Apple, an early champion of AI and machine learning, is without question behind virtually everyone else when it comes to generative AI – which is why I can almost buy the rumor that Apple is ready to hop into a prompt-based bed with often rival and constant search partner Google to license the latter’s powerful Gemini models for iPhone-based AI activities.
I can almost buy it – but there are many reasons why ultimately I don’t.
First, though, let’s recall how Apple has first run and then stumbled its way to a less-than-middling position in the AI market.
If you think it started with Siri, you’d be wrong. When I emailed him about the rumor, Apple analyst and Creative Strategies Chairman Tim Bajarin reminded me about Knowledge Navigator, a 1987 software-agent-based search engine concept, not entirely unlike an AI chatbot, that presaged digital voice assistants like Siri, which Apple launched almost 25 years later. Apple has been thinking about this stuff for a long time.
Siri was arguably the first on-phone digital voice assistant, and, for a hot second, was pretty impressive. However, despite numerous brain transplants and Apple’s knack for using AI, neural networks, and machine learning for everything from battery management to autocomplete and photo processing, Siri soon surrendered the pole position to Amazon Alexa and Google Assistant.
By the time we arrived at the age of generative AI, Apple was unaccountably absent, offering nothing generative-AI-worthy at WWDC 2023, and instead making a whole lot of fuss about an expensive mixed-reality headset. Granted, the Apple Vision Pro is one excellent distraction but, owing to its price and the fact that you wear it on your face, it’s unlikely to have the same impact as a sizable AI update.
Does this AI partnership ring true?
Since the summer we’ve heard rumors (also from Bloomberg’s Mark Gurman) that Apple was hard at work on Project Ajax, which may include something called Apple GPT (home-grown, prompt-based generative AI) and maybe some ways of accessing the (large language model) LLM of your choice from various partners.
I think it’s also accepted wisdom that this year’s WWDC will be 90% AI. Apple can’t afford to disappoint.
Gurman’s latest rumor, though, that Apple might simply license Google’s powerful, multi-modal Gemini AI for various on-board iPhone operations doesn’t entirely hold water – at least for me.
First, the only part of Gemini that meets Apple’s strict benchmarks for privacy is Gemini Nano. It can run locally and, as Apple’s Head of AI John Giannandrea made clear a few years ago, running any other way is “technically wrong.” I don’t suspect that Giannandrea has suddenly changed his tune and would allow more powerful Gemini models on iPhone when they’ll all need Cloud access to work.
Gemini Nano is so limited, though ( text suggestions, grammar checking, summarization), that it won’t provide anything like the generational leap Apple and its next iPhone need.
DIY AI
I’m not suggesting that Apple would never partner with Google. Your iPhone search defaults to Google because the search giant pays Apple billions every year. It’s a fantastic deal for Apple, but I think the company would love to have its own search engine (there have been rumors over the years of it trying to build one).
There is a reason that Apple Maps, which also uses Apple’s own AI, exists. I think it’s a point of pride for Apple that after a disastrous launch in 2012 the company updated, iterated, and improved Apple Maps until its embarrassing origins were a distant memory.
It’s the same reason that Chrome isn’t your iPhone’s default web browser. Apple likes to build things on its own, especially for the full-stack control it gives it.
AI is arguably the most important innovation in our lifetimes, and I think Apple wants its spot on that commemorative plaque. If it licenses Google Gemini at scale, its generative AI efforts will never recover. Apple’s massive user base would also boost Gemini well past OpenAI‘s GPT, Microsoft‘s CoPilot, and Met’s LLaMA.
Bajarin, though, sees it differently. He notes the cost of delivering a powerful generative AI architecture. With that in mind, the shortcut provided by using those “base AI architectures,” could be hard to resist.
There’s also one key way in which Apple can’t compete with Google, “Even if Apple had its own Gemini level model it probably would not have the infrastructure to serve its massive base of customers,” said Bajarin in an email.
Maybe it’s not about the infrastructure, or about having a solid base on which Apple could build a bespoke model (think how Apple Silicon is built on the ARM base). Perhaps it’s a financial decision.
Follow the money
Gurman postulates that Apple and Google might do this to reset the balance if the European Union forces Apple to divest from Google Search and let consumers choose the default. Leaving aside the confusion and frustration that might cause, there’s the billions in lost revenue Apple would have to recoup. Yes, Google could pay a similar amount to have its Gemini AI front and center on the iPhone.
Still, none of it makes sense to me.
Apple’s goal with the iPhone 16, iOS 18, and future iPhones is to differentiate its products from Android phones. It wants people to switch and they’ll only do that if they see a tangible benefit. If the generative tools on the iPhone are the same as you can get on the Google Pixel 8 Pro (and 9) or Samsung Galaxy S24 Ultra (and S25 Ultra), why switch?
I’ve seen reports that Apple has met with multiple AI companies (while constantly acquiring smaller ones). I believe the goal here is to learn how they do it, and add in whatever Apple’s AI efforts are missing to not just achieve parity, but surpass the AI competition – and that includes Google Gemini.