Why Trust Must Be Built Into AI from the Start
A recent Applause survey, conducted between January and March 2025, found that 65% of users had encountered problems with AI applications, including bias, hallucinations and factual errors. These findings highlight the shortcomings of traditional software testing, which relies on deterministic unit tests unsuited to the probabilistic nature of AI systems. Unlike conventional code, AI interacts with people who evaluate it not only on functionality but on fairness, transparency and reliability.
Industry data reveals a consistent mismatch between how AI models perform in controlled tests and how users experience them across different demographics. A recommendation engine might achieve 95% accuracy in testing but still produce skewed results for specific groups—undermining both user trust and commercial success.
To close this gap, developers must adopt human-centred testing from the outset. This involves building test communities that reflect the diversity of real users—across age, ethnicity, language and accessibility needs. Internal or homogenous testing teams are unlikely to catch bias or fairness issues. Frequent testing cycles using diverse human evaluators, and tracking quality across demographic segments rather than aggregate scores, offer a clearer view of real-world performance.
Bias in AI is not a one-off problem. Models that adapt to new data can develop fresh biases, meaning continuous oversight is required. Transparent AI systems—particularly in sensitive sectors such as healthcare, finance and recruitment—must provide users with clear, understandable explanations for their decisions. These explainability features should be tested with real users to ensure clarity, not just technical correctness.
Robust, inclusive feedback mechanisms are another key component. Basic star ratings or thumbs-up/down tools are too blunt to capture nuanced experiences. Developers should use qualitative methods, such as interviews and focus groups, to understand how trust develops or erodes over time. Feedback channels must accommodate different communication styles, from quick surveys to in-depth discussions.
Environmental context also shapes AI behaviour. Voice assistants, for example, may work well in quiet offices but struggle in noisy public settings. Cultural norms further influence how AI content is interpreted. These realities underscore the need for contextual testing in varied, real-world conditions through partnerships with diverse user communities.
Monitoring trust must go beyond conventional metrics like uptime. Developers should track KPIs that reflect user experience, such as confidence scores, bias detection rates and explanation clarity. These metrics must inform ongoing model updates, ensuring feedback directly shapes system improvements.
Honest communication about AI limitations also plays a crucial role. When users understand what a system can and cannot do, they are more likely to respond with patience and trust. Premature launches followed by reactive fixes risk damaging reputations and user confidence.
Longitudinal data backs these concerns. Applause surveys from 2023 to 2025 show persistent worries about AI bias and inaccuracy, even as generative AI adoption increases. In 2023, 90% of users were concerned about biased content; by 2024, half still reported bias and over a third cited errors in chatbot responses. A 2025 study by the European Broadcasting Union and the BBC found that 45% of replies from major AI assistants contained major errors when reporting news, raising serious questions about accountability.
Despite rising AI investment, essential quality assurance often lags behind. Another 2025 Applause survey exposed a stark disconnect between AI deployment and the adoption of rigorous testing in software development lifecycles.
To secure the UK’s leadership in responsible AI, developers must embed user-centred testing at every stage. This means auditing current processes for diversity and fairness, integrating real human evaluations, creating inclusive feedback systems and defining trust-specific metrics. These changes do more than improve performance—they help earn lasting user confidence.
In the rapidly advancing world of AI, trust is not a byproduct of functionality but a product of design. By making trust a foundational element of development, the UK can lead the way in building AI systems that are not only powerful but equitable, reliable and sustainable.
Created by Amplify: AI-augmented, human-curated content.
