Google Keynote (Google I/O ‘22)
[VIDEO PLAYBACK] - It began with a problem, and it was kind of a big one. - There was so much information in the world, and yet it was so difficult to find. - So we tried to solve that problem. But along the way, we found new ones, like how to reply to all those emails? What's the best way to get from A to B? How do I get that look? Or how do I get rid of that? Or this? - My computer just crashed. - We built that.
From that, came this. To stop this, we made that. The more we look, the more we find. Find a problem.
Build a solution. - A climate crisis. - Our goal is to make the sustainable choice an easier choice. - There's someone that can't see, can't hear, really make it accessible to as broad a set of people as possible.
- Take a selfie. - OK. Get ready. - We know that the starting point is uneven, but we can generate equitable outcomes and give everybody a level playing field.
- I can actually just be inspired by the world around us. - Technology has the power to make everyone's lives better. It just has to be built. [MUSIC PLAYING] [END PLAYBACK] SUNDAR PICHAI: Thank you. Thank you. All right.
Mic check, one, two, three. Can you guys hear me? AUDIENCE: [INAUDIBLE] SUNDAR PICHAI: It's on. After two years of starting many meetings on mute, I thought I should double check. All right. Good morning, everyone, and welcome.
Let's actually make that welcome back. It's so great to be back in Shoreline Amphitheatre after three years away. To the-- [APPLAUSE] Well, thank you for joining us, and to the thousands of developers, partners, and Googlers here with us, it's great to see all of you.
And to the millions more joining us around the world on live stream, we are so happy you're here, too. Last year, we shared how new breakthroughs in some of the most technically challenging areas of computer science are making Google products more helpful in the moments that matter. All of this work is in service of a timeless mission-- to organize the world's information and make it universally accessible and useful. I'm excited to show you how we are driving that mission forward in two key ways-- by deepening our understanding of information so that we can turn it into knowledge and advancing the state of computing so that knowledge is easier to access, no matter who or where you are. Today, you will see how progress on these two parts of our mission ensures Google products are built to help, and I'll start with a few quick examples. Throughout the pandemic, Google has focused on delivering accurate information to help people stay healthy.
Over the last year, people used Google Search and Maps to find where they could get a COVID vaccine nearly two billion times. We've also expanded our flood forecasting technology to help people stay safe in the face of natural disasters. During last year's monsoon season, our flood alerts notified more than 23 million people in India and Bangladesh, and we estimate the support of the timely evacuation of hundreds of thousands of people. In Ukraine, we worked with the government to rapidly deploy air raid alerts.
To date, we have delivered hundreds of millions of alerts. In March, I was in Poland, where millions of Ukrainians have sought refuge. Warsaw's population has increased by nearly 20% as families host refugees in their homes and schools welcome thousands of new students. Nearly every Google employee I spoke with was hosting someone. In countries around the world, Google Translate has been a crucial tool for newcomers and residents trying to communicate with one another. We are proud of how it's helping Ukrainians find a bit of hope and connection until they are able to return home again.
Real time translation is a testament to how knowledge and computing come together to make people's lives better. More people are using Google Translate than ever before, but we still have work to do to make it universally accessible. There is a long tail of languages that are underrepresented on the web today, and translating them is a hard technical problem.
That's because translation models are usually trained with bilingual texts. For example, the same phrase in both English and Spanish. However, there's not enough publicly available bilingual texts for every language. So with advances in machine learning, we have developed a monolingual approach, where the model learns to translate a new language without ever seeing a direct translation of it. By collaborating with native speakers and institutions, we found these translations were of sufficient quality to be useful. Today, I'm excited to announce that we are adding 24 new languages to Google Translate.
[APPLAUSE] This includes the first Indigenous language of the Americas. And together, these languages are spoken by more than 300 million people. Breakthroughs like this are powering a radical shift in how we access knowledge and use computers, yet so much of what's knowable about our world goes beyond language. It's in the physical and geospatial information around us.
For more than 15 years, Google Maps has worked to create rich and useful representations of this information to help us navigate. Advances in AI are taking this work to the next level, whether it's expanding our coverage to remote areas, reimagining how to explore the world in more intuitive ways. Around the world, we've mapped around 1.6 billion buildings and over 60 million kilometers of roads to date. Some remote and rural areas have previously been difficult to map due to scarcity of high quality imagery and distinct building types and terrain.
