Google Launched 10 Products at Its Annual Conference — Hardware, Software, Services, and in the End All of It Is AI | Reporting Live From Google I/O

None of them can yet build Google a business to follow advertising. But investing in technology this steadily beats not investing at all.

At 10 a.m. San Francisco time on May 17, the Shoreline Amphitheatre in Mountain View, California was packed with 7,000 developers from around the world. The outdoor venue was hosting Google I/O, the company's most important conference of the year.

This developer conference was originally convened to launch new versions of Android. For the past two years it has been a progress report on Google's artificial intelligence.

AI was again the biggest block of this year's program. Google CEO Sundar Pichai opened by restating the shift from "mobile-first" to an AI-first company, and said Google would rethink every one of its products accordingly, and push AI into academic research and medicine.

He had described the same shift on the same stage a year earlier, when he also launched an AI voice assistant and put it inside a smart speaker, phones, and social apps.

The room was much cooler this year, rather like the weather outside — Mountain View, usually hot enough in the sun to send you looking for shade, dropped to 17 degrees Celsius today. Across more than two hours of talks and demos there was little hearty applause or cheering. The audience stayed calm.

There was not much that was genuinely new; most of it was improvements to products that already exist. Google's AI assistant can now hold a conversation in text. The smart speaker can make phone calls. AI has been pushed deeper into email, photos, social apps, and other internet products. The assistant has even reached the iPhone. It all felt like a repeat of last year's conference with small revisions.

That is not necessarily a bad thing. I/O used to be a matter of launching a pile of new things to a cheering crowd, after which they died within a year or two and, the next year, several thousand people pretended those products had never existed.

That was never the best way to push technology forward.

This year's I/O showed that most of the AI projects announced last year have survived, and are getting better features and more uses.

Google Lens: It Recognizes Everything Around You, From Passwords to Restaurants

This was the first product shown at today's I/O. Google Lens is a search feature built on image recognition. Pichai demonstrated it on stage: point the camera at a flower and Google Lens tells you the flower's name — rather like Microsoft's flower-recognition app.

It also reads barcodes, and Google supplies the product information. More interesting: point it at any shop sign on the street and Google tells you the details of that business, and can even connect you to the Wi-Fi inside. Reviews show up too, telling you which dishes are good and what other people made of the place.

Google Lens is not an app of its own. It is a feature, and it will be built into Google Assistant and Google Photos.

Google Assistant: Everywhere, Including the iPhone

This year's I/O invoked "AI first" again, but Google Assistant itself has not moved much.

As expected, Google's voice assistant got an official iOS version, which will exist as a standalone app. Google did not spend long on it: iOS limits on app permissions mean the app cannot do its full job, and can only call Google's own apps and services, such as YouTube.

Updates to Assistant itself were small. The first new feature introduced was, of all things, keyboard input: you can now type to the voice assistant instead of only speaking to it.

Then there are languages. Google Assistant added French, German, Japanese, Korean, and Italian.

Scott Huffman, Google's vice president of engineering and the head of Assistant, said the voice assistant has to reach more devices: "The Assistant is becoming more conversational, more available, and always ready to help."

To get Assistant everywhere, Google released an SDK, and the big screen on stage filled with the logos of more than 70 partners. The kit lets software and hardware makers build Assistant into their own products, so their products can be run by voice.

There is also a new feature called Google Actions, which makes carrying out a task feel more like a conversation. Product manager Valerie Nygaard demonstrated ordering food delivery on stage. The whole thing played out as one continuous conversation, the assistant walking her step by step through choosing dishes, picking a drink, choosing an address, and finally paying with a fingerprint.

Google Home: It Can Make Calls and Run Your TV Now

Google Home, the smart speaker launched at last year's I/O, got no second-generation model either — only a few new features.

Hands-free calling is one. Rishi Chandra, Google Home's vice president of product, really did use the speaker on stage to call his own mother. Tell Google Home who you want to call and it dials the number.

The person placing the call doesn't need a phone at all; the speaker can reach any mobile number or landline directly. The service is free in the United States and Canada for now.

Google Home can tell voices apart, so it knows who is giving the call command and dials from that particular user's contacts.

