Apple has announced a groundbreaking approach to enhance its artificial intelligence models without relying on user data from iPhones and Macs. In a blog post highlighted by Bloomberg, the company detailed its innovative strategy to utilize a synthetic dataset and allow devices to compare this data against recent emails or messages from users who have consented to its Device Analytics program.
This method enables Apple devices to identify which synthetic inputs most closely resemble real user samples, sending only a signal to the company indicating the closest match. This approach ensures that user data remains secure and does not leave the device, thereby maintaining privacy. By analyzing the most frequently selected synthetic samples, Apple intends to improve the outputs of its AI, like summarizing emails more effectively.
Currently, Apple’s AI relies solely on synthetic data for training, which may lead to less relevant or effective responses. According to Bloomberg’s Mark Gurman, the company has faced challenges with its flagship Apple Intelligence features, delaying the release of certain capabilities and changing leadership within its Siri team. However, the introduction of this new training system, set for beta testing in the upcoming iOS and iPadOS 18.5 and macOS 15.5, represents a significant turnaround in its strategy.
Since launching iOS 10 in 2016, Apple has emphasized its commitment to privacy through a technique called differential privacy. This method introduces randomized elements into broader datasets, ensuring that individual users cannot be identified from the data.
Apple’s new AI training strategy continues to uphold this principle of privacy while seeking to enhance the functionality and responsiveness of its AI features, including the recently improved Genmoji feature. With these developments, Apple aims to bolster its AI capabilities while reinforcing its dedication to user privacy.