To address this, we are using computer vision to detect buildings at scale from satellite images. As a result, we have increased the number of buildings on Google Maps in Africa by five times since July 2020, from 16 million to nearly 300 million. [APPLAUSE] We have also doubled the number of buildings mapped in India and Indonesia this year, and globally, over 20% of buildings on Google Maps have been detected using these new techniques. We've gone a step further and made the data set of buildings in Africa publicly available. International organizations like the United Nations and the World Bank are already using it to better understand population density and to provide support and emergency assistance.
We're also bringing new capabilities into Maps. Using advances in 3D mapping and machine learning, we are fusing billions of aerial and street level images to create a new high fidelity representation of a place. These breakthrough technologies are coming together to power a new experience in Maps called immersive view.
It allows you to explore a place like never before. Let's go to London and take a look. What a beautiful city. Say you're planning to visit Westminster with your family. You can get into this immersive view straight from Maps on your phone, and you can pan around the sides. Here's Westminster Abbey.
And if you're thinking of heading to see Big Ben, you can check if there is traffic, how busy it is, and even see the weather forecast. It's London, so I'm guessing it's rain. Now, if you're looking to grab a bite during your visit, you can check out restaurants nearby and get a glimpse inside. [APPLAUSE] What's amazing is that this isn't a drone flying in the restaurant.
We use neural rendering to create the experience from images alone. And Google Cloud Immersive Stream allows this experience to run on any smartphone. This feature will start rolling out in Google Maps for select cities globally this year. Another big improvement to Maps is eco-friendly routing. Launched last year, it shows you the most fuel efficient route, giving you the choice to save money on gas and reduce carbon emissions. Eco-friendly routes have already rolled out in the US and Canada, and people have used them to travel 86 billion miles, helping save an estimated half million metric tons of carbon emissions, the equivalent of taking 100,000 cars off the road.
I'm happy to share that we are expanding this feature to more places, including Europe later this year. In this Berlin example, you could reduce your fuel consumption by 18% taking a route that is just three minutes slower. These small decisions have a big impact at scale. With expansion into Europe and beyond, we estimate carbon emissions savings will double by the end of the year. [APPLAUSE] And we've added a similar feature to Google Flights. When you search for flights between two cities, we also show you carbon emission estimates, alongside other information, like price and schedule, making it easy to choose a greener option.
These eco-friendly features in Maps and Flights are part of our goal to empower one billion people to make more sustainable choices through our products, and we are excited about the progress here. Beyond Maps, video is becoming an even more fundamental part of how we share information, communicate, and learn. Often, when you come to YouTube, you're looking for a specific moment in the video, and we want to help you get there faster.
Last year, we launched auto generated chapters to make it easier to jump to the part you're most interested in. The results are great for creators, because it saves them time making chapters. We are now applying multimodal technology from DeepMind. It simultaneously uses text, audio, and video to auto generate chapters with greater accuracy and speed. With this, we now have a goal to 10x the number of videos with auto generated chapters from eight million today to 80 million over the next year. Often, the fastest way to get a sense of a video's content is to read its transcript, so we are also using speech recognition models to transcribe videos.
Video transcripts are now available to all Android and iOS users. Next up, we are bringing auto translated captions on YouTube to mobile, which means viewers can now auto translate video captions in 16 languages, and creators can grow their global audience. We'll also be expanding auto translated captions to Ukrainian YouTube content next month, part of our larger effort to increase access to accurate information about the war.
[APPLAUSE] Just as we are using AI to improve features in YouTube, we are building it into our workspace products to help people be more efficient. Whether you work for a small business or a large institution, chances are you spend a lot of time reading documents. Maybe you felt that wave of panic when you realized you have a 25 page document to read ahead of a meeting that starts in 5 minutes.
Google, whenever I get a long document or email, I look for a TLDR at the top. TLDR is short for too long, didn't read. And it got us thinking, wouldn't life be better if more things had a TLDR? That's why we have introduced automated summarization for Google Docs. [APPLAUSE] Using one of the machine learning models, Google Docs will automatically parse and pull out the main points.
This marks a big leap forward for natural language processing. It requires understanding of long passages, information compression, and language generation, which used to be outside of the capabilities of even the best machine learning models. And Docs are only the beginning. We are launching summarization for other products in Workspace. It will come to Google Chat in the next few months, providing a helpful digest of chat conversations, so you can jump right back into a group chat and look back at the key highlights. What's for lunch? [APPLAUSE] What's for lunch is definitely a highlight, in my opinion.
We are working to bring transcription and summarization to Google Meet, as well. So you can catch up on some of the most important meetings you missed. Of course, there are many moments where you really want to be in a virtual room with someone, and that's why we continue to improve audio and video quality, inspired by Project Starline, which we introduced at I/O last year. We have been testing it across Google offices to get feedback and improve the technology for the future, and in the process, we've learned some things that we can apply right now to Google Meet.
