Apple has announced a new approach to enhancing its artificial intelligence models that does not involve utilizing data from its users’ devices. In a recent blog post, highlighted by Bloomberg, the company shared how it plans to leverage devices to compare a generated synthetic dataset against emails or messages from users who have opted into the Device Analytics program.
The devices will identify which synthetic inputs most closely match the actual samples and communicate this to Apple by transmitting “only a signal indicating which of the variants is closest to the sampled data.” According to the company, this method ensures that user data remains private and does not leave the device. The frequently selected synthetic samples will then be employed to enhance Apple’s AI outputs, including features like email summaries.
Currently, Apple relies solely on synthetic data to train its AI models, which may lead to less effective responses, as noted by Bloomberg‘s Mark Gurman. The company has faced challenges in launching its flagship Apple Intelligence features, which included delays for some functionalities and changes in leadership within the Siri team.
In an effort to revitalize its AI initiatives, Apple is set to introduce this new training system in the beta versions of iOS and iPadOS 18.5, along with macOS 15.5, according to Gurman.
Apple has emphasized the use of differential privacy to safeguard user information since at least 2016, starting with the launch of iOS 10. This technology has previously been applied to enhance features like the AI-powered Genmoji and is now being incorporated into the firm’s new AI training strategies. By introducing randomization into a larger dataset, Apple aims to ensure that data cannot be traced back to any individual.
