Apple on Monday unveiled the iPhone 16 to be released Sept. 20 with one marquee feature: Its A18 chip is “designed for Apple Intelligence,” the company’s still-unfinished artificial intelligence software. It‘s supposed to do things like summarize messages, write emails and clean up photos.
So will Apple’s AI be worth it? Testing a prerelease version of the software, I’ve found it sometimes helpful — and sometimes laugh-out-loud weird.
Weird like the time it alerted me that Donald Trump had endorsed Tim Walz for president. (Ha.) And the time it made up the idea that I’m teaching at UC Berkeley. (No.) And the time it elevated an obvious social security scam to my “priority” inbox. (Yikes). And the time it edited a selfie to make me bald. (Double yikes.)
In the preview I’m using, Apple Intelligence does an uncomfortable amount of making things up. This usually happens on low-stakes information like summaries of alerts from apps — but it feels weird nonetheless to see fabrications and misinterpretations of your life appear on your lock screen, inbox and other core parts of your iPhone.
What that means for anyone considering spending $799 on a new iPhone 16: You might want to hang on until Apple Intelligence figures out the right balance of helpful and nonsense and until we understand its impact on battery life. Until then, an older model or refurbished iPhone will likely serve you just fine and cost less. And next year, they’ll have an even better iPhone to sell you.
Apple says the AI software I’ve been testing, the third developer preview version of iOS 18.1, isn’t finished and isn’t necessarily representative of the final product.
Normally, I wouldn’t bother writing about beta software. But I’m writing about Apple Intelligence now because the company spent more than 15 minutes marketing it during its iPhone 16 launch event. “It marks the beginning of an exciting new era,” said CEO Tim Cook. If Apple wants you to buy a new iPhone for it now, I’m going to independently test what’s available.
There are Apple Intelligence features that weren’t yet available to test, such as customizable generative-AI emojis. We don’t yet know if the Siri assistant will really be smarter. (I asked it about a vaccine and it failed to give me the best information from the Centers for Disease Control; Apple says it didn’t use Apple Intelligence for that — it was regular old dumb Siri.) We also don’t know what kind of impact the final version will have on your iPhone’s battery life. (In my preview tests, my battery depletes four or more hours earlier each day.)
If I sound skeptical, it’s for two reasons: First, Apple has for years stretched the truth about the real capabilities of its existing AI tool, Siri. Second, I’ve tested many generative AI products that gin up hype but let users down with untrustworthy information or snake oil.
Beyond the marketing bonanza, Apple knows Apple Intelligence isn’t ready. A version with some features will be available “in beta” in October, it says, for those who choose to turn it on. All the promised Apple Intelligence features may not come to regular iPhone 16 and 15 Pro users for weeks or months.
The bad news is it’s possible the accuracy problem I keep encountering is inherent to the AI technology that Apple is using. We may just have to get used to some level of nonsense now spread all over our iPhones.
When Apple Intelligence makes things up
In some ways, Apple Intelligence is a smarter and likely safer way to bring AI to a mass consumer product than we’ve seen from other Big Tech companies. It’s not just an all-knowing chatbot grafted onto your phone. At its best, Apple Intelligence integrates its capabilities into existing apps in small ways to speed along tasks.
I found it helpful in many ways. For example, in the Photos app, Apple Intelligence lets you use more natural language to find photos from a particular occasion. Or you can highlight a part of a picture you find distracting, and it will use generative AI to make it disappear.
The way you’ll most notice Apple Intelligence at work is trying to summarize your mountains of incoming information. For example, the traditional Mail app has long shown two lines of preview of every email to help you decide whether to open it. With Apple Intelligence, those two lines are now an AI-written summary, which can give you a better sense of whether you want to open the email. (It knows to skip right over the “Hope you are well.”)
The unavoidable problem is that Apple Intelligence sometimes gets it wrong — misinterpreting the meanings of text, inverting people’s names, or taking images in an unpleasant direction.
This happens because factual errors — which the industry tries to downplay as “hallucinations” — are inherent to this generation of AI technology. Generative AI isn’t designed for truth, it’s designed to look for and re-create patterns. Google’s Gemini and OpenAI’s ChatGPT have the same problem.
I told Apple about the many times I saw Apple Intelligence get facts wrong (I’ve had at least five to 10 laugh-out-loud moments per day). It says it is working to improve accuracy. But so is every other AI company — and that has proven to be a giant challenge. Recent research has found that the latest AI tech just stinks at summarizing.
For iPhone users, the question is how well Apple will be able to contain errors to places where it doesn’t really matter.
“Apple is leaning into use cases that are low-stakes and high-utility. This is smart,” says entrepreneur Phil Libin, the former CEO of Evernote who now runs the video presentation start-up mmhmm. (Like me, he’s also been testing the prerelease software.)
To his point, Apple is not acting like Google, which added to its search engine high-stakes AI-written answers that it sometimes gets completely wrong.
The largest category of errors I experience are, as Libin says, low-stakes: annoyances that make me roll my eyes, but don’t cause problems.
- It summarizes app alerts from my video doorbell as nonsense, such as “Multiple people at front door and house, with recent motion at front door.”
- It doesn’t understand humor, so its summaries of chats with friends and families or photos can be laughably dumb. Once, it described a photo of my beautiful baby as “group of people in a room with white walls.”
- I asked its “clean up” function to tame my wild hair in a selfie, and it made me look bald.
But things get hairy (pardon the pun) when Apple Intelligence invents wrong information. It suggested a rewrite of an email I drafted to my neighbors that left out a key question I was trying to get them to answer.
Even worse, there were times its mistakes nudged me to take the wrong action. For example, it once summarized and elevated to “priority” status an email that it said told me I needed to change my password immediately. That’s not what the email actually said — it only warned me to change my password if I wasn’t responsible for a recent login.
It also summarized and elevated a phishing scam email about my social security number being deactivated. I wonder what might have happened to someone not paying closer attention.
There is some reason to be optimistic. Even if Apple can’t get rid of errors, it can, and hopefully will, minimize the higher-stakes places and types of information where Apple Intelligence activates. For example, it ought to avoid trying to summarize news and social media entirely. (That’s how it ended up telling me that Trump had endorsed Walz from a misreading of an alert about Walz’s brother.)
But there is a risk that regular goofs could make us feel our smartphones are less smart, a repeat of the way many already feel about the incompetence of Siri.
It’s hard to think of another time Apple asked to take your money now for something that’s so unfinished.