Decentralising AI: Can Blockchain Technology Solve Data Privacy Challenges in Machine Learning?
摘要
Blockchain has been charged as a solution looking for a problem. Could machine learning be that problem? Machine learningMachine Learning (ML) (“ML”) is a subset of Artificial IntelligenceArtificial Intelligence (AI) (“AI”) that covers machines which process and identify patterns in data to make predictions. Machine learning powers an increasing number of applications involving data and predictive analytics, across many industries including healthcare and finance. Generally, machine learningMachine Learning (ML) relies on a single system or server to receive data and perform all computational tasks required to train the ML modelML model. This centralised approach to training of a ML modelML model (or “ML training”), raises significant data security and privacy concerns. In an attempt to resolve these challenges, alternate forms of machine learningMachine Learning (ML), namely federated learning and distributed learning, are being considered. These alternative approaches divide the training responsibilities, and reduce data flows and communication, amongst cooperative participants. However, extensive literature identifies that data security and privacyPrivacy challenges persist, and new challenges relating to “trust” arise between trainersTrainers and the central server. In the context of federated learning, these trust challenges are manifested in ensuring active and honest participation in the ML training. Without proper incentives, encouraging participation in the federated system and providing guarantees to the quality of the siloed ML training is difficult to ensure, particularly where ML modelsML model are trained by separate and distinct entities (as compared to different computers or devices controlled by the same entity). BlockchainBlockchain technology, which enables enhanced security, particularly through the immutability and traceability of data, may offer a solution to these critical privacy issues. Significantly, a blockchainBlockchain-enabled solution may also leverage smart contracts, the consensus mechanism and crypto assets to offer opportunities to provide economic and non-economic incentives for ML trainersTrainers. Ultimately, this chapter analyses the technical and legal implications of this blockchain-ML integration, in particular its efficacy as a solution to the privacy challenges in ML training.