Apple's new AI features get it all wrong. I am not disappointed, because that would mean I had hope.
Think differently? Only if "differently" means "the exact same but three years behind schedule".
While everyone else in the UX world is talking about the disaster that is Liquid (gl)ass, I wanted to turn your attention to something more important in the grand scheme.
By now, we all know that Apple is behind the curve on AI adoption. There is some debate over whether Apple’s recent paper that debunks the “intelligence” part of AI was only published as a post-facto justification for Apple’s failure to keep up with everyone else.
I certainly agree with Apple’s premise in the paper. The manner in which most tech companies have been approach AI is disastrous, with an infantile credulity around the potential of AI to replace human cognition. Just as I was beginning to doubt Apple ever really cared about humans at all, they published an implicit affirmation of the human mind. Perhaps there was hope for Apple after all, right?
I held out a sliver of hope that, when Apple finally did release a major AI integration, it would feature one key differentiator from every other implementation of AI: user input translation. By that I mean that the AI would be connected to a universal database of every function in the operating system, enabling the user to execute any desired function by prompt rather than memorizing an arcane menu tree.
We shouldn’t have to memorize this stuff
Here’s an example. Let’s say you want to set your phone to “Do Not Disturb”, but you want to allow certain people to get through and make your phone ring. How do you do that? It is a multi-step process, and the toggle to activate this function is hidden in the ringtone screen of all things. Oh, and you have to manually disable this setting once you enable it. So, if you only want to allow someone’s calls through for a specific purpose, but otherwise don’t want them bothering you at all hours, you’re out of luck.
The alternative to this crappy design is to build an AI prompt layer into the interface, where the user verbally issues their command to the OS, which then identifies the necessary settings and toggles them automatically.
The prompt might be worded literally, such as: “allow Fred Smith’s calls through the do not disturb until I turn it off”
It could also be worded in a more anthropomorphic, implicit style: “I’m going to nap for an hour but Fred Smith might be calling about my insurance claim so you can wake me if he calls”
It might even be worded somewhat ambiguously if the user is sleepy “wake me if Fred Smith calls”
All three prompts would result in the exact same settings being toggled. The OS would use its LLM to standardize the diverse and chaotic forms that any human speech can take into the rigid and deterministic format of computer commands.
This is the kind of thing I was hoping that Apple would give us. iOS has tons of hidden functionality that even people who have owned iPhones for years don’t know about. A prompt interface would open all that functionality up to anyone since only the will to enter a prompt or speak a request to Siri would be required to access it. It could even hold the possibility of ever more complex functionality and customization, since any new feature would be far more likely to get adopted by users.
So did they? LOL no. Here’s what we got instead.
So close but so far
One of the AI features that Apple boasts about on their website is the fact that Siri has been trained on all the features of the device. This means that you can ask Siri how to do a thing, and it will tell you. But you still have to do it yourself. I can’t tell from the screenshot below if you can keep that info bubble open while you perform the task, but if you can’t, then it’s even worse.
I don’t want to fill my brain with this crap. This is not useful human knowledge. I want the phone to understand my commands and execute them for me. Even when I know the sequence of button presses, it’s probably still quicker for me to speak the command to Siri each time.
And yet this is what Apple gave us. I held out a sliver of optimism that Apple wasn’t behind on AI, but were blazing a different trail, something that prioritized usefulness over the infantile flashiness of the HAL-9000 delusions that OpenAI and Anthropic are peddling. But no, Apple is walking off the same cliff as everyone else, just more slowly.
Can nobody grasp the obvious?
This is one more AI-enabled company that isn’t picking the lowest, yet juiciest, fruit from the AI tree. AI is no more a substitute for human intelligence than any earlier information technology, yet it’s being treated as though it is.
Previous to AI, information technologies were created to automate away activities which are not ideal for the human brain, freeing more cognitive bandwidth for humanistic activities.
The printing press replaced the miserable and boring process of hand-copying books, freeing us to spend more time composing literature.
