With the web, you can just click on a link and land on a web page. That’s one click and a few seconds. What if you could run any app with one tap? That’s what we’re working on. We’re evolving Android apps to run instantly, without installation. We call this Android Instant apps. We’re going to share a quick preview of what we’ve been working on, and I can’t tell you how much I’ve been looking forward to this moment for a really long time.
So let’s say my friend Michael sends me this link to Tasty on BuzzFeed video. And let’s be completely clear. Let’s be honest with each other. I do not have the BuzzFeed video app installed on my phone. So what we’re going to do is we’re going to tap the URL, and it’s going to take me right into BuzzFeed video’s Android app, without installing it.
Ficus, go! What’s happening here is Google Play is fetching only the pieces of the app that we need right now — there. We are in the Android app and I didn’t even have to install it. It was pretty fast too; right? So here you can see there’s a bunch of different videos showing to make a whole bunch of different recipes, and the videos start playing automatically. Remember, it’s a real Android app, right? And I can just swipe and go to the next video really quickly. The app was able to open so fast because it’s been split up into modules. When Ficus tapped the link, Google Play downloaded only the code that was necessary to display this screen. And if I want to keep BuzzFeed video on my home screen, it’s simple to install the app right here.
Here’s another example. B&H photo and video has a beautiful Android app but I don’t have it on my phone because I don’t shop for cameras every day. Now if I’m searching for something specific, like a camera bag I can still get that same experience. With one tap, the app opens up right to the bag I want to buy. Technically, this is a deep link to the Android activity B&H wrote to display this product page. And that’s all Google Play needed to download. I can also swipe here and see more details about the bag.
Now, when I add it to my cart, the animation there, it was pretty sleek and at checkout time Android Pay works, just like if I had the app installed. I don’t have to pull out my credit card or type in my name and address. With Android Instant apps, I’m already signed in and I am ready to pay. So it’s going to take me two taps, not two minutes.
Finally, let’s see how Android Instant apps can help me when I’m out and about. So I walk up to a parking meter and I need to pay. I’m in a real hurry today and I don’t have time to install a parking app, but what if I could just tap my phone, and with NFC it could bring up the parking app immediately? All I have to do is choose how long I want to park. I’m already done. And now Ficus can run off the stage to his meeting and even add more parking time later if he really needs to.
So that’s Android Instant apps. As a user, it’s totally seamless, from launching the app, to signing in, to making payments.
Now, as a developer, you’ll update your existing app. It’s the same Android APIs, the same project, and the same source code. And it can take less than a day of work, depending on how your app is built. You’ll modularize your app and Google Play will download only the parts that are needed on the fly, as we saw here.
We’re really excited to give developers more ways to get their apps into the hands of users. In addition to discovering your apps in the Play Store and installing them, Instant apps will provide another on-ramp. People can use your Instant app directly, and as we showed earlier, if they want to install it, that’s easy, too. Most importantly, you will be in control of the experiences you build. And when you do make the update, your app will be just one tap away for over a billion people.
Oh, yeah, I almost forgot to mention one really important detail. So this demo that Ficus did, he actually did that on a phone running KitKat. You should know that Instant apps is going to be compatible all the way back to Jelly Bean. And we really want for all of you to be able to try this, and we’re really going to make it available to you just as soon as we can. But it’s a big change in how we think about apps. We want to get it right. And so that’s going to take some time for us. We’re working now with a small set of developers and we’ll be gradually rolling out access and actually rolling out Instant apps to users later this year.
We are so excited about all of the opportunities that this will open up, and we can’t wait to see what you are going to build when your app is just a tap away.
And with that, I’ll hand it back to Sundar.
Sundar Pichai – CEO, Google
Thank you, Jason and Ellie. Firebase is the most comprehensive developer offering we have done till date. I’m glad it’s available today and look forward to hearing your feedback.
We talked a lot today about machine learning and AI. We think there’s an opportunity to accelerate computing by working on this with everyone else. And so we’re trying to do that in two ways. First, we are opening up core components of our machine learning systems. Last year, we open-sourced TensorFlow so that developers can embed machine learning and deep neural nets with a single API. In 2015, it was the most forked project on GitHub, and it is the number one machine learning project on that site.
