Tim Cook has promised “incredibly exciting” generative-AI news for 2024. But the constant dripping of reports that claim Apple is talking to Google, OpenAI, and Chinese search giant Baidu to use their AI engines hints at a future that is far more boring. It’s a future potentially filled with commoditization, privacy risks, and fragmentation of the iOS and macOS user experiences.

Last week, news broke that Apple had already inked a deal to use Baidu’s AI, Ernie, in mainland China. The report was later denied, but it further advances an idea that everyone has been suspecting for a while: Apple is behind on AI. The company already uses some AI capabilities in its operating systems—from heavily processing your photos to the automated creation of image stickers to the talking kitchen timer that is Siri—but it seems that Apple is behind other big players after pumping the brakes on generative AI in favor of Vision Pro and iCars.

WHAT IS ERNIE?

But back to Ernie. What the hell is Ernie, you may ask, aside from Bert’s long-life lover? It is Baidu’s answer to ChatGPT-4—a large language model (LLM) that started development in 2019. Ernie 4.0—its latest version—was introduced this past October by the company’s cofounder, chairman, and CEO Robin Li in this very long keynote.

In case you don’t have time to watch three hours of Li extolling Ernie’s abilities, here’s the gist of it: First, Ernie has memory. Unlike Gemini or ChatGPT, Ernie supposedly can learn from you over time like Dot, adopting a persona that adjusts to you, from the way it talks to how it can help find the best shopping deals for you (the memory aspect is important, but more on this later).

According to Li, it can also generate novel-length text from scratch, solve practical problems that involve math and geometry, and do all kinds of things for the user, like identifying objects around it. It appears to work like a regular, chat-based bot, but it seems it can help anywhere you are on your phone at any given time.

THE UX PROBLEM

While the report of the deal has been denied, we know that China requires companies to use a Beijing-approved AI by law. Apple could create its own generative AI and put it through the approval process, but that seems like an impossible prospect now.

So let’s assume that the Baidu deal is going to happen out of necessity for Apple. Let’s also consider the very real possibility that, just like Samsung, Apple could use Ernie inside China and Google Gemini or OpenAI ChatGPT in the rest of the world. This will basically mean that Cupertino has no real technological edge over its main rival in the most important user experience revolution to happen since the original iPhone.

Apple being Apple, it will almost certainly be able to shape these external AI engines into a unique and consistent user experience. But which of Ernie’s AI capabilities will Apple choose to use? And how far can Apple leverage certain AI features without introducing UX fragmentation?

To keep the experience consistent for users across the world, Apple will have to pick and choose wisely. Take Ernie’s memory capability, for example, which will only be available to users in mainland China. By introducing a technology—however compelling it may be—to just a subset of users, Apple will face a conundrum. Will a company known for its tightly controlled UX really offer a different AI experience across the world?

Even if it does, it presents another problem when it comes to the output of these AI engines. Just like Midjourney, DALL-E, and Stable Diffusion produce dramatically different images with the same prompt, using different AIs could also result in different outputs depending on which part of the world a user is located. The quality and style of generated AI text, the images, the way it favors some news and not others—this may seem inconsequential for you, but it will have an effect on how users experience this technology.

THE PRIVACY PROBLEM

For the sake of argument, let’s assume that Apple will be able to create the same UX front end for everyone, coming up with clever work-arounds that can camouflage the UX inconsistencies and the output problems caused by using different engines. Even then, there is another issue at play.

Ernie depends on Baidu’s cloud infrastructure. Which, of course, opens a can of worms the size of the Tibetan Plateau for Apple if the deal ever comes to pass. The idea of an ever-present AI that can learn from you does sound “exciting,” but when that AI is storing everything in the cloud, excitement can quickly turn to worry and panic.

The South China Morning Post, which is owned by Baidu’s rival Alibaba, reported that Ernie has ties with the People’s Liberation Army cyber warfare division, something that Baidu has strongly denied. Apple’s commitment to privacy could be seriously compromised, deeply affecting its users’ experience, at least in China.

But privacy will be an issue to contend with regardless of location if Apple decides to use third-party AI engines. With so much computation happening in the cloud, it opens Apple up to more security issues. Safeguarding its users’ privacy was one of the reasons Apple moved all its Siri features that don’t depend on the internet to local devices. But for the time being, these powerful new AI engines require the cloud to work.

Despite all these problems, people will likely be very happy to see the current form of Siri turn into Ernie or Gemini. Even Grover will be better than what we have now. But what is clear to me is that Apple has an Herculean task ahead of itself in creating a new UX experience that is distinctive from the competition. It may take months or years until Apple catches up on AI, and the consequence of this lack of vision could prove to be the biggest existential threat Apple has ever faced.