The Inside Story of Apple Intelligence

Google, Meta, and Microsoft, as well as companies like OpenAI and Anthropic, all had well-developed methods for relational AI by the time Apple suddenly announced its own force this June. Conventional wisdom claimed that this entry was inappropriately later.

Apple thinks. The company’s officials claim that it is arriving just in time and that it has been secretly preparing for this time for decades.

That’s one of the things I learned from speaking with senior Apple managers about how they came up with what is now known as Apple Intelligence in the fall. Craig Federighi, the senior vice president for application engineering, is a well-known figure from the continuous it net series keynote product launches. John Giannandrea, senior vice president of machine learning and AI approach and who formerly led machine learning at Google, is less well known. In a separate meeting, I spoke with Greg” Joz” Joswiak, Apple’s senior vice president for global marketing. ( These exchanges made me better suited for my sit-down with Tim Cook, which I had the following day. ) Cook and other senior managers reaffirmed that Apple would solve this destructive technology with the same care and attention to detail that the business is renowned for, despite the enormous potential of AI. The staff at Cupertino was generally waiting for this moment to arrive, to borrow a song from some players who likewise co-founded a company called Apple.

In 2015, Joswiak says,” We were doing intelligence, like helping to identify routes in maps and which apps you would employ next. ” We didn’t often talk about it publicly, but we were there and ahead of the curve”.

In 2018, Apple poached Giannandrea from Google, a walk that Cook told me showed that Apple anticipated the coming AI change. He was hired by the business, which was strange for Apple and went against tradition when it hired him. Giannandrea was surprised to discover how heavily Apple was utilizing cutting-edge AI in some of its most well-known materials when she arrived. ” Face ID is a feature you use every day, many, many times a day to access your telephone, and you have no idea how it actually works”, he says. ” Many profound learning operations are conducted personally on your phone just to make that have work. But to the consumer, it only disappears”.

Federighi says that experimenting with OpenAI’s GPT-3 design, which was released in 2020, stoked his mind. He claims that” stuff that seemed on their way to becoming conceivable suddenly appeared extremely possible.” The real problem was “obviously” whether the systems could be used in an Apple method.

Apple immediately had several teams working on transformer-based AI models. Therefore, there was no need for Apple to create an internal task force to develop AI products when ChatGPT captivated the market in November 2022; instead, work was already being done to develop characteristics that would” only disappear.” ” We have methods of drawing up useful expertise across the business to achieve larger product transformations”, says Federighi. We put up many of those strands in a way that is really incredibly well-known to us at Apple when it came to making a bigger move in a public manner.

Apparently, Apple had likewise relocated some AI-savvy designers from its no-go-getter smart car job to the Apple Intelligence work. When I brought this up, Federighi gave me a sigh that signaled,” Hey, I’m not going that”.

Not that any of this was simple. ” This is a place along a journey”, says Giannandrea. ” Computer knowledge is changing. For more and more of the items that we want to perform, like statement recognition, speech understanding, and summary, the only way to do it is to develop. And so this is a development”.

Apple made the early decision that Apple Intelligence would not be a standalone solution but rather something that was integrated into the system. Unlike a number of its rivals, Apple had no curiosity in producing artificial general intelligence, a pursuit that to the company seems impossible and nearly silly. Giannandrea asserts that there are numerous unanswered issues and advances required from the field’s most reliable experts. The notion that you’re using these systems to transfer to AGI is really naive. He says that Apple may very well be involved in significant breakthroughs—not to kickstart the Singularity, but to increase its products. He claims,” We likely have more professionals working on what we call “investigations” than we do working on what’s going to send subsequent year, referring to what appears to be the name for the business’s fundamental analysis. ” I do say that Apple employees are somewhat more concerned with the impact of their work on consumers.”

” Apple is laser focused on things that are going to create your day-to-day living much”, says Joswiak. Making use of personal data is finally necessary, whether it’s knowing who your close friends are when you search for a particular image, recalling places you’ve been when you’ve visited when you use charts, or keeping track of what you’ve saved from Safari. Apple would need to organize the personal information of its users in a comprehensive manner in order to fully utilize AI. Because of its very public focus on privacy, the company felt it was uniquely qualified to pitch to its customers. Protecting that privacy, however, turned out to be a major technical challenge.

” We had to innovate at the data center level, at the system level, at the OS level, at the cryptographic and security protocol level, at the distributed AI inference level,” says Federighi.” We had to innovate at every level up the stack. We had to extend an on-device processing level of security that you have on your phone, to advance processing in the cloud.” ” I hope this is how everyone goes about carrying out this kind of process.” Even if it means losing its competitive advantage, he claims that others will follow his example because of how strong his conviction is. There are many instances where we have very conflicting opinions about how others may copy what we do, but we were happy to set an example and inspire, according to Federighi.

