Apple Intelligence's third-party hooks are the right direction for phone AI

Summary

It’d be great if the third-party AI hooks in Apple Intelligence extended to all phones.

There would be numerous benefits to privacy, personalization, and avoiding problematic platforms.

There’s no sign that anyone’s following Apple’s lead, however.

When Apple revealed Apple Intelligence at WWDC 2024, one of the things that raised some eyebrows was its plan to push some Siri requests to ChatGPT. It’s strange, on the surface, that a company worth over $3 trillion would need help with AI, especially when it’s famous for favoring in-house solutions. Sometimes whether it makes sense or not — as people led into the Australian outback by the 2012 version of Apple Maps can attest.

With the apparent confirmation that Google Gemini is coming to Apple Intelligence (via 9to5Mac), however, it occurs to me that third-party AI hooks are wise for a variety of reasons. Some of these are specific to Apple, but I feel like phone AI, in general, should move to a modular approach, mostly for the sake of us end users.

How I travel-proofed my iPhone.

Privacy barriers and Siri’s shortcomings

Physician, know thyself

Apple

Practically speaking, Apple probably didn’t have a choice if it wanted to hurry Apple Intelligence out the door for the iPhone 16. The company is rumored to be working on its own large language model (LLM) similar to ChatGPT, but that’s coming in 2026, at the earliest. Even then it’s unlikely to go toe-to-toe with rivals, which have enjoyed a lengthy headstart.

A modular approach also improves privacy.

ChatGPT gave Apple a way to add an LLM shine to Apple Intelligence without having to fix how Siri answers knowledge questions. On its own, Siri is infamously bad at answers, frequently prompting users to tap a web result link on their iPhone — even if they’re talking to a HomePod. Isn’t bypassing tedious web searches part of the point of a voice assistant?

A modular approach also improves privacy. In the way Apple is handling things, at least, add-on AI doesn’t get automatic access to everything on your phone. There are advantages to that — which is probably why Apple is working on an LLM — but we should be encouraging the idea of selective data handoff, certainly in an era when you can’t trust what every company is going to do with it.

Apple’s goal is worth pursuing, but the company may be a little overeager to get there.

Personalization and the monopoly on thought

Don’t trust any one algorithm

Midjourney / Pocket-lint

Already, some AI models are proving better than others at common tasks. While ChatGPT might be the best for knowledge questions, Gemini is ahead of Siri in smartphone and smart home control. When it comes to image generation, I’d rather turn to a dedicated tool like Midjourney.

When everything is handled by a single AI model, there’s the risk of users being funneled down a single path of thought by algorithms.

It’s probably not going to happen for competitive reasons (more on that later), but in a more user-friendly world, every phone would let us mix plug-in AIs instead of forcing us to use separate apps. Putting aside the issue of which AI is best at a given task, we all have personal preferences. I like Midjourney for AI art, but someone else might genuinely prefer DALL-E or Gemini, and they should have that option. Certainly, Apple should consider supporting image generator add-ons given how weak Image Playground is at the moment.

A greater concern is that when everything is handled by a single AI model, there’s the risk of users being funneled down a single path of thought by its algorithms. Consider that Grok briefly blocked results, saying that Elon Musk and Donald Trump spread misinformation (via Fortune). Even ignoring politics, any AI platform is going to reflect the priorities and biases of its creators — Apple sometimes plays it a little too family-friendly with content, which is one reason Image Playground isn’t that useful. The medium influences the message, to twist Marshall McLuhan a little bit, so stacking one platform against another can be important for opening up perspectives.