Compounding advantage: Medical AI and structural injustice

Artificial intelligence in clinical medicine is widely presented as a democratising advance, bringing the interpretive capacity of leading institutions to those that lack it. This paper argues that, under prevailing market governance, AI does not democratise but compounds: it operates as a self-amplifying mechanism that converts each increment of institutional advantage into the means of acquiring the next. The distinctive injustice this produces is not a snapshot maldistribution of a beneficial resource but a dynamic one - a ratchet. Standard distributive principles capture it only awkwardly. A Rawlsian analysis, and Daniels's extension of fair equality of opportunity to health care, can condemn any given unequal distribution, but they are framed to assess states of affairs rather than the self-reinforcing processes that generate them. Iris Marion Young's account of structural injustice is the more adequate lens, for the compounding mechanism has just the features her account describes: large-scale group disadvantage produced by the aggregation of individually reasonable actions, under accepted institutions, with no culpable author. The medical-AI case also sharpens her account, displaying a feature it does not emphasise - that the engine of the injustice is the distribution of a genuine good, so that the structure deepens disadvantage faster precisely as the technology improves. A second channel, in which tools are less reliable for under-represented populations because training data accumulate where infrastructure already concentrates, is shown to be the same mechanism operating on quality rather than access. Normative implications follow for the dynamics of diffusion, and professional obligation is grounded in Young's social-connection model of responsibility.