Starline inspired machine learning powered image processing to automatically improve your image quality on Google Meet. And it works on all types of devices, so you look your best wherever you are. [APPLAUSE] We're also bringing studio quality virtual lighting to Meet. You can adjust the light position and brightness so you'll still be visible in a dark room or sitting in front of a window. We are testing this feature to ensure everyone looks like their true selves, continuing the work we have done with Real Tone on the Pixel phones on the Monk Scale, which we will tell you about in just a moment. These are just some of the ways AI is improving our products, making them more helpful, more accessible, and delivering innovative new features for everyone.
Now, I'll turn it over to Prabhakar to share the progress we are making on our original and most important product, Google Search. [APPLAUSE] PRABHAKAR RAGHAVAN: Thanks, Sundar. People's quest for knowledge often starts with a search.
And over time, we've worked hard to understand the trillions of questions that people ask us every year to deliver the most helpful information possible. But the way people search for information shouldn't be constrained to typing keywords in the search box. It's human nature to gain knowledge through multiple senses and inputs as we go about our day. For instance, if I hear a bird chirping outside the window, I might point to it and ask my wife, what kind of bird is that? And while it's a huge challenge for computers to understand information the way we do, advances in technology are helping bridge the gap. With AI technologies like natural language understanding and computer vision, we're transforming search to be far more natural and helpful than ever before. Imagine the future.
We can search any way and anywhere and find helpful information about what you see, hear, and experience in whatever way is most intuitive to you. This is our vision for the future of search, and it's one we've already taken steps towards. When Google started, our breakthrough was in understanding text based queries. Over time, people have asked us more complex and nuanced questions, and our investment in natural language understanding has significantly improved our ability to answer these, even when the query feels like a brain teaser. I don't know this one either.
It's "Hachi-- A Dog's Story." That's the answer. But for many questions, it can be easier to speak than type, which is why, over a decade ago, we introduced voice search. We now get hundreds of millions of voice queries every day, and adoption is even higher amongst new internet users. In India, for example, nearly 30% of Hindi queries are spoken.
But often, seeing is understanding, so we re-imagined Google Search yet again with Google Lens to help you search what you see using your camera right from your search bar. Lens is now used over eight billion times a month, which is nearly triple last year. Now, we are re-defining Google Search yet again. We're combining our understanding of information across multiple modes to help you express your needs more naturally than ever before. Just last month, we launched multisearch, one of our most significant updates to search. On the Google app, you can now search by taking a photo and asking a question at the same time.
You can snap a pic of a spill proof water bottle and ask for one with rainbows on it to brighten your kids day. Or in my case, I was able to take a photo of my leaky faucet and order the part to fix it. The funny thing is, I still don't know what the part is called. [LAUGHTER] But this is just the beginning of what we can do with Multisearch. Later this year, we'll add a new way to search for local information with Multisearch Near Me.
Just take a picture or long press one you see online and add "near me" to find what you need from the millions of local businesses we serve on Google. Near Me will work for a multisearch for everything from apparel to home goods, to my personal favorite, food and local restaurants. So let's say I spot a tasty looking dish online.
I don't know what's in it or what it's called, but it's making me hungry. With this new capability, I can quickly identify that it's japchae, a Korean dish, find nearby restaurants that serve it, and enjoy it no time. [APPLAUSE] While this all seems simple enough, here's what's happening under the hood. Google's multimodal understanding recognizes the visual intricacies of the dish and combines it with an understanding of my intent, that I'm looking for local restaurants that serve japchae. It then scans millions of images and reviews posted on web pages and from our active community of Maps contributors to find results about nearby spots.
Multisearch Near Me will be available globally later this year in English and will come to more languages over time. [APPLAUSE] Today, this technology recognizes objects captured within a single frame, but sometimes you might want information about a whole scene in front of you. In the future, with an advancement we're calling scene exploration, you'll be able to use Multisearch to pan your camera, and ask a question, and instantly glean insights about multiple objects in a wider scene. Let me give you an example.
Let's say you're trying to pick out the perfect candy bar for your friend, who's a bit of a chocolate connoisseur. You know they like dark chocolate and have an aversion to nuts. And of course, you want to get them something good. If you went to the store today to find the best nut-free dark chocolate, you'd be standing in the aisle for a while. You'd look at each bar, figure out which type of chocolate it is, whether it's nut-free, compare and contrast the options, and maybe even look up reviews online.
