Apple Inc. is embarking on a new strategy to enhance its artificial intelligence platform by analyzing data directly on customers’ devices. This approach aims to strike a balance between improving AI capabilities and upholding stringent user privacy standards, a critical area where Apple seeks to catch up with its AI rivals such as OpenAI and Alphabet Inc.

Traditionally, Apple has trained its AI models using synthetic data, which mimics real-world inputs without incorporating personal details. However, the limitations of synthetic data have become apparent, as it often fails to accurately represent actual customer data, thereby hindering the effectiveness of Apple's AI systems. The new methodology seeks to rectify this by processing user data on-device, ensuring that the data remains under the user's control and is not directly utilized to train AI models.

The core of this technology involves comparing Apple's synthetic data with recent samples of user emails within the iPhone, iPad, and Mac email applications. By cross-referencing the synthetic data against real-world messages, Apple aims to identify which elements of the synthetic dataset align most closely with actual user behavior. These insights will be instrumental in refining text-related features within Apple Intelligence, including notification summaries, Writing Tools for synthesizing thoughts, and user message recaps.

In a recent post on its machine learning blog, Apple stated, "When creating synthetic data, our goal is to produce synthetic sentences or emails that are similar enough in topic or style to the real thing to help improve our models for summarisation, but without Apple collecting emails from the device.” This underscores the company's commitment to enhancing AI functionalities without compromising user privacy.

Large language models (LLMs) are central to modern AI and power the features in Apple Intelligence, which debuted last year. Besides synthetic data, Apple has also leveraged licensed third-party information and data obtained from scanning the open internet to train its models. However, the over-reliance on synthetic data has led to certain shortcomings, such as misrepresentations in notifications and inaccuracies in text summaries. The new system holds the potential to significantly improve Apple’s AI models, a crucial step for the company to establish itself as a major player in the competitive AI landscape.

The company's artificial intelligence team has faced challenges, resulting in products that lag behind competitors, which prompted a recent management restructuring involving the Siri voice assistant and related projects. The new system is slated to roll out in an upcoming beta version of iOS and iPadOS 18.5 and macOS 15.5, with a second beta test already provided to developers.

Apple is also implementing privacy-focused strategies to enhance other Apple Intelligence features, including Image Playground, Image Wand, Memories Creation, and Visual Intelligence. The company has previously utilized differential privacy to improve its Genmoji feature, which allows users to create custom emojis. This system helps identify popular prompts and patterns while ensuring that unique or rare prompts remain undiscovered.

The company tracks how the model responds to common requests, such as generating an image of a dinosaur carrying a briefcase, and uses this data to improve results. These features are available only to users who opt-in to device analytics and product improvement capabilities, which can be managed in the Privacy and Security tab within the Settings app.

Apple noted, "Building on our many years of experience using techniques like differential privacy, as well as new techniques like synthetic data generation, we are able to improve Apple Intelligence features while protecting user privacy for users who opt in to the device analytics program.”

The artificial intelligence team at Apple has experienced turmoil in recent months, marked by organizational struggles, leadership issues, product delays, and executive changes. In March, Apple restructured its AI group, reassigning responsibility for Siri from executive John Giannandrea to Vision Pro creator Mike Rockwell and software chief Craig Federighi. While Apple plans to announce Apple Intelligence upgrades in June, the implementation of long-awaited features for Siri has been postponed until next year.