For years, Google I/O, the internet giant’s annual conference for software developers, emphasized the degree to which mobile phones were morphing in a profound way. They’d evolved from mostly stand-alone devices, albeit connected to voice and data networks, to devices that did relatively little of the actual machine processing that was making them so useful. The heavy lifting increasingly took place inside the megabrain called Google. It was powered by complex and opaque software and hardware that massaged and moved information—mostly collected from you and me—around globe-spanning data centers. Our devices were becoming displays (and input systems) for Big Tech’s universe, where the real intelligence lived.
The power of that model was astonishing. Nowhere was it more obvious than in Google Maps, which for me (and I’m betting a lot of other people) is the single most essential use of my mobile phone, after conversations of various kinds.
But at this week’s I/O in Mountain View, California, Google announced that it would shift some of the most essential intelligence back to Android devices.
Not tomorrow, but soon. What Google CEO Sundar Pichai announced in his keynote talk was what he termed a “significant milestone”—made possible by the seemingly never-ending improvements in hardware and software—that would put vastly more processing power into the phone itself for one of the most important features Google and the rest of the tech industry have been working on: voice recognition.
Until now, Pichai said, Google has needed about 100 gigabytes of storage on its servers, plus a constant network connection, to make voice recognition work at a good-enough level for each user’s everyday use. Pichai said advances have given the company the ability to bring that 100 gigabytes down to 500 megabytes and move it directly into the user’s mobile device. This eliminates network-caused delays and, he said, will be transformative in how people use their mobile devices.
The first example of this will be in Google Assistant, the voice-driven feature of newer Android phones that lets you tell the device what you want it to do for you and get answers to your questions. The keynote demo of this running solely on the phone was more than impressive. We users of devices haven’t seen this yet—Google says this won’t arrive until the next generation of phones appears later this year, and it’s not clear whether the capability will be added to current phones.
What’s most intriguing about this shift in power back to the users’ devices is what it suggests. A related I/O announcement gives us a hint at some of the possibilities. For one thing, Google says it’s working on ways to keep more of users’ personal data on the devices and not in the company’s data banks.
Google calls a key element of this initiative “federated learning,” and is deploying it first in its Gboard mobile keyboard software. It’s creepy to realize that a mobile screen keyboard could easily be capturing and sending back to someone else’s storage everything you type. One reason to do this is to learn how people are using devices and improve the quality of data entry for the user. Federated learning means that what you type stays right in your device. Only when a lot of other people are entering the same kinds of words and phrases does the mother ship learn that—and no one’s personal data ends up in the corporate servers.
This kind of system, which Apple already uses to a degree, has great potential to help protect privacy. Whether the promise will survive corporate imperatives remains to be seen, but as our personal devices get more power and storage, there are fewer reasons—other than corporate surveillance—for companies to hold on to it.
The main reason they’ll keep a lot of our data is that we’ll ask them to, unfortunately. The convenience of “cloud” storage, for backups and having access from any device, has made it part of how we work. But there are other ways to back up and have access to our personal information without handing it all over to Google and other giant companies. We should be learning more about them and using them.
Not everything will move to our phones anytime soon. Google Maps is a classic example of why not. If we want real-time traffic information from Google, there’s only one way we’ll get it: from the network.
This will be true of some other apps and services, but over time, as the devices we hold in our hands get more and more powerful in every way, the possibilities grow for new ways to use them. Personally, I’m looking forward to a genuine, full-function Star Trek tricorder.