Apple CEO Tim Cook is in a big mood. He has just returned from dinner in Washington with four Republican lawmakers at Central, a neo-French brasserie near Capitol Hill.
Cook has a methodical, no-nonsense style that’s different from that of his predecessor. His subordinates watch for the slightest change in his rocking chair pace.
1. It’s too early
During Monday’s WWDC keynote, Apple unveiled a number of new features for iPhones, iPads, and Macs. But it was the company’s AI advancement—called simply “Apple Intelligence”—that caught the most attention from tech observers.
The big question is whether the company’s version of AI will be as good as its competitors’. That’s a tough one to answer: Despite all the hype, AI is still in its early innings. And it’s not even clear if it will ever be as good as humans. But that’s not stopping the likes of Google, Microsoft, and Facebook from trumpeting their AI prowess in the hopes of luring users away from their services.
During an interview with The Washington Post, Apple CEO Tim Cook acknowledged that Apple Intelligence won’t be error-free. The company will do everything it can to prevent AI hallucinations—or the occurrence of a system that produces incorrect or misleading information with confidence enough that other systems accept it as valid—but it can’t guarantee the accuracy of its answers all the time.
And that’s okay. As long as Apple Intelligence isn’t trying to take over the world, it can probably get by without getting too many things wrong. But if it starts making weird emoji suggestions, or advising you to wear blue instead of green, that may be a sign that the technology has gone rogue and needs to be shut down.
2. It’s too late
Apple is a latecomer to the AI party, but it wants to make sure it gets on the train in a way that will be beneficial to users. That means avoiding AI hallucinations and ensuring that the integrations are secure, well-documented, and have guardrails to prevent mistakes.
It also means focusing on the immediate benefits of AI, such as saving time and making things more efficient. It also means taking the time to test integrations and ensure that they meet high standards before releasing them to users. This was evident in the WWDC keynote, where Apple’s revamped Siri, writing features across the system, and GenAI-powered emoji all came with a warning to check them for mistakes, reminding users that generative AI is still prone to producing nonsense results — the kind of inaccuracies referred to by tech billionaire Elon Musk as AI hallucinations.
The most important thing, though, is that Apple is implementing these new AI capabilities in ways that are both secure and private. Its on-device AI models are tiny, and they’re trained with 3 billion parameters. That’s more than most rivals’ models, but it’s nowhere near the amount of data that Google Gemini or Facebook DeepMind can handle.
As a result, users on older iPhones and iPads will not be able to use some of these features until the fall when iOS 18 rolls out along with updates for macOS, WatchOS, and VisionOS. The good news is that Apple will provide a workaround so people can continue to use Siri and its other new integrations.
3. It’s too complex
CUPERTINO — If you’re looking for new AI features that will propel Apple ahead of market leaders Microsoft and Google, you may be disappointed by the company’s Monday event. Wall Street was lukewarm on the announcements, with the company’s stock closing down nearly 2%.
Apple’s big promise is that its Apple Intelligence software will understand context better than rivals’ offerings. It’s deeply integrated into iOS, iPad and macOS Sequoia, and can use generative models to create text, images and take action within apps. It also uses your personal data to understand what’s important to you and simplify and accelerate everyday tasks.
The software can, for instance, gather a variety of factors to help you decide whether you should go to a school play or a work meeting. It can also take into account traffic, your schedule and who you’re meeting with. It can even suggest an alternate meeting time or reschedule the one you have to attend, Apple said.
To do all that, Apple Intelligence relies on a variety of small models that run on the chip in your phone or tablet. And if the on-device model can’t answer your question, it’ll reach out to a new ultra-secure cloud that’s designed to keep your personal information private. Apple is betting that this is a crucial distinction it can use to separate itself from the competition, though analysts say privacy concerns will likely remain high.
4. It’s too expensive
Apple doesn’t plan to let a few bad AI headlines scare away customers. It’s going to focus on deploying generative AI in a safe and secure way. That may not sound like a bold bet for investors, but it’s the path that makes sense for Apple, says Rensselaer Polytechnic Institute computer scientist Jim Hendler.
During Monday’s WWDC keynote, Apple’s software chief Craig Federighi made clear that Apple Intelligence features won’t feel creepy. For one thing, they won’t rely on “cloud intelligence,” where data is sent back and forth to servers. Instead, all of the work is done on your device. And you’ll only see the results if you want to.
Another key point is that the generative models Apple uses are small and designed to excel at specific tasks, such as creating small images or translating conversations in real time. In that way, they’ll avoid teetering on the edge of the uncanny valley, where realistic AI-generated images have led to a surge of fake news and misinformation.
It’s also worth noting that Apple Intelligence will only be available on the iPhone Pro 15 and the upcoming iPhone 16 this fall, because of their hefty processing requirements. That may be a bummer for investors, but it’s a smart strategy that will prevent users from over-promising what an AI can do.