Apple hopes to fix misfiring AI by letting it learn from your emails

Table of Contents

A brief summary of AI training

How is Apple planning to fix its AI?

Why is it a crucial step forward?

Apple’s AI efforts have not been as successful as competitors like Google’s Gemini, Microsoft Copilot, or OpenAI’s ChatGPT. The company’s AI stack, Apple Intelligence, has not significantly improved iPhone and Mac user experiences, leading to internal management issues.

To address this, Apple has published a Machine Learning research paper outlining a new approach to training its onboard AI using data stored on iPhones, starting with emails to enhance features like email summarization and Writing Tools.

A brief summary of AI training

Before delving into specifics, it’s important to understand how AI tools function. Training involves feeding a large amount of human-created data to an artificial brain, such as books, articles, and research papers, to improve responses. Large Language Models (LLMs) like ChatGPT, integrated into Siri and Apple Intelligence, analyze patterns and relationships between words to predict text.

However, training AI with limited data is time-consuming and costly. Using AI-generated data can “poison” models, leading to inaccurate responses. Apple aims to refine its AI responses by indirectly accessing user data on iPhones without compromising privacy.

How is Apple planning to fix its AI?

Instead of relying solely on synthetic data, Apple plans to improve AI responses by analyzing human data stored on iPhones. By comparing user emails to synthetic data, Apple can identify realistic conversation patterns for AI training. This approach aims to enhance email summarization and Writing Tools features, without compromising user privacy.

Apple assures that sampled emails never leave the device and are not shared with the company. Users can opt-in to share Device Analytics data with Apple for AI training, ensuring privacy protection.

Why is it a crucial step forward?

Current Apple Intelligence features like email summarization often provide confusing or inaccurate responses. By training AI with realistic human data, Apple aims to improve contextual understanding and response accuracy. The upcoming iOS 18.5, iPad 18.5, and macOS 15.5 beta updates will implement this new AI training approach, with an opt-in feature for users to participate.

By integrating user data for AI training, Apple hopes to enhance Siri and Apple Intelligence features, providing users with more accurate and contextually relevant responses.