³Ô¹ÏºÚÁÏ

Problem

Few technologies have moved as quickly, or reached as deeply into everyday life, as artificial intelligence. Yet a purely technical account of AI misses what matters most: the human judgements, organisational practices and social consequences that determine whether these systems serve people well. Our research approaches AI as a human-centred phenomenon, examining not only what AI can do, but how it's imagined, built, deployed and held to account.

Science

Dr Ozum Demir Caliskan and her colleagues studied AI developers directly, using in-depth interviews to understand how they think about ethics while they're building. They found that developers imagine the downstream moral consequences of their ideas in two distinct ways: some set moral concerns aside and trust that safeguards will catch problems later, while others fold ethical thinking directly into the design process, letting it reshape the very ideas being developed. Professor Susan Scott at ³Ô¹ÏºÚÁÏ Business School and Professor Wanda Orlikowski at the MIT Sloan School of Management argue against treating AI as a fixed, finished 'thing', proposing instead a view of 'AI-in-the-making': the idea that AI's effects are produced through ongoing relations between people, data and machines, not fixed in the technology alone. And as part of a large research project with NHS breast-screening services, Dr Namrata Malhotra and her collaborators tested whether AI could take on a specific clinical role: acting as a second reader in the process where two doctors each check a scan independently before a third steps in if they disagree.

Outcome

The study found that moral engagement in AI design is itself a generative, constructive part of the work, not a brake on creativity and that the choices organisations make about how they use AI matter as much as the technology they choose. Dr Malhotra's team found that AI, used as a second reader in NHS breast cancer screening, can match the performance of two human readers working together. Even though human arbitration occasionally overruled correct AI recalls, these findings suggest that using AI as a second reader can help ease unsustainable staff workloads and, potentially, improve patient outcomes.

Change

Together, these findings make a clear case. Responsible AI depends on human insight at every stage: the moral imagination of creators, careful thinking by scholars about how we frame these tools, and rigorous evaluation before large-scale deployment. Our research shows that AI can ease real pressures, including the staffing shortages putting NHS breast-screening services under strain, but only when the human decisions around it are made carefully. Screening services, NHS England and the organisations building AI tools now have evidence to act on. 

Read more 

• Hagtvedt, L. P., Harvey, S., Demir-Caliskan, O., & Hagtvedt, H. (2025). 'Bright and dark imagining: how creators navigate moral consequences of developing ideas for artificial intelligence'. Academy of Management Journal.
• Scott, S. V., & Orlikowski, W. J. (2025). 'Exploring AI-in-the-making: sociomaterial genealogies of AI performativity'. Information and Organization.
• Kelly, C. J., et al. (2026). 'Diagnostic accuracy, fairness and clinical implementation of AI for breast cancer screening: results of multicenter retrospective and prospective technical feasibility studies'. Nature Cancer.
• Warren, L. M., et al. (2026). 'Impact of using artificial intelligence as a second reader in breast screening including arbitration'. Nature Cancer.