Apple's approach to AI: observations from the World Wide Developers Conference 2024

Apple's approach to AI is unique in the market. It's not a product you use directly, like an app. Rather, it's a service, infused throughout its products, that helps users get things done.?

Caveats: I'm someone who uses and loves Apple products; but not uncritically so. I use AI in my work and life, and regularly run up against its obvious flaws and limitations. I'm intensely interested in privacy, not as a promise or an email of a legal update, but as an architecture, something I can understand and negotiate. Against this background, I'm excited for Apple's foray into the AI market.?

Consider the leading brand approaches to AIs. OpenAI's ChatGPT arguably transformed our thinking about what AIs can do and brought it to the masses through its easy-to-use chat interface. Microsoft invested billions in OpenAI to create Copilot, its branded LLM based on ChatGPT. But I can't help but think of old Clippy when I use Copilot: it's a button in Microsoft's various apps, that summons up a context-limited assistant that is as often wrong as helpful. Amazon approached its AI-driven digital assistants, embodied as apps behind Alexa, as voice app extensions for brands. Users would (could? Didn't!) download a brand's voice app to interact with it. A few were very successful, and it's a fine platform to set a kitchen timer so as not to burn the stovetop caramel. Few would argue it moved much past that, or timely announcements that the Amazon delivery has arrived in the garage. Google, empowered by its unparalleled knowledge graph and massive index of the internet, is making attempts to answer everyday questions — but it has been dramatically hamstrung by its addiction to ad revenue, as well as reality-warping internal politics.?

In the midst of all this, Apple unveils its approach to AI at WWDC24 this June, and starts by rebranding it as "Apple Intelligence." Its vision??An assistant to help users get things done: sorting emails and photos, surfacing key notifications, helping find files, and making summaries. Apple Intelligence has secure, private access to the user's "Personal Context": where you are, what's on your calendar, where you're headed, who you communicate with, what's on your list of things to do. By making secure, private use of this Personal Context, Apple's vision is AI, infused throughout, that helps users get things done. So users will move from "What's the weather today?" to "When is mom's flight landing?" or "How much time do I need to get to my afternoon appointment?" and expect reasonable answers.?

What's the privacy promise? More than just words. The AI architecture uses on-device processing whenever possible — with on-device, data is processed locally and never leaves the phone. When needed, additional external resources are available via encrypted access, which Apple has branded Private Cloud Compute. Of note is more than the promise, but the ability to go look and see: the proof, the architecture, is public, and Apple is welcoming eyes upon it.?

All of this is integrated, invisible to the user, and not a "thing" or an app that you work with directly except, notably, access to ChatGPT.?Here, Apple created a user experience that will train users over time: a "hey you're leaving Apple" interface that has become familiar on secure websites, directing users elsewhere (and, by implication, potentially less secure.?Apple is building, in the minds of its customers, a notion of "what are your capabilities, right here" versus "what must be handled externally". ChatGPT happens to be today's external LLM — it could be anything else in the future.