There’s a lot to like about Apple Intelligence, although not always in the places you’d expect. I had high hopes for making ‘Memories’ videos from a text prompt. And it worked, catching me unawares by how great a job it made of building a video about my dearly departed dog. Good boy, iPhone. Good boy. But I got a very different and unexpected type of joy from notification summaries. Not through them being useful and saving me time, but because they were often deeply and amusingly surreal.
In attempting to compress messages from friends, Apple Intelligence resorts to bizarre and almost haiku-like constructions as algorithms strip context and attempt to squash what remains into a handful of words. “7cm snow reported; complaining about tax dodging by landowners”, it said last week. I was grateful for the semi-colon, inferring that it probably wasn’t the snow itself that was very grumpy.
“Photo shared, joking about grasping, wondering about long-term impact,” said another notification, making me concerned about the sentience of a shared photo and what ‘grasping’ meant in this missive. But nothing beat the summarised motion alerts from our Arlo doorbell, which suggested a house under siege: “Multiple people and an animal detected at doorbell!” I wondered if I needed to immediately hide.
Actual Inelegance
The problem isn’t Apple Intelligence, really – it’s artificial intelligence. The clue is in the name. This isn’t intelligence – it’s various flavours of fancy autocomplete, using masses of training data to figure out what comes next – be that the next word in a sentence, or the next pixel in a photo. Which is why output from these systems varies wildly between faintly magical to unbelievably boneheaded.
Apple’s writing and image-editing tools span the entire range. Sometimes, they get things right. Automated email responses and rewriting tools can save you time. The ‘Clean up’ feature in Photos means you don’t have to mess around with a cloning tool. Except not always. Because the writing tools sometimes merely play ‘swap the synonym’ and remove every crumb of personality from your words, or just shorten text rather than summarising it. And removing objects from images sometimes fails, because Apple Intelligence doesn’t explore the full context of a picture.
The result is the same: you must remain vigilant when it comes to errors. And in writing, there’s a particular danger. You’ll more likely spot strange glitches in a photo than an errant fiction in a wall of text.
Cattle intelligence
To be fair, Apple is far from alone. Every AI is riding a hype cycle and making claims it cannot deliver. And while AI can be useful, it’s no magic wand. It won’t write your debut novel from a prompt when it struggles with an email.
Where Apple is alone is in being very Apple about its approach to AI. Apple is being precise and careful, and for that should be lauded. Image Playground churns out anodyne illustrations, but they’re OK for a presentation and aren’t going to put jobbing creatives out of work. And the company prioritises privacy over blazing ahead. That can make its AI appear less intelligent – and often less capable – than rivals, but Apple appears to prefer that over risking your data.
Still, one area in which Apple strays from its usual path is in its tone-deaf Apple Intelligence adverts. They feature a series of idle people – an office worker who’d not read a document before a meeting; a spouse who forgot their loved one’s birthday – solving problems with Apple Intelligence while “I’m a genius” blares forth.
It’s a long way from a “bicycle for the mind”, which in one reading became a metaphor about how tech could help people to be more creative, rather than more creative in getting out of scrapes when caught short. In fact, the series almost comes across like someone goofing off at an ad agency used an AI to come up with it. Hmmm.