Apple Will Read Users’ Conversations to Improve Its AI
Apple plans to start analyzing data on users’ devices to improve its artificial intelligence platform.
Apple currently trains its models, including those that power Apple Intelligence, primarily using synthetic data — information that mimics real-world scenarios but doesn’t include any personal data from users. However, the company acknowledges that such synthetic data doesn’t always accurately reflect users’ real behavior and information. This can cause Apple’s AI algorithms to perform worse on certain tasks, such as text summarization. In addition to synthetic data, Apple also uses information licensed from third parties and data from open sources on the internet.
The new approach will address this issue while ensuring that user data remains on customer devices and is not directly used to train AI models.
The technology works like this: An algorithm takes synthetic data generated by Apple and compares it to a live sample of user emails in the company’s Mail app for iPhone, iPad, and Mac. By using real emails to check the accuracy of the generated synthetic data, Apple will have a better understanding of which parts of the synthetic data are most consistent with real messages, and therefore suitable for training AI models.
