Google proposes ‘federated learning’ to address privacy concerns in AI solutions
Artificial intelligence (AI) algorithms used by Internet companies to develop applications that can read emotions or auto-complete sentences rely on the processing of user data. In most cases, these algorithms involve machine learning approaches which require the centralisation of ‘training data’ in a data centre, thus raising privacy concerns. Google says it can address such concerns via ‘federated learning’. As explained by the company, ‘federated learning enables mobile phones to collaboratively learn a shared prediction model while keeping all the training data on device, decoupling the ability to do machine learning from the need to store the data in the cloud’. In practice, this means that an AI training model is downloaded on the user’s device, where it learns the users’ patterns, and summarises them in encrypted ‘updates’ which are then sent to a Google training server. These updates do not include raw data about user’s habits, but only describe what the AI has learnt. According to Google, ‘Federated Learning allows for smarter models, lower latency, and less power consumption, all while ensuring privacy’.