Federated Learning: Revolution or Evolution in Machine Learning? A Legal Analysis

Federated Learning (FL) is a form of machine learning (ML) where algorithms are trained across different data sets, without the sharing of these data sets. The promise of FL is that it enables the training of algorithms without infringing on privacy, even if an algorithm is trained on personal data. There is no lack of articles on FL in the computer science literature, but there is a dearth of writings on, and understanding of, FL among lawyers and legal academics. This article aims to fill the lacuna by providing a comprehensive legal analysis of FL. This article argues that FL should largely be considered outside the material scope of data protection regulation, like the EU’s GDPR or Singapore’s PDPA. However, it will also be argued that it is a mistake to see the potential of FL purely in its application to personal data. As FL may also provide significant benefits when applied to non-personal data. This article will also analyse how the regulatory frameworks for artificial intelligence apply to FL, arguing that they do not pose insurmountable obstacles to FL. Nevertheless, the objective to create ‘ethical’ AI may pose practical challenges for FL. The ultimate purpose of this study is to aid developers who are considering FL to appreciate the legal and regulatory implications of their product, as well to assist legal practitioners and policy makers to evaluate the impact and effects of FL with a view to formulating the appropriate legal response.