Ten years ago today, Apple’s digital assistant Siri was revealed with the promise that the iPhone maker had finally figured it all out.
“For decades, technologists have teased us with this dream that you’re going to be able to talk to technology, and it’ll do things for us,” said Apple exec Phil Schiller, taking the stage at the launch of the iPhone 4s. “Haven’t we seen this before, over and over? But it never comes true.”
The problem, said Schiller, was that voice interfaces were too reliant on simple syntax. Call Mom. Dial 555-2368. Play Beethoven. “What we really want to do is just talk to our device,” he said, “and your device — in this case, your phone — will figure out what you mean and help you get what you want done.” He paused before bringing up the next slide, showing a familiar icon of a microphone cut-out in burnished aluminum. “That’s a feature on the iPhone 4s we call Siri.”
Described by Schiller as a “humble personal assistant,” Siri gave 2011’s iPhone a dose of star power during a difficult time for Apple. Just months before the phone’s unveiling, a relative unknown at the company, then-chief operating officer Tim Cook, had been named CEO. The day after Schiller’s presentation, Apple’s legendary co-founder Steve Jobs would die from pancreatic cancer. Analysts were cool on the company’s prospects but praised Siri as a potential game-changer. One called it “a powerful harbinger of the future use of mobile devices,” while another said it was “the beginning of a new user experience [for] all of Apple’s mobile and Mac products.”
A decade later, the sheen has worn off Siri’s star. “It is such a letdown,” was how Schiller described the promise of voice interfaces past, and such a description could easily be applied to Apple’s contribution to the genre. Everyone who uses Siri has their own tales of frustration — times when they’ve been surprised not by the intelligence but the stupidity of Apple’s assistant, when it fails to carry out a simple command or mishears a clear instruction. And while voice interfaces have indeed become widespread, Apple, despite being first to market, no longer leads. Its “humble personal assistant” remains humble indeed: inferior to Google Assistant on mobile and outmaneuvered by Amazon’s Alexa in the home.
Looking back on a decade of development for Apple’s personal assistant, there’s one question that seems worth asking: hey Siri, what happened?
Looking back to 2011, initial reactions to Siri were incredibly positive, with reviewers impressed by the feature’s responsiveness and accuracy. “The crazy thing about Siri is that it works — at least most of the time — better than you’d expect it to,” was The Verge’s judgment; “It’s kind of like having the unpaid intern of my dreams at my beck and call,” said CNN; “Siri saves time, fumbling and distraction, and profoundly changes the definition of ‘phone,’” said The New York Times. All in all: Apple seemed to be living up to its promises.
But reading these reviews now, it’s clear Siri was graded on a curve. Its novelty and ambition invited generous appraisals, but when reviewers noted frustrations, they caveated that the software was only in beta and that any rough patches would surely be smoothed away in due time. A detailed run-down of Siri in 2011 from Ars Technica highlights problems familiar today, with the assistant dinged for mishearing instructions in loud spaces and mangling complex commands. An instruction to “Send a text to Jason, Clint, Sam, and Lee saying we’re having dinner at Silver Cloud” is interpreted with Siri texting Jason: “Clint Sam and Lee saying we’re having dinner at Silver Cloud.”
Siri had a first-mover advantage, but it didn’t take long for rivals to emerge. Samsung introduced S Voice on the Galaxy S3 in 2012; that same year Google Now was launched for Android (replaced by Google Assistant in 2016); in 2014 Microsoft brought out Cortana for Windows Phone; and also that year, Amazon went its own way by introducing customers to Alexa on the Echo smart speaker. Speaking to your computer quickly became an expected feature not just on mobile devices, but a whole range of gadgets.
Looking through reviews and comparisons of digital assistants in this period, two things stick out. The first is that people soon get bored of Siri. As reviewers tackle iPhones after the 4s, they often note incremental updates to the assistant but never dedicate much space to it its features. Partly, this seems just because the changes are so small (e.g., retrieving sports results in iOS 6; integrating Wikipedia in iOS 7; introducing ‘Hey, Siri’ in iOS 8), but also because the novelty has worn off.
By the time we get to reviews of the iPhone 8 in 2017, Siri is mentioned in passing, if at all. Our own review summarizes the assistant’s contribution in a single line: “Siri sounds a lot nicer as well, although it’s not any more capable than before.”