Last week we open-sourced a natural language parser which is also built on top of TensorFlow. We are doing these things so that we can engage the external community and work on this together with everyone.
Second, for developers and companies out there, we are also exposing our machine learning capabilities through our Google Cloud platform. We already have a cloud machine learning platform underway, and you get access to computer vision, speech, language, and translation APIs. And we are working on bringing many more machine learning APIs so that you can get access to the same great capabilities we have inside at Google. We believe this will be one of the biggest differentiators for the Google Cloud platform over time.
And, by the way, when you use Google Cloud platform, you not only get access to the great software we use internally, you get access to specialized hardware we build internally. And talking about that, for machine learning, the scale at which we need to do computing is incredible and so we’ve started building specialized custom hardware. We call these Tensor Processing Units or TPUs.
TPUs deliver an order of magnitude higher performance per watt than all commercially available GPUS and FPGAs. And when you use the Google Cloud platform, you get — you can take advantage of TPUs as well. TPUs are what powered AlphaGo, DeepMind’s AlphaGo, in its game against Lee Se-dol. Go is an ancient Chinese board game, has a simple 19-by-19 grid but it is one of the most complex games humans have ever designed. It has more possible board configurations, many more possible board configurations than there are atoms in the universe.
Beating Go for computers was widely considered to be the grand challenge for AI and most people thought it wouldn’t happen for another decade or so. So we are really thrilled that AlphaGo was able to achieve this milestone recently. One thing worth calling out, in the second game there was a move 37 by AlphaGo. It changed the course of the game and is now widely considered one of the most beautiful Go moves ever seen in tournament play. It was not just an intuitive move but a very creative move. We normally don’t associate computers with making creative choices and so to us, this represents a significant achievement in AI.
By the way, Lee Se-dol has won every single game since he played against AlphaGo and he has even replayed some of the moves he learned from AlphaGo in that game.
As a state-of-the-art capabilities in machine learning and AI progress, we see them becoming very versatile and we think it applies to a wide range of fields. I want to give you two examples. First, robotics. At Google, a bunch of 20%ers decided to help robots — train robots to pick up objects. You can see it behind me. This is not new. You usually do it by writing control system code. You program the robots with rules. But this time they decided to use deep learning techniques. So you create a continuous feedback cycle so that the robots can learn hand-eye coordination by themselves using deep learning. As you can see, they keep doing it and they keep learning it over and over again and over time, they get better and they even learn natural and useful behaviors. So, for example, the robot is nudging the stapler away to pick up that yellow object. We didn’t write that rule. The robot learned it automatically using deep learning. So it’s an amazing example of what machine learning can do.
Another example in healthcare, diabetic retinopathy is the fastest growing cause of blindness. It affects 4.2 million people in the US and many more worldwide. To detect it, you need to do a scan, a scan of the eye like you see behind me and a highly trained doctor can detect it. If detected early, the treatments are effective. If detected late, it causes irreversible blindness. And it’s very, very difficult to have highly trained doctors available in many parts of the world. So we set out to work with a small team of engineers and doctors and, again, use deep learning and started training on eye scans. And over time, our computer vision systems have gotten really good at detecting diabetic retinopathy early. This is still early, and there’s a long road ahead, and we’ll work with the medical community to get it in the hands of as many people as possible. But you can see the promise, again, of using machine learning.
When I hear about advancements like these, I’m reminded that we live in an extraordinary period for computing. Whether climate change, healthcare, or education, the most important struggles already have thousands of brilliant and dedicated people working to make progress on issues that affect everyone.
Now, consider what the best climate change researchers, doctors, or educators can do with the power of machine learning assisting them. As you have seen today, I’m incredibly excited about the progress that we are making with machine learning and AI. We believe that the real test is whether humans can achieve a lot more with the support of AI assisting them. Things previously thought to be impossible may, in fact, be possible. We look forward to building this future together with all of you.
Thank you for joining us at Google I/O.