Public services, personal data and machine learning: prospects for infrastructures and ecosystems

There is a widespread belief that machine learning tools will improve decision-making in health and social care. Equally there are concerns that the new tools, used with large personal datasets, will jeopardise privacy and erode trust. We reflect on experiences gained in the course of the Quanticode research and development project in England. These suggest that the opportunities are real: it is possible to generate insights that are valued by health and social care planners. The concerns are also real, though, indicating that there is a need to address them, and to balance opportunities and risks. The terrain is also contested, with evidence of differences in values relating to the ownership of datasets in particular. We argue that developments in the governance of tools and datasets will be substantially shaped by the concerns and by debates over values.