The calculator replaced the computation of mathematical functions on large numbers, freeing (some of) us to spend more time advancing the state of mathematics through new theorems and proofs.
The camera replaced the rote documentation of reality, freeing us to create expressive art that captures what no camera can.
By comparison, the most touted applications of AI include things like composing prose, and generating images, music, and videos. These are human activities, and ones that AI frankly sucks at. Yet this is the stuff the AI companies are pouring billions into.
I don’t want AI to write for me. I want AI to manipulate the complex text formatting features of my word processor to style the presentation to my taste.
I don’t want AI to compose music for me. I want AI to locate a good drumkit VST, generate a 5/4 beat, then alter the beat per my feedback, until I have a drum backing I can play guitar and bass over.
I don’t want AI to generate photographs out of a collage of dubiously sourced images. I want AI to help me render photorealistic 3D scenes by executing the complicated commands of a 3D modeling/rendering software, sourcing off-the-shelf models, and then altering them per my commands, or by interpreting sketches I provide.
I don’t want AI to generate video for me. I want AI to take those photorealistic 3D scenes I generated, animate them, and then apply post-production filters to further enhance the photorealism, all under my control through the mediation of prompts that tell the software exactly what to do.
In short, I want AI to eliminate all the arbitrary machine knowledge and allow me to develop my human knowledge. The more time I spend worrying about formatting commands, the less time I have to create compelling turns of phrase. The more time I spend thinking about precise settings in ProTools, the less time I have to compose catchy melodies. The more time I spend learning how to use Maya or Blender, the less time I have to write a compelling story or envision captivating scenes.
Why don’t the AI companies get this?
So much for thinking different
I had hoped that Apple would offer an alternative vision of the future to what the techbabies are hawking. I don’t know why I hoped that. Apple have never been as brilliant as the fanboys want us to believe. But they have had one core strength that other tech companies don’t, which is that they have at least a vestigial trace of a user-as-owner business model. Unlike Google and Meta, Apple is a hardware company at its core, and the sense that their technology is meant to be a useful tool to its owners rather than an insidious espionage and propaganda device still survives in their products… somewhat.
But that ethos has atrophied so badly that, instead of driving an alternative vision of human-centric AI, it simply gives us a watered-down version of what all the other techbro dingleberries have been doing for the last three years. There is no major tech company left with Apple’s level of resources that can offer a mainstream counter-narrative to the dystopian nightmare that a certain clique of misanthropic, dephysicalized geldings is pushing as our inevitable future.
Their anemic attempt at AI is just one of Apple’s increasingly random and chaotic decisions. I guess that’s what happens to your survival instinct when you’re worth $3 trillion.
Nope, no one can grasp the obvious. A failure of imagination. VUIs are a total joke to me, like a CLI but you cannot backspace and edit before you press enter, so better not make a mistake in diction or syntax or you get to start over.
You are right on the money that if we want to be able to use snazzy features, let's admit that people DNGAF about which menu it is in, and will never remember that in 2 months when they need it again. If you want to exploit natural language interfaces — and Siri implies that yes they do — then why not do something /useful/ with it. I say just because everyone is navel gazing, looking at some specific metrics like use or accuracy not user value.
Lots of great gripes to be had about how we mostly do the wrong things because a few levels above us is measuring the wrong thing, the wrong way. Streaming starting to collapse is from loosing the vision so turning a windfall (also get money from Netflix streaming our back catalog!) to an opportunity (could we make more money if we streamed ourselves?!) to a cost (oh, it costs money to run a service...) to a liability (cut costs, increase revenue, you have to be a profit center or else!).
The problem is that the behaviour you dream of probably requires a top down + bottom up approach, which -at the moment- is far from being reached (and nobody is seriously working on, Apple included). To be clear : you would need some software behaving more like AlphaGo than ChatGPT... LLMs are too unreliable even to perform the somple pre defined tasks you mentioned