Google Home has no screen, but it can work with a television. Through a Chromecast dongle, Google Home taps into the TV's operating system, and the TV becomes the speaker's screen. Ask "what's on my schedule this week" or "what's the weather today" and the TV shows the answer.

Not many people check their schedule or the weather on a TV. The more useful move is to ask Google Home for "something fun on YouTube" (or another video service), and the TV throws up a row of recommendations to pass the time.

Google Home also added content partners, including HBO Now and the free tier of Spotify, plus YouTube TV, the streaming service YouTube launched this year.

The speaker finally supports Bluetooth, too, so you can connect a phone and use it as a plain playback device.

Google Photos: Google's Fastest-Growing New App Will Now Pick Photos and Send Them to Your Friends

Google Photos, the company's AI photo album, may have been the biggest surprise of this I/O. Anil Sabharwal, vice president of Google Photos, said the app has 500 million monthly active users who upload 1.2 billion photos and videos a day. It is getting three new features.

The first is Suggested Sharing. Image recognition tells Google Photos whose faces are in a picture, and if you match those faces to your contacts, it suggests which photos to send to whom.

You can also set the sharing to happen on its own. Sabharwal demonstrated this on stage: he set his wife to automatic sharing, so every time he takes a photo with his daughter's face in it, Google Photos sends it to his wife. If she had not installed the app, she would get an email with the photo attached.

This solves a familiar problem. After an evening out with friends you usually want the photos with you, or a particular person, in them — and getting hold of them is a nuisance.

"What used to take a few hours now takes a few minutes," Sabharwal said.

To spare you even more picking, Google Photos has launched Photo Books, a service that prints paper albums. You select a pool of candidates in the app and the AI picks the keepers. In a magazine-style layout screen you arrange them as you like. Google then makes the album and delivers it to your door.

Photo Books is open only in the United States for now. A softcover album costs $9.99, and the price climbs if it runs past 20 pages.

YouTube Built TV Channels and Tipping for Creators, but the Tipping Is a Bit Weak

YouTube added features as well. CEO Susan Wojcicki said on stage that YouTube generates more than 1 billion hours of viewing a day. What makes YouTube "different," she said, is that it is open and interactive: "everyone can have a voice."

The new feature that carries that idea is tipping during live broadcasts. YouTube product manager Sarah Ali said the number of livestreams grew fourfold over the past year, so they have added Super Chat to live shows: viewers can post comments in a white-backed comment panel below or to the left of the stream.

The panel looks a little like a chat window, and the comments refresh so fast they resemble danmaku. Anyone can talk to the streamer there, and anyone can pay to sponsor, which pins their ID, the amount, and their comment to the top. Chinese livestreaming apps do much the same, though next to virtual sports cars and yachts, YouTube's tipping has nothing like the visual punch.

Bid higher and you can push the interaction further. Google livestreamed a show by the Slow Mo Guys — a two-man act with more than 9 million followers on YouTube — from the lawn outside the venue. Ali tipped them $500 through the Super Chat window, and 500 water balloons promptly rained down on the hosts. It looked like a good time.

Another small update is 360-degree video. YouTube is putting the format on the living-room television; until now it worked only on the web and on phones. Swing the remote left and right to turn the view. And 360-degree video can now be live.

Android O: Not Much You Can See, Mostly About Disciplining Sloppy App Developers

The Android operating system is a standing item at I/O, but this year's release is mostly tuning, with the updates aimed at how the phone feels to use.

On stage, Dave Burke, Android's director of engineering, showed Android O's picture-in-picture mode, which lets you handle something else while on a Duo video call.

In the new system, an app's most recent notification appears if you press and hold its icon.

If you regularly sign in to sites with Chrome on your desktop, Android O's autofill saves you retyping a username or email address: one tap logs you in.

Copying text on a phone now uses machine learning too. Double-tap an empty spot on the screen and the system selects the field you probably want — an address, a phone number, an email, a name. No more dragging the cursor back and forth, which makes selecting text quicker.

On security, Google introduced Google Play Protect, which it says uses machine learning to scan apps and check whether they are safe. Every Android phone with the Google Play store on it is covered.

