Table of Contents

• Overview of AI Training
• Apple’s Strategy to Enhance Its AI
• The Importance of This Development

Overview of AI Training

Before diving into the details, let’s briefly explore how AI tools function. The training phase involves supplying a vast amount of human-generated data to an “artificial brain.” This data may include books, articles, research papers, and more. The more data the AI processes, the more effectively it can generate relevant responses.

Chatbots, known as Large Language Models (LLMs), analyze patterns and relationships between words. Programs like ChatGPT, which are integrated into platforms such as Siri and Apple Intelligence, operate as advanced word predictors.

However, there are limitations to the amount of usable data available for training AI systems, making the process both lengthy and costly. One might wonder why AI-generated content isn’t utilized for this purpose. Unfortunately, research indicates that using synthetic data can skew AI models, resulting in inaccurate and misleading outputs.

Apple’s Strategy to Enhance Its AI

Rather than depending exclusively on synthetic data, a more effective way to enhance an AI tool’s performance is through continual refinement and tuning. The ideal method for training an AI assistant is to utilize authentic human data. The wealth of information stored on users’ devices represents a valuable resource, but accessing this data directly would raise significant privacy concerns and potential legal challenges.

To circumvent these issues, Apple is exploring a method of analyzing users’ emails indirectly. The data will never leave the user’s device, nor will it be transmitted to Apple’s servers. Instead of reading emails, Apple’s approach will involve comparing them to a collection of synthetic emails. The goal is to identify synthetic data that closely resembles human-written text, thereby gaining insights into natural conversation patterns.

Historically, Apple has relied heavily on synthetic data for AI training, as reported by Bloomberg. The company explains, “This synthetic data can then be used to test the quality of our models on more representative data and identify areas for improvement in features like summarization.” The anticipated outcome could lead to more accurate responses from Siri and Apple Intelligence.

Apple aims to enhance its email summarization features and other components within its Writing Tools using insights from this realistic human data. The company ensures that the sampled emails will not be transferred off devices or shared with Apple, highlighting a commitment to user privacy similar to those applied in the Genmoji system.

The Importance of This Development

Currently, the summaries provided by Apple Intelligence in Mail can be confusing and, at times, nonsensical. This issue has progressed to the point where Apple temporarily suspended its notification feature after it received negative feedback for misrepresenting news stories.

The inadequacies of the summaries have even become a source of humor among teams, as the AI frequently combines unrelated sentences that fail to convey the intended message.

The fundamental challenge is that AI continues to have difficulty understanding context and human intent. Improving this understanding necessitates training on data that is aware of specific situations and contexts. Although newer AI models designed for reasoning have been developed, they still do not fully address the issues at hand.

Apple’s proposed method appears to strike a perfect balance. The company states, "This process allows us to improve the topics and language of our synthetic emails, which helps us train our models to create better text outputs in features like email summaries while prioritizing privacy."

Importantly, Apple will not be accessing all emails on iPhones and Macs globally. Instead, it will adopt an opt-in approach, allowing only those users who consent to share Device Analytics data with Apple to participate in the AI training process. Users can enable this feature by navigating to Settings > Privacy & Security > Analytics & Improvements. Reports indicate that these initiatives will coincide with the release of upcoming updates for iOS 18.5, iPadOS 18.5, and macOS 15.5 in beta.