Deployment of medical AI and emergence of platform solutions

This thesis examines medical AI deployment as an ongoing process of experimentation and collective learning, not the implementation of a finished object. I study how AI vendors, healthcare institutions, professional service providers, and regulators work to make AI functional in clinical settings, and how these efforts, repeatedly encountering barriers, give rise to platform approaches. Methodologically, I develop a responsive approach that adapts to shifting strategies and problem definitions, combining landscape mapping, organisational vignettes, and a longitudinal case study. Four interconnected studies trace this process. The first examines early radiological AI deployments, where technical promise encountered organisational, regulatory, and local infrastructural barriers that catalysed demand for “platform” solutions. The second follows the trajectory of a platform provider, Plathub, that evolved from an imaging research group to a start-up to a deployment platform, before acquisition by a pharmaceutical multinational whose subsequent withdrawal underlies the sector’s volatility. The third analyses how different actors mobilise “platformisation” in overlapping but divergent ways. The fourth examines how one innovative hospital procured contracts with multiple platforms to create a novel platform ecology. These studies show that platform arrangements emerge through dispersed social learning: diverse actors borrow, imitate, adapt and mobilise one another’s strategies to manage deployment complexity, constituting a heterogeneous arena of competing interpretations and governance visions. A key finding is that making AI deployable depends on ongoing local infrastructuring work. Governance arrangements effective in one clinical context cannot simply be transferred to another. Yet, policy and regulation often treat “medical AI” as a stable and unitary category—a gap that generates momentum towards platform approaches. Platformisation itself, however, becomes multiple: vendors, institutions, and intermediaries enact different platform ontologies and accountability boundaries. Rather than resolving disorder, platforms create new sites of negotiation over what the system is, where responsibilities lie, and how clinical risk is governed. Contrary to narratives of planful disruption and eventual dominance, healthcare institutions retain power through procurement leverage, standards control, and existing systems. In a key case, a hospital created a “platform of platforms” that actively orchestrating multiple platform relationships to promote competitive supply and limit dependency on any single arrangement. This thesis contributes an STS account of how actors collectively learn to deploy emergent technologies, foregrounding infrastructuring work, contingency, and social learning over strategic execution. Conceptually, it bridges development arena and emerging ecology perspectives, tracing how sense-making unfolds locally while following how interpretations stabilise, travel, and acquire authority across the wider ecology. Methodologically, I develop a reflexive, responsive design capable of tracking shifting practitioner understandings in real time. For policymakers and practitioners under decision-making uncertainty: deployment is neither a discrete phase nor a purely technical problem, but an ongoing accomplishment requiring sustained investment in coordination infrastructure, standards work, and mechanisms for collective learning.