Android O also cut startup time sharply, and clamped down on background apps and services and on background location polling, to protect battery life and keep the system smooth.

One thing Google did not explain is what the letter O in the code name stands for. Every previous version was named after a dessert, so some people guess Oreo.

Android Go: A Hardware-and-Software Package That Makes Cheap Phones Usable

Android One, part of Google's plan to find its next billion users in poorer countries, has a new name after years of trying: Android Go.

From Android O on, every version of the operating system will have a stripped-down Go edition. Phone makers in the program get a simplified build of stock Android, designed for low-spec handsets, so that devices with only 512MB to 1GB of memory can run Android smoothly.

Android Go does require preinstalling Google Play and a set of redesigned Google apps. Chrome's built-in Data Saver, for instance, loads only the parts of a page the user needs, which cuts data use.

YouTube Go lets users see screenshots of a video before downloading it, and choose the resolution by hand. It also lets users share downloaded videos with each other offline.

For languages, Google's Gboard keyboard now offers 191 of them, and can translate as you type.

The Play store inside Android Go will recommend apps that suit low-end handsets. Google will also steer developers toward apps that work offline and come in installers under 10MB.

On tracking and monitoring, Google is giving developers finer detail — a developer can see, for example, the clock speeds of the CPU cores while a user is in the app.

That should help them build lighter apps and games that use fewer system resources.

Daydream Virtual Reality: More Partners, and a Standalone Headset This Year

Google's VR project has finally signed up some sizeable manufacturers. Clay Bavor, Google's vice president of virtual reality, said LG's latest flagship phone and Samsung's Galaxy S8 and S8 Plus will support Daydream soon.

In the platform's first year — its aim is to standardize both VR hardware and software — it had only 3 partners, and only 4 phones supported it, Google's own Pixel included.

Google's VR strategy has not changed: turn more phones into Daydream phones.

But Google also confirmed for the first time that it is not only making boxes you slot a phone into. It is developing a standalone headset that needs no phone or computer. Bavor did not make clear at the conference what the device will actually look like. What is known is that Daydream has partnered with Qualcomm, which will supply the chips. The hardware partners also include HTC Vive and Lenovo, and the product will go on sale later this year.

The Visual Positioning System: Indoor Positioning From Images

Google has not forgotten augmented reality. Project Tango has been folded into the VR team, but it is still moving. Asus's new phone will support Tango — meaning several cameras on the back — and on the screen you see a virtual figure laid over the real scene.

Tango's features are getting practical, too. It now plugs into Google Maps and Google Lens, and has become Google's Visual Positioning Service.

The demo looked rather cool. The phone uses image recognition to work out where an object is and which way it faces. In a big hardware store like Lowe's you can see on the screen what is on the shelves and what each item is, as though you were wearing a pair of Google Glass.

Expeditions, once a VR app for schools, now supports AR. Holding up their phones, students can watch a tornado rise off the desk, or a DNA model appear on the floor.

Google for Jobs: Helping People Find Work, and Making Money on the Side

When Pichai came back on stage to sum up, two lines of large type appeared behind him: TensorFlow + Everyone.

Pichai said he believes Google's computing power should be open to everybody free of charge, and that this is why the free, open-source machine-learning framework TensorFlow was designed the way it was. A video at the event told a small story: a 17-year-old high-school student used machine learning to build an image-recognition tool for detecting breast cancer early.

Besides the open-source tools, Pichai introduced a job-hunting tool for the general public, Google for Jobs.

It is a job-hunting version of Google Search, using machine learning to help people find work. You can search for a role in the search engine — "retail," say — and see the kinds of jobs different companies are offering, along with location, commute, whether the work is part-time, and so on.

The listings are pulled together from partnerships with recruiting sites such as LinkedIn and Monster, which send their information over. So far 500,000 people have used the product, and companies including Johnson & Johnson and FedEx say applications with resumes rose 80% after they started using it.

The jobs in the demo were not well-paid technical roles like engineering. They were the jobs an ordinary person might look for: retail, construction, education. "We want to serve the job-hunting needs of every occupation and every level of experience."