Only when it began developing those privacy systems did the business officially introduce Apple Intelligence, and later large numbers of features were made available in large numbers and much fanfare. However, the truth is that the initial public iteration of Apple Intelligence isn’t quite blowing people away. Critics complain that its inbox summaries, email rewrites, and photo search, as well as a more conversational Siri, don’t seem much different than the gen-AI offerings already unveiled by competitors. But just as Apple crashed its rivals ‘ parties when it came to digital music streaming and smart watches, the company is confident, if not hubristic, that its Apple-tude will ultimately prevail. ” This is a multi-decade thing”, says Giannandrea. ” I was very excited about the stuff that we’ve announced this year, but I think Craig and I are much more excited about, like, what’s in the next 10 years”.

I posed the two executives ‘ questions, which might include details of what those potential future products might be. And also naturally, they refused. ” You know us better than that”, says Federighi. Apple will take it seriously, even if some rivals debut similar innovations first. This crowd prides itself on being not first, but best. If that philosophy still holds true, then generational AI might serve as the ultimate test.

This is not the first time I got an exclusive look at Apple’s AI journey. In a day of interviews with Federighi, Phil Schiller and Eddy Cue, as well as scientists Tom Gruber and Alex Acero, the company gave me a sneak peek into how it was putting the most recent AI techniques into practice in August 2016. The message then, as now, was that Apple was on it, but doing AI in its own way.

Even as Apple is bear-hugging machine learning, the executives caution that the embrace is, in a sense, business as usual for them. Deep learning and ML are only seen by the Cupertino illuminati as the most recent addition to a steady flow of cutting-edge technologies. Yes, yes, it’s transformational, but not more so than other advances, like touch screens, or flat panels, or object-oriented programming. In Apple’s view, machine learning isn’t the final frontier, despite what other companies say. Cue points out that it’s not like there weren’t other technologies that have helped to alter how we interact with devices over time. And no one at Apple wants to even speculate on the bizarre/scary speculations that frequently surface during AI discussions. As you’d expect, Apple wouldn’t confirm whether it was working on self-driving cars, or its own version of Netflix. However, Apple was not working on Skynet, the team made it abundantly clear.

” We use these techniques to do the things we’ve always wanted to do, better than we’ve been able to do,” says Schiller. ” And on new things we haven’t been able to do. As it develops both internally and through the processes we create products, it will ultimately be a very Apple way of doing things.

Luana asks,” Can Intel resuscitate, or is it going to become Xerox”?

Thanks for the question, Luana. At this week’s WIRED Big Interview event, I kept thinking about Intel’s situation, which had once dominated the chip industry with similar triumphalism, as I watched Nvidia CEO Jensen Huang speak with such great confidence. It invented the microprocessor! Building on that innovation, Intel became the default chip for the personal computer revolution. But ultimately, it fell victim to the Innovator’s Dilemma. The mobile revolution and the significance of graphic chips, earth-shattering events that its rivals exploited, could be forgiven for their failures. The rise of custom silicon by businesses like Apple and Amazon, which likely reduced their reliance on Intel’s offerings, may have been the coup de grâce. At this point, who needs Intel?

I wouldn’t equate Intel with Xerox, though. The advances of the latter’s PARC division were never really exploited. The clueless top brass at Xerox’s headquarters sat by while Apple, and eventually everyone else, copied its graphical user interface. Intel, in contrast, built a fantastic business—so successful that it was easy to fall into complacency. I can’t say whether resuscitation is possible. ( If twice-former CEO Pat Gelsinger can’t figure it out for a huge paycheck, don’t expect me to do it for free. ) However, Intel does possess a lot of valuable resources, particularly its chip-making facilities. It also has billions of dollars in funding from the Biden administration to produce those chips in the United States, at least until Trump stops. If Intel doesn’t acquire one of its rivals, it might wait until the next significant opportunity arises, and a hungry new CEO is wise to take a gamble on it. In order to keep the mighty from falling, Huang might think about embroidering the Intel logo on the lining of his renowned leather jackets.

Questions can be sent via email to [email protected] or by leaving a comment below. In the subject line, write ASK LEVY.

End Times Chronicle

It’s now official: Björk declared that the apocalypse has already happened. But don’t worry, “biology will reassemble in new ways”.

Last but Not Least

For this year’s final Plaintext, which I will use up my vacation days, this special Steven Levy-themed collection of links is available.

Here’s the complete Tim Cook Big Interview. I enjoy Tim’s response to Apple’s request to demonstrate the Vision Pro mixed-reality headset to Stevie Wonder.

If you want to watch me interviewing Cook, here’s a video.

Figma CEO Dylan Field apologized at the WIRED Big Interview event after receiving an offer to buy Adobe. He apologized for telling me that he wasn’t selling his business. ( The deal ultimately fell apart under scrutiny from regulators. )

Mira Murati, the former OpenAI CTO, stated at the same time that she still believes AI won’t end humanity, but that it’s up to us to ensure that this is the case.