires Qualitative Research 

Human-centric AI is about designing technology around people, whether they are the users or subjects of analysis. It answers the problem of “how to make technology work for people, rather than making people work in ways that put technology first”. The Ministry of Defence explicitly prioritises human-centricity as an ethical principle of its AI policy to ensure “AI systems enhance human well-being and do not introduce unintended harms”. While other government policies imply this requirement, providing government services is an inherently human-centric endeavour. This principle of human-centricity is, therefore, fundamentally relevant across all aspects of government and essential to maintaining the legitimacy of public services.

Sustaining the public’s trust in government requires integrating qualitative research into the development and assessment of human-centric AI. Qualitative research has long been critical in uncovering the values, beliefs, and motivations that shape human behaviour. Nonetheless, a long-standing tension between qualitative and quantitative research has been a persistent issue. In the 1940s and 1950s, quantitative methods—focused on statistical analysis and large-scale surveys—dominated the social sciences. A shift began in the 1960s and 1970s with the emergence of interpretive and ethnographic approaches, focusing on lived experience and social context. These resulting methodological tensions, known as the “paradigm wars”, gave rise to mixed-methods research, championed by scholars like John Creswell, synthesising quantitative scale with qualitative depth.

Revisiting this synthesis of qualitative and quantitative research is essential for designing AI systems that reflect the complexity of human experiences. Current designs predominantly rely on quantitative methods, such as statistical models and large datasets, to identify patterns, automate processes, and assess AI system performance. While this quantitative foundation enables measurable insights at a massive scale, it risks sidelining what truly matters in providing public services: the meaning, values, intentions, and behaviour of people and populations. Paradoxically, as AI becomes more "sophisticated," its reliance on purely quantitative approaches replicates regressive research paradigms of the 1940s.

Conducting Qualitative Research About Human Experiences

The value of qualitative research in delivering effective policy-making is well established. As DSTL explains, “many qualitative methods are more appropriate for understanding some of the human-centred approaches to decision-making”. Ethnographic methods, in particular, have proven essential for capturing the complexity of human experiences and translating them into meaningful insights. By embedding themselves in real-world contexts, policymakers gain a deeper understanding of the people and populations they serve. In national security, activities such as undercover work, intelligence gathering, and agent handling are extreme, dangerous, and highly coveted forms of qualitative inquiry. A common thread across these diverse applications is the generation and analysis of text-based reports, which serve as critical artefacts for understanding human behaviour.

Today, advances in AI and the availability of vast digital text sources offer new opportunities to scale qualitative research. The internet provides a rich source of text data for conducting online ethnography to inform the design of public services. Other opportunities lie in narrative analysis to detect problems like hate speech, disinformation, and online radicalisation. The increasing digitalisation of government services also presents opportunities to unlock otherwise hidden insights in text-based reports. Embedding qualitative approaches into AI systems ensures that technological advancements enhance, not bypass, our understanding of the people governments seek to serve.

Qualitative Approaches to Evaluating AI Systems

In addition to understanding people, qualitative research is crucial for evaluating AI systems themselves. Traditional metrics - such as accuracy, precision, and F1 score - are useful for assessing technical performance, but they fall short in capturing how systems behave in real-world, socially complex environments. These metrics cannot explain how AI systems handle the complexity, ambiguity or nuance of understanding human experience.

Going beyond F1 scores requires understanding how the system responds to human-machine interactions of researching human experiences. The UK Government’s Magenta Book provides guidance on government evaluation and offers qualitative approaches for conducting evaluation, such as ethnographic research, in-depth interviews, and structured interviews. These approaches help uncover unintended consequences, systemic biases, and context-specific failures that quantitative testing often misses. Combining both approaches provides a more comprehensive understanding of how AI systems operate—and, crucially, how they can be enhanced to ensure safety, fairness, and reliability in public-facing applications.

Qualitative Insights into Public Trust of AI Systems

Many people are scared of AI. Public trust in the government's use of AI cannot be assumed; it must be earned and continuously evaluated to meet the ambition for the widespread adoption of AI. The current Government’s AI action plan explains that the “Government must protect UK citizens from the most significant risks presented by AI and foster public trust in the technology, particularly considering the interests of marginalised groups”. The previous government’s AI strategy employed the term “qualitative intelligence” to inform the development of AI systems and promote corresponding public trust.

Building and assessing trust requires more than metrics; qualitative approaches capture the nuanced social factors that influence public attitudes. These approaches can help policymakers understand the expectations, concerns, and values that people associate with technology use in public services. When embedded into the development and deployment of AI, these findings inform systems that are more transparent, accountable, and aligned with public interest. The result is not only more effective AI but also faster, more confident adoption, anchored in public trust.

Conclusion: Embedding Human-Centricity into the Future of Government AI

As the UK government accelerates the adoption of AI across public services, the path forward must be guided by a commitment to human-centric design. This means recognising that the success of AI does not rest solely on technical performance, but on how well these systems reflect, respond to, and respect the complexities of human experience. By embedding qualitative approaches alongside quantitative methods into the development, testing, and evaluation of AI, policymakers can ensure that these technologies earn public trust, uphold democratic values, and enhance the legitimacy of government action. The opportunity is not just to deploy more intelligent systems but also to build more responsive, inclusive, and ethical public services that fit the digital age.

References

• UK Gov (2025) AI Opportunities Action Plan

• MOD (2024) JSP 936: Dependable Artificial Intelligence (AI) in defence (part 1: directive)

• DSTL (2023) Human-centred ways of working with AI in intelligence analysis

• UK Government (2020) Magenta Book: Central Government guidance on evaluation