The second major trend is that once competitors did arrive, Apple’s advantage evaporated quickly. A comparison of Siri and Samsung’s S Voice in 2012 notes that the latter already “offers a very good approximation” of Apple’s digital assistant, while a head-to-head test in 2014 shows that “Google Now crushes Siri.” By 2017, The Verge noted that Siri “feels largely half-baked,” and a dissatisfaction with digital assistants in general had begun to creep in. We point out that these assistants can answer basic questions just fine but fail to reliably do things for the user — like booking cinema tickets or ordering food. At least, not without creating new problems of their own.
Looking back, it’s clear that Siri’s big problem is that it failed to maintain momentum. The basic roster of tasks that made Apple’s digital assistant so appealing in 2011 — setting alarms, taking notes, and so on — were never significantly expanded upon. The ability to answer trivia questions and retrieve sport scores is fun, but not as significant an upgrade as telling a computer to complete tasks using just your voice. Meanwhile, rivals replicated and then exceeded what Siri could do. They began to offer more reliable dictation, better language understanding, and integration with third-party skills. Siri just didn’t keep up.
So, where did things go wrong? How did Apple lose its lead? The answer is complicated.
Many suggest Apple’s dedication to privacy means it can never keep up with rivals like Google whose business involves collecting users’ data because that data is incredibly useful when it comes to improving AI systems. I don’t buy this as a reason for Siri’s failure, though. First, because Apple’s love of user privacy is far from absolute. (In 2019, for example, The Guardian revealed that “a small proportion” of Siri recordings were being passed to contractors for analysis, with a whistleblower claiming they overhead discussions “between doctors and patients, business deals, seemingly criminal dealings, sexual encounters and so on.”) And secondly, because Apple is a two-trillion-dollar company. If it wants to circumvent the messy problems of collecting user data, it can, simply by paying to generate this data. Yes, analyzing random Siri interactions is helpful, but there are other ways to achieve the same improvements.
A more convincing explanation is management dysfunction. In 2018, The Information published a damning report on the comings-and-goings at team Siri. It noted that there was deep-seated disagreement within the company about how Siri should work (is it a feature focused on search and retrieval, or an assistant that carries out complex tasks?). These disagreements stemmed back to Jobs’ original plans for Siri but had devolved into “petty turf battles and heated arguments” between rival factions. They were exacerbated by a lack of leadership and continuity in the Apple execs overseeing Siri. As one former employee told The Information: “When Steve died the day after Siri launched, they lost the vision […] They didn’t have a big picture.” This tallies with the feature’s stalled development after its initial launch.
Other problems are rooted in Apple’s ideology of technology development. For example, The Information’s report claims that Apple exec Richard Williamson made the decision to only update Siri once a year, following the company’s cadence of new hardware and iOS updates. This seems to have slowed progress. (Williamson, who refuted this claim, left Apple in 2012 after spearheading the disastrous Apple Maps launch. Scott Forstall, another executive involved in Siri and Apple Maps, departed that same year. Read into that what you will.)
There’s also Apple’s walled garden approach, which means Siri has always worked well with iOS features but played badly with third-party services. While testing Siri in preparation for this story, I was consistently surprised by its failure to execute simple tasks on popular iOS apps. Siri can’t send a voice memo on WhatsApp; can’t post a story to Instagram; can’t record a run on RunKeeper; and can’t open up The New York Times crossword. Sure, some of the blame for this lack of interoperability lies with outside developers, but it’s also Apple’s job to encourage such functionality through toolkits and the like. The company certainly doesn’t lead by example, either. When I ask Siri for information I know is stored in iOS, like “show me photos from last August,” it just performs an image search for the phrase “last August.”
Instead, Apple uses Siri to herd people back to its own inferior apps like a shepherd directing sheep off a cliff-face. If I ask Siri for directions, it prompts me to reinstall Apple Maps (when I habitually use Google Maps and Citymapper). If I try to send an email to my boss, Siri tells me, “I’m sorry I can’t do that,” and then directs me to the App Store to download Apple’s default mail app (I use Outlook). And, here’s a sign of how slipshod Siri’s development is right now: when this happens, Siri sets up a search for “mobilemail” on the App Store. This, of course, isn’t the name of Apple’s mail app, but an ID used by iOS developers, and so it draws a blank when you search for it on the App Store. That’s the kind of broken functionality you get when a company isn’t thoroughly testing its own product.
