Beyond the Hype: Thinking Critically about AI in Healthcare

Artificial Intelligence is often branded as the miracle cure for crumbling global healthcare systems, but what actually is AI and how can digital code care for analogue bodies? This lecture provides an introduction to the different techniques that sit under the umbrella heading of AI, explaining how each technique links to specific healthcare use cases. It traces how these technical capabilities have been translated into a powerful deterministic rhetoric promising hope to thousands, and argues that understanding what AI can and cannot do is the first step towards thinking critically about its role in healthcare, and distinguishing hope from hype. 

AI is championed as being key to empowering individuals to take better care of their own health.  This lecture examines the underpinning assumption that combining data from wearables and healthcare records to identify risk and provide personalised advice will automatically lead individuals to take better care of their health. It challenges this information deficit model by arguing that health is relational and systemic, not the consequence of individual choices made in isolation, demonstrating why we cannot code our way out of structural health inequalities. The lecture reveals how current AI applications often reinforce rather than resolve existing disparities, and proposes that we redirect our focus from individual behaviour change to addressing the structural determinants of health outcomes.

AI consistently performs better in the lab than it does in the clinic. This lecture explains why, examining the sociotechnical nature of healthcare information infrastructure and the complexities of the ‘last mile’ problem. It explores the technical foundations required for successful deployment, the hidden labour clinicians must perform to keep AI working, and the workflow integration challenges that emerge in practice. By critically analysing why the “plug and play” mentality is dangerous in complex adaptive systems, the lecture reveals that implementation failures are not merely technical glitches but symptoms of a deeper misalignment between how AI is designed and how healthcare actually operates. It argues for a more realistic approach to AI deployment that accounts for the messy realities of clinical practice and proposes design principles that centre on facilitating AI to fill information needs not information wants.

Healthcare is a heavily regulated industry, governed by policies, regulations, and ethical frameworks that were designed before AI existed. This lecture examines how existing governance mechanisms struggle to address AI’s unique characteristics, particularly its capacity for continuous change and emergent behaviour. Rather than cataloguing individual regulations, it identifies common themes challenging policymakers, regulators, and ethicists globally: the evidence standards problem, the accountability gap, and the tension between innovation and precaution. It proposes adaptive governance approaches that can evolve alongside the technology, emphasising the need for regulators to explicitly consider how their frameworks shape not just AI safety, but the very model of care AI enables. This sets the stage for understanding how governance choices directly influence the systemic transformations explored in the next lecture.

AI is not simply a tool within the existing healthcare system; it is changing the nature of the system itself. This lecture explores how current AI implementation represents second-order rather than first-order change, driving key shifts including from evidence-based medicine to algorithm-based medicine, from patient-centric to data-centric care, and from one-to-one relationships to many-to-many networks. It examines the changing doctor-patient relationship, the risks of automation bias, and the transformation of what constitutes medical knowledge. By critically analysing how AI is quietly restructuring healthcare practice, the lecture demonstrates that these shifts are not inevitable consequences of technology but reflect specific choices about what we optimise for and whose interests we serve. It argues that we must make these transformations explicit and subject them to democratic deliberation. Having traced how AI is reshaping healthcare in ways that often deepen existing problems, the series concludes by asking: what would it look like to build AI for a different vision of health?

The previous lectures have established that individual optimisation is the wrong goal for algorithmic healthcare. This lecture presents a manifesto for building AI that serves population health and collective flourishing instead. It redefines the purpose of healthcare AI—shifting from the mere absence of disease to the maximisation of the population’s capability to flourish, ensuring all individuals can work, play, and participate in their communities. The lecture reimagines AI as a ‘systems companion’ rather than a digital coach, explaining what this means both theoretically, and more practically,  offering a roadmap for moving from critique to construction.