But thanks to Scene Exploration, you'll be able to scan the entire shelf with your camera and see helpful insights overlaid in front of you. [APPLAUSE] Yeah, insights overlaid, so you can find precisely what you're looking for. Try doing that with just keywords. Here's how it works. Scene Exploration uses computer vision to instantly connect the multiple frames that make up the scene and identify all the objects within it.
Simultaneously, it taps into the richness of the web and Google's Knowledge Graph to surface the most helpful results. In this case, which bars are nut-free, dark chocolate, and highly rated. Scene exploration is a powerful breakthrough in our device's ability to understand the world the way we do. And it gives us a superpower-- the ability to see relevant information overlaid in the context of the world around us.
You could imagine using this in a pharmacy to find a scent-free moisturizer or at your local corner store to find a Black-owned wine label to support. AUDIENCE: That's right. [APPLAUSE] PRABHAKAR RAGHAVAN: This is like having a supercharged Control-F for the world around you. Looking further out, this technology could be used beyond everyday needs to help address societal challenges, like supporting conservationists in identifying plant species that need protection or helping disaster relief workers quickly sort through donations in times of need. From Multisearch Near Me to Scene Exploration, the advancements we've talked about today are in service of a broader vision to make Search even more natural and helpful. So you can search your whole world, asking questions in any way and anywhere.
To deliver on this promise, Google needs to serve billions of people's diverse information needs. And it's critical that they see themselves reflected in our products. Building for everyone is a core value at Google.
You might recall, at I/O last year, we announced Real Tone, a multiyear initiative to build more equitable camera and imagery experiences on Pixel 6. [APPLAUSE] We are now expanding our commitment to skin tone equity across Google products. We partnered with Harvard professor Dr. Ellis
Monk, who spent the past decade researching the impact of race, ethnicity, and skin tone in social inequality. And his research and expertise are shaping how we approach inclusivity in our products. [APPLAUSE] Before I hand it over to my colleague, Annie, to tell you more about the work we're doing in the space, let's hear from Dr. Monk himself. [VIDEO PLAYBACK] - Oftentimes, when it comes down to representation, I'm amongst the darkest that you would find in media, when I know that there are much darker folks, even of different races. - Color biases and colorism are really a global phenomena. The reality is life chances, opportunities-- all of these things are very much tied to your phenotypical makeup.
- It's facts. I feel like ever since I was a kid, if you were a dark, you were not considered pretty. - From a really young age, I did feel a really strong sense of responsibility to dig more deeply into issues of colorism.
So for the last 10 years, I've been working on issues of racial inequality. And I developed a 10-point skin tone scale with the goal of making sure that everyone across the skin tone continuum feels represented. - Do you feel like right now it's easy to find your skin tone? - When we type "beauty tutorial," if we only see white women, it just ties into the bigger picture of not being represented. - We can weed out these biases in our technology from a really early stage and make sure that the technologies that we have work equally well across all skin tones. I think this is just a huge step forward.
- Being represented is a right. If the internet truly is for everyone, then everyone should be represented. - There is great potential for the tech industry to adopt the scale, because the world is a better place when technology works equally well for everyone. [END PLAYBACK] [MUSIC PLAYING] ANNIE JEAN-BAPTISTE: At Google, we build products for the world.
Billions of people with varied backgrounds and experiences rely on our products every day. As Dr. Monk mentioned, skin tone is one of many important dimensions that shapes people's identity and experiences. And we all deserve to feel seen and validated. Today, we're excited to share how we're starting to use the Monk Skin Tone Scale to build more inclusive products across Google. Developed by Dr. Monk, the scale adds a critical step
to our product development and testing to ensure the technology powering our futures works well for everyone. Our research has shown that more people in the US find the Monk Scale to more accurately reflect their skin tone, compared to the current industry standard. This was especially true for people with darker skin tones. We're testing the scale globally and in different product, settings and will continue to improve it to reflect people everywhere. At Google, we've started using the Monk Scale to help improve how we understand and represent skin tone in products like Photos and Search.
Every day, millions of people search the web expecting to see images that match the diversity of the world around them. We've started to roll out improvements to images in Google Search to show a range of skin tone diversity so that people from all kinds of backgrounds can find more relevant results. And for makeup queries, like everyday eye shadow and bridal makeup looks, users will have a new way to filter by relevant skin tones to find more helpful results. I can't tell you the number of times that complimentary shades-- [APPLAUSE] I can't tell you the number of times complimentary shades from my skin tone haven't been available. And so I'm personally thrilled to be able to easily filter for images that I can relate to.