This last point, though, highlights a problem particular not just to Apple’s assistant, but to voice interfaces more generally, and that is one of expectations.
When Schiller introduced Siri in 2011, he stressed time and time again that Siri would understand users — that it knows what they are saying, just like a real person. This set the bar too high for Siri’s functionality. If you treat voice interfaces as if they have the same level of fluency and knowledge as a human being, you will always be disappointed. We speak, and they stumble. We guess what they’re capable of, and they disappoint. Usually because they don’t support the app or command we thought they would. Each failed interaction then teaches users: don’t trust this feature. By comparison, screens and displays tell us clearly what we can and cannot do. They offer menus, directions, and buttons. A voice offers only itself and our projections of intelligence. For Siri, users have been guided by Apple’s flair for the theatrical. They expect too much, and Apple delivers too little.
Here’s a true story. I took a break just now from writing this article to make a cup of tea and remembered that I had a meeting in an hour’s time. Worried it might slip my mind, I did what I often do in these situations: I asked Siri to set a reminder. “Siri, remind me at ten to five that I have a call,” I said. “Okay,” said Siri, “Setting a reminder for tomorrow at five: you have a call.” I tried again. This time Siri created a reminder for ten o’clock in the evening. The third time, I paused mid-command, trying to think of a clearer way to word my query. Siri got tired of waiting and beeped at me: “What do you want me to remind you of?” And at that, I gave up.
It’s true that Siri and its ilk are often disappointing, but they still tempt users because they hold great potential. Despite the problems associated with voice interfaces, the technology represents a genuine advance. I regularly use Siri for quick tasks, like taking notes, setting timers, and making searches. And when it works, it works seamlessly and unthinkingly. It’s a genuine time-saver. Siri can do much more, too, especially if you’re willing to dive into the world of Apple Shortcuts and smart home commands.
As an accessibility tool, voice controls and dictation have opened up modern gadgets to many more users, and since Apple introduced Siri in 2011, the company has launched a number of products that rely heavily on voice. This is either because screen real estate is limited (the Apple Watch) or it’s nonexistent (the AirPods and HomePod). In years to come, we can expect Apple’s augmented reality glasses to be added to this list. With this in mind, Apple urgently needs to fix Siri — not ignore it.
If you want to be optimistic, then there are some signs the company is turning the ship around. AI, in general, has received much more attention from Apple in recent years. From poaching Google’s head of machine learning in 2018 to designing its own AI processors to the regular launch of AI-enhanced features, the company is clearly paying more attention to the field. And, best of all, Siri itself has seen a few significant improvements, with on-device processing and availability on third-party hardware added this year.
I’m still skeptical, though. For a start, even when it comes to basic commands, it often seems Siri is not just standing still but moving backward. With iOS 15, Apple removed a decent chunk of Siri’s functionality, including tasks related to notes and photos, and third-party integrations like ride-hailing and payments. Other basic commands, like checking voicemail, also seem to have recently disappeared (whether temporarily or not isn’t clear).
The big problem, I think, is that Apple still doesn’t know what it wants Siri to be. Is the feature simply a way to control your phone with your voice — letting you navigate apps and find content? Or is it something more ambitious — an actual assistant capable of carrying out complex tasks on your behalf? Apple tends to present Siri as the latter in marketing materials, while users find its functionality limited to the former. As someone who reports on AI and machine learning, I think we’re still many, many years away from building computers that truly understand us. Language is just too complex, too deeply rooted in human experience and culture, to be brute-forced by the sort of statistical models we’re throwing at the problem. And while, yes, there are lots of impressive new language systems out there, none of them are reliable enough to create a flawless digital assistant.
If Apple wants to salvage Siri, I think it needs to reset expectations and focus on core competencies instead. It’s interesting to compare Siri’s launch with that of its competitors. When Google introduced Google Assistant in 2016, for example, the focus was less on solving complex tasks and understanding users’ every whim, and more on making the company’s basic search functionality accessible in more places. It was a tighter focus that gave Google the space to surprise, rather than disappoint. (Though the company has certainly over-promised in later ads, too.) Siri, by comparison, surprised us all when it launched in 2011, but has since burned out that goodwill. Apple needs to re-focus on the basics rather than push into a future that doesn’t yet exist. It needs to start listening.