Decentralized Data Markets

Abstract

Acquiring access to massive amounts of data has become fundamental to state-of-the-art artificial intelligence systems. However, as data value increases, data owners have challenged current norms and practices of data acquisition. Data marketplaces have been promoted to fairly compensate data producers and incentivize greater data sharing. In this thesis, I describe a decentralized model of data markets to overcome privacy concerns in siloed, data-limited domains such as healthcare. I propose two federated techniques to automatically select a subset of data sellers and datapoints for a buyer given some sample data. I also examine the socio-technical implications of emerging data markets for medical data and synthesize ethical principles for medical data marketplaces. Decentralized data markets have the potential to enable new AI economies through more robust, transparent, and participatory data sharing platforms. Through the contributions in this thesis, I hope to make a positive step towards realizing a future where transformative data-enabled technologies such as general-purpose machine learning systems are developed more responsibly and the benefits are distributed more equitably.