Trustworthy AI: UK Air Traffic Control Revisited

Trust in AI requires understanding real workplace practices: The study found that trustworthy AI cannot be designed in isolation from how people actually use tools daily. In air traffic control, this means AI systems must integrate with existing work practices rather than replace human judgment, as controllers rely on tacit knowledge and contextual understanding that formal requirements often miss.

Safety-critical domains demand human-centered design: Air traffic control revealed that trustworthy AI in high-stakes environments requires ongoing human oversight and the ability for workers to intervene. The research emphasizes that AI adoption in safety-critical settings must preserve human agency and decision-making authority rather than automating them away.

Socio-technical challenges are often overlooked in AI development: The paper highlights a significant gap in AI trustworthiness literature—most research focuses on technical requirements while ignoring organizational and social factors. The ethnographic findings suggest that adoption failures often stem from misalignment between AI system design and workplace culture, not technical limitations.

Building trust requires transparency about system limitations: The study indicates that air traffic controllers need to understand when and why AI systems make recommendations, and what their failure modes are. Trustworthy AI in safety-critical domains must communicate its confidence levels and constraints clearly to users who depend on it.