The Role of Government as a Provider of Data for Artificial Intelligence

Access to government data has the potential to be a catalyst for AI development. To achieve this, a multi-faceted approach that enhances data-sharing is necessary. Progressive models for data sharing are central to the promotion of responsible AI including equitable access and transparency in use of government data. As governments embrace the use of AI as the case studies in this report show, positive use cases are needed to further encourage the notion of AI for public good. Various data sharing models are emerging to enable the supply of data by governments to AI developers. These models include contracts, open data, data stewardships, and public-private partnerships. Key issues related to these models for data sharing include the legal basis for the sharing of data by governments, compliance with data processing principles such as purpose limitation and data minimisation in disclosure. Four case studies are explored in this report to understand current models of data sharing by the government with the private sector. We assess the objectives of data sharing by governments and the mechanisms for data sharing with the aim of identifying benefits, risks and cross-cutting recommendations that can advance the goal of data sharing by governments for AI development. In the UK, the NHS and DeepMind entered into a data sharing agreement for DeepMind to develop an app to quicken the diagnosis of acute kidney injuries. Through an information sharing agreement, the NHS released the personal health information of 1.6 million patients to DeepMind. The parties did not follow data governance safeguards in place and the public objective of improving diagnosis faced a public backlash and sanctions from oversight institutions when the project became public knowledge. Some of the issues arising from the case study include the legal basis for NHS to share data with DeepMind, data minimization safeguards, mechanisms in place for data subjects to exercise their rights and transparency in AI design, testing and use. This case study shows the need for evolved understanding of data governance and oversight institutions with strong enforcement powers. In Taiwan, the government introduced an app, the Health Passbook, where users of the app can select which third party providers can access their personal health information to provide customised healthcare services. The government does not directly share data with third parties in this case but serves as a facilitator by verifying third parties in line with data governance principles before they become eligible to receive data directly from data subjects. This case study provides an alternative for governments who are risk averse to play a different role in how data is shared for the development of AI.