AI Still Cannot Solve Google's Most Important Problem

"I think AI should be used to solve the most important problems," Pichai said as he closed this year's Google I/O keynote.

I/O is a developer conference, and its most important job is to get developers building new things with Google's newest technology, adding bricks to a Google ecosystem that already reaches billions of people.

Google has been the most important AI company for a long time. Search, photos, voice, and "simple" functions like spam filtering all rest on AI. When Google's TensorFlow development platform appeared, developers gathered behind it fast and packed out every related event in Silicon Valley.

Some of the technology at I/O is genuinely startling, above all the Visual Positioning System, dull name and all: from what the camera sees, AI knows at once where you are standing indoors, whether that is a big supermarket or an airport.

The grand plans by carriers, Apple, and others to fit sensors throughout indoor spaces have never worked. Google's new feature may settle the difficulty of indoor positioning for good.

Behind useful features lie technical problems that are hard to imagine. That is probably why Google is the one that built them first.

But Google — Alphabet, strictly — is a listed company. Before AI can solve Pichai's most important problem, it has to solve the problem of making money. Google Labs and Google Reader were both useful and ahead of their time, and then were shut down.

Turning AI into a business will be Pichai's biggest challenge.

As mobile search advertising grows, Google is still posting healthy growth: ad revenue passed $20 billion last quarter, and its market value is second in the world, behind only Apple.

But the ceiling on Google's advertising business could not be plainer. Revenue per click fell 19% year on year: customers will pay less and less for a single ad.

Last quarter was the first time the growth in what Google spends to acquire traffic outran the growth in its ad revenue.

At the same time, Google has run out of places to put more ads. Mobile search results are already stacked with them, and YouTube has had to look into paid subscriptions.

For a company whose main business is advertising, AI has not been a good source of revenue so far. You search for something on your phone, Google returns a list of results, and the ads slipped in among them are the revenue. But when you speak a question to Google's AI, what you want is plainly the single most reliable answer; it can hardly make you sit through an ad first.

What Google is doing is stuffing AI into every product: AI to sell dearer hardware (the Pixel phone, Google Home); AI to fight for the social market (Allo, Duo, Photos sharing); AI to get into recruiting (Google for Jobs).

Allo and Duo went unmentioned this time. Probably because not many people have downloaded them even on Google's own Android. AI has not won Google back control of social networking. That also leaves Google Photos' elegant feature for picking out and sharing photos of friends looking like dead weight. Contacts and channels for reaching people are bound up with social networks — unless all your friends use Google+ and Gmail day to day.

If Tencent or Facebook copied the Google Photos features, they would get more out of them. A better photo album cannot pull the billions of users already soaking in WeChat, Facebook, and WhatsApp away from them.

Apple and Samsung both exist, which makes it hard for Google to get anyone to pay flagship prices for a voice assistant — and that is exactly what Pixel is trying to do. Hardware, brand, design, the Mac, better third-party apps: there are too many reasons the iPhone holds on to 700 million users who like spending money.

It is also why Google Assistant, Pixel's point of difference, arrived on the iPhone today. iPhone owners will not buy a Google phone to get a "better Siri."

Even Samsung, a company with no gift for software, has made Tizen a bigger watch operating system than Android Wear — the watch is still a hardware business. And now that Samsung is using the enormous profits from its semiconductor business to subsidize its new flagship phones, Pixel's chances of surviving get slimmer.

Shopping is no different. Amazon holds an overwhelming share of e-commerce in nearly every Western country, and China belongs to Taobao and JD.com. Making it slightly easier to place an order through Google Assistant does not offset fatal weaknesses in range, price, and delivery.

The only product at the whole conference that looked commercial enough was the recruiting service, Google for Jobs. Whether it works in the end may depend on whether Google is willing to do the drudgery beyond search and sending traffic along.

In nearly every market Google's AI is trying to upend, the competition has either built a monopoly or is on its way to one.

A lead in AI is still not enough for Google to open a new source of revenue to replace the advertising income that must eventually stop growing. But investing in technology this steadily leaves more room for possibility than not investing at all.

Cover image and some photographs in this article by Cui Qiwen.