We all knew Apple was going to unveil its AI strategy this week at its Worldwide Developers Conference. What we didn’t know was how Apple was going to square the most confounding circle of AI, which is how to deal with its inherent weaknesses: reliance on big, cloud-based models, the fact that it knows nothing about you, and those pesky instances where it ‘hallucinates’ incorrect facts.
Like an expert marksman, Apple targeted all of those weak spots with a strategy under the umbrella label of Apple Intelligence (which — how about that? — can be shortened to “AI”). If I had to characterize Apple’s big AI push in a word, it would be “comprehensive.” Apple may be calling it a “beta,” but it’s definitely not a hobby: Apple’s version of AI will permeate the entire user experience on iPhones, iPads and Macs.
We’ve still got a few spots left in our June AI Fundamentals class, which will be laser-focused on tools and techniques for PR work, happening June 20. We’ve curated a custom toolset of apps specifically crafted to speed up PR writing, pitching, and creating on-brand media (think: articles, images and videos).
If you’re new to AI, or if you’ve plateaued in your use of it in your work, this class will take you to the next level. This PR class is happening one time only, so don’t wait! Click the button below to sign up:
It starts with Siri. Apple’s assistant has rightly taken a lot of flak over the years for never really delivering on the idea of an ever-present digital assistant. To be fair, no one else did either — Amazon’s Alexa and Microsoft* Cortana didn’t get much closer. But Apple stuck with Siri, striking the right balance with it being present enough to help with basics (timers, weather) but not interesting itself where it clearly wasn’t useful (pretty much everything else).
Siri’s AI Upgrade
Empowered by AI, though, Siri becomes something different: a bridge between apps that can now infer meaning in the messages and files on your device, understand your commands, and start to really handle specific tasks for you without you needing to find, tap, and swipe through individual apps. In AI parlance, Siri is taking its first steps toward becoming an agent.
For an AI agent to be truly effective, however, it needs to understand who you are. Not just the context of a request, but your preferences, habits, connections, schedule, and everything else that might affect how you’ll react to a specific response to that request. We’re talking way beyond ChatGPT custom instructions.
In a word, an agent needs to be personalized. And Apple has historically owned the lockbox where we put all our personal information: the iPhone. If only AI could access all of that information in a way that is secure and respects user privacy, every iPhone owner could have their own AI agent, with a full understanding of not just what you want, but who you are.
That’s clearly Apple’s goal, and it’s ambitious. It also leans into Apple’s strengths: It achieved this level of trust with consumers not by saying “you can trust us,” but that “you can’t trust anyone.” Generally, Apple’s approach to privacy has been to lock away your personal data not just from the world, but even Apple itself. It has historically been a leader in deprecating tracking technologies on its devices and services, going so far as to make “burner accounts” an iPhone feature.
Offloading the Blame for Hallucinations
With respect to AI, though, Apple can’t do it all. That’s why it’s partnering with OpenAI for when users ask Siri general knowledge questions. In those cases, Siri will get an answer from ChatGPT after it checks with the user that it’s OK. No ChatGPT account required, and Apple says queries won’t be harvested by OpenAI for training data or any other purpose.
While some are saying the OpenAI partnership is evidence Apple isn’t very good at making its own models — and there may be some truth to that — it also happens to be a very smart move from a public-relations standpoint.
Follow: Apple Intelligence is mostly concerned with app actions (e.g. figuring out when to pick up your mother from the airport and where to go to lunch with her), rewriting (shortening a long Note), and interpreting (summarizing the content of emails in Mail). In other words, language tasks: cases where an AI is working with existing information, not going into its training data to find an answer.
For knowledge tasks — when, say, you urgently need to know LeBron James’ average vertical — Siri turns to ChatGPT, which gives Apple cover when things (inevitably) go wrong. In those instances, any glue-on-pizza hallucinations can simply be blamed on OpenAI. Apple won’t even have to actively ascribe that blame; the mere fact that Siri says “Working with ChatGPT” will create that association in the user’s mind.
It’s a savvy approach, and Apple took pains to note that the partnership is not exclusive. That’s no doubt to end-run any suggestion of colluding with other Big Tech players to pick winners — which tends to make government regulators very interested — but it also gives Apple a lot of wiggle room for Siri to eventually pick from a menu of models, and perhaps in time replace them all with its own.
Defining Mainstream AI
We now see Apple’s vision for consumer AI very clearly, and it’s no small thing. This is not just “slapping ChatGPT on the iPhone” and seeing what happens. There are still many ways it can go wrong, but by offloading knowledge queries and emphasizing privacy, Apple has shown it can be ambitious and thoughtful at the same time. The company may not end up redefining what “AI” stands for, but Apple just gave itself the best chance of defining how AI is actually used by the masses.
*Disclosure: Media Copilot staff have done consulting work for Microsoft.
Leave a Reply