Apple Intelligence: A Different Approach to Generative AI
Among the biggest questions surrounding generative AI models like ChatGPT, Gemini, and Midjourney is how they will fit into our daily lives. Apple aims to answer this question with its own take on generative AI, Apple Intelligence. Unlike other companies that prioritize bigger models with a vast amount of information, Apple takes a more pragmatic approach. Apple Intelligence is built specifically for its operating systems, prioritizing a frictionless user experience.
To achieve this, Apple focuses on creating smaller models trained on customized datasets designed for specific functionalities required by its users. While this may limit the variety of prompts, it increases transparency as users can understand why the system makes specific decisions. The inclusion of “adapters” allows Apple Intelligence to cover a spectrum of requests while still considering factors like size, speed, and compute power.
Apple also opens up to third-party models like OpenAI’s ChatGPT and Google’s Gemini when its in-house models are out of scope. To address privacy concerns, Apple prompts users to opt-in or opt-out of using third-party platforms, providing transparency and control over data access. Additionally, Apple maintains privacy standards across its devices and servers, ensuring that queries are processed with the same level of privacy protection.
Certain operations are always performed on-device, such as Image Playground and text generation. Image Playground generates images in three different house styles, while text generation is available in friendly, professional, and concise styles. Apple’s models are trained on a combination of licensed datasets and publicly accessible information obtained through its web crawler, AppleBot.
Apple emphasizes responsible AI by empowering users with intelligent tools while avoiding perpetuating stereotypes and biases. The company takes precautions at every stage of the process to prevent misuse or potential harm and actively seeks user feedback to improve its AI tools. Privacy protection is paramount, utilizing on-device processing and Private Cloud Compute infrastructure without using users’ private personal data or interactions for training.
As Apple’s OS betas reach general availability, the company faces the challenge of balancing user experience and privacy. It aims to offer as much or as little information as each user requires, catering to both those who prioritize accuracy and efficiency and those who value transparency. While some aspects of the black box problem are unavoidable, Apple strives to provide transparency whenever possible, respecting users’ preferences and publishers’ content sourcing concerns.