Apple manages to talk about AI without ever talking about AI: use is king

This was one of the big questions we had ahead of WWDC 2023, which is considered the most important of the decade. Would artificial intelligence be the focus of the conference? After all, AI has become the hot topic of technology. Google, Facebook, Amazon, Microsoft and all the industry heavyweights have jumped into the fray with both feet.

The only thing that was missing, in short, was Apple. The conference seemed like a good opportunity to reveal to the world its intentions in this new playground – especially that of generative AI, as this is where all the attention is focused. For example, we think of Midjourney for creating images and ChatGPT for creating texts.

The AI ​​was present, but hidden under a precise lexicon

And Apple responded well during WWDC 2023, but in a way that turned out to be pretty clever: the American company distilled references to artificial intelligence throughout its event, but never directly called it that. A sacred achievement, which has not gone unnoticed by several observers.

Does Apple smell a certain overdose around AI, with a concept that has been dished up to us in all sauces for months? Still, the Cupertino-based company has applied methodically to avoid this buzzword. However, this did not prevent him from discussing his progress in this area, but using more precise terms than an all-encompassing acronym: “AI”.

So there was never any talk of generative AI, nor GPT (short for Generative Pre-trained Transformer, a fashionable technique). Instead, the company used a more elaborate jargon, which allowed it to occupy the ground without giving the impression of following the herd. People talked about ‘neural engine’, ‘machine learning’ or ‘transformer model’.

These concepts are, of course, not new. For example, the term “neural engine” is employed by Apple since 2017 with the arrival of the Apple A11 Bionic system-on-chip. This processor is precisely optimized for tasks in AI. At its presentation nearly six years ago, Apple indicated that this neural system ” is designed for machine learning algorithms. »

It can still be found in 2023 when the M2 Ultra chip is presented, with an even faster “neural engine” than the previous generation (up to 40% compared to the M1 Ultra). The company never quotes AI in his press releasebut ensures that the number of cores (32) and the number of operations per second (31.6 trillion) are specified.

This use of technical jargon was rightly reflected when Apple fleshed out the M2 Ultra chip. ” It can run transformer models “, it was said at one point during the conference. A formula that may escape those who do not follow the AI ​​sector diligently. But again, Apple isn’t using this term arbitrarily.

A transformer is a kind of neural network architecture — we find this name in GPT, by the way. This mechanism, developed by a Google research team in 2017, can now be found in image generation tools (Midjourney, Stable Diffusion, DALL-E). The transformer originates from deep learning, a process of machine learning.

Concepts of AI always associated with use

The AI ​​concepts Apple mentioned at its conference are never mentioned outside of the applications they fit into. An example with iOS 17, which one was also announced during the conference: autocorrect gets a full update with a transformative language model […] for word prediction. »

When Apple mentioned adaptive audio for the AirPods, there was no mention of AI with the “adjusted volume” feature. Apple preferred to describe this breakthrough as machine learning,” to understand environmental conditions and listening preferences over time so you can automatically adjust the sound. »

And the examples abound. FaceTime doesn’t use AI, but ” most advanced machine learning techniques to create a digital avatar. There is no more AI to detect the fillable fields in a PDF document, but ” new machine learning models “. And the wallpaper? They also rely on advanced models.

If we were teasing, we could say that Apple set itself the mission of putting as many keywords around AI as possible without ever going directly into the subject. If there had been a drinking game where you had to have a drink every time you said the word AI, you would have walked out of WWDC sober (unlike other conferences where you would be dead drunk).

Apart from this surprising peculiarity, this undoubtedly reflects Apple’s vision of AI: not to make it an above-ground subject, but a means of extending already existing applications. We see it with adaptive audio, where the addition of certain processes aims to further improve the sound experience. Ditto with the M2 Ultra chip which is there to provide more computing power.

For further

Show what we’re talking about

It’s also a way for Apple to show they know what they’re talking about by using specific words at the right time. “Machine learning”, “neural engine” or even “transformer model” are much rarer in the communication of other companies than AI, a collective formula that means nothing anymore because it is used on all occasions.

This is clearly noticeable and smart. We’re showing AI because it can extend already established practices, rather than presenting potential applications with much more uncertain adoption (like these articles that list the x things you can do with ChatGPT). Use dictations. A policy that is not new at Apple.

However, there is a flip side to this coin. Apple has certainly shown that it knows what it’s talking about (although other tech companies just have a different vocabulary). They use specific terms, which will appeal to those who know about them. But that’s where the shoe pinches: it is also jargon that eludes the general public. The general expression IA certainly has its shortcomings, but it is also a way of addressing the population.

If you liked this article, you’ll like the next ones: don’t miss them by subscribing to Numerama on Google News.