Building on our work with Real Tone, we're also using the Monk Scale to improve imagery experiences in products like Google Photos. And later this month, we'll be launching new Real Tone filters that were designed to work well across skin tones and evaluated using the Monk Scale. These filters were crafted by a diverse range of renowned image makers, who are celebrated for beautiful and accurate depictions of their subjects. With Real Tone filters, you can apply the beauty and authenticity of professional editing to your own photos with just a few taps.
[APPLAUSE] Building more inclusive experiences is a long-term commitment, one that involves close collaboration with brands, publishers, and researchers. And that's why we're pleased to announce that today we are open sourcing the Monk Skin Tone Scale so anyone can use it as a more representative skin tone guide-- [APPLAUSE] --in research and in product development. By open sourcing the scale, our goal is to improve it over time in partnership with the industry. In the coming months, we'll also be developing a standardized way to label content for images on Google Search. Creators, brands, and publishers will be able to use an inclusive schema to label images with attributes like skin tone, hair color, and hair texture.
All of this is part of our ongoing commitment to ensuring that the web is as representative as our world. And now, let's check in with our watch party in Sao Paulo. [MUSIC PLAYING] SUNDAR PICHAI: Thanks, Annie and team, for such inspiring work. And olá, Sao Paulo.
So far today, we have talked about how we are advancing access to knowledge as part of our mission from better language translation to improved search experiences across images and video to richer explorations of the world using Maps. Now we are going to focus on how we make that knowledge even more accessible through computing. The journey we have been on with computing is an exciting one. Every shift from desktop to the web, to mobile, to wearables, and ambient computing has made knowledge more useful in our daily lives.
As helpful as our devices are, we have had to work pretty hard to adapt to them. I've always thought computers should be adapting to people, not the other way around. So we continue to push ourselves to make progress here.
To share more about how we are making computing more natural and intuitive with the Google Assistant, here's Sissie. SISSIE HSIAO: Thanks, Sundar. It's amazing how quickly voice is becoming such a common way to access computing.
Every month, over $700 million people around the world get everyday tasks done with their Assistant. They can just say, hey Google, to get help on the go, in their homes, and even in the car. But these interactions are still not as natural as they could be. First, you should be able to easily initiate conversations with your Assistant. So today, we're introducing two new options so you don't have to say, hey Google, every time.
[APPLAUSE] First is a new feature for Nest Hub Max called Look and Talk, which is beginning to roll out today. You can simply look directly at your device and ask for what you need, like when you make eye contact to start a conversation with another person. Once you opt in, Look and Talk is designed to activate when both Face Match and Voice Match recognize it's you.
And video from these interactions is processed entirely on device, so it isn't shared with Google or anyone else. [APPLAUSE] Let me turn the camera back on and show you how it works. GOOGLE ASSISTANT: The mic and camera are back on. SISSIE HSIAO: Walking into the kitchen to start a weekend with my family, I can simply look over and ask, show me some beaches in Santa Cruz. GOOGLE ASSISTANT: I found a few beaches near Santa Cruz. [APPLAUSE] SISSIE HSIAO: Pretty cool, right? How long does it take to get to that first one? GOOGLE ASSISTANT: By car, the trip to Natural Bridges State Beach is 51 minutes.
[APPLAUSE] SISSIE HSIAO: That's so much easier than saying the hotword over and over. The ability to distinguish intentional eye contact from a passing glance requires six machine learning models that are processing over 100 signals, like proximity, head orientation, and gaze direction, to evaluate the user's intent all in real time. We've also tested and refined Look and Talk to work across a range of different skin tones using some of the same principles of inclusion behind Real Tone on Pixel 6 camera and Monk Scale. [APPLAUSE] We're also excited to expand Quick Phrases on Nest Hub Max. Quick Phrases already let's you skip the hotword for things like answering calls on Pixel 6 and stopping timers on Nest devices.
And in the next few months, you'll be able to ask your Assistant for many common requests, like setting alarms, asking for the time, and controlling lights from your Nest Hub Max, all without saying the hotword. [APPLAUSE] All right, check this out. Turn off the living room light. That was so easy.
I just said what I wanted. Designed with privacy in mind, you choose which Quick Phrases are enabled for your Nest Hub Max. So those are two ways that it's getting easier to start talking to the Assistant. We're also improving how the Assistant understands you by being more responsive as you just speak naturally.
If you listen closely, people's conversations are full of ums, pauses, and corrections. But that doesn't get in the way of understanding each other. And that's because people are active listeners and can react to conversational cues in under 200 milliseconds.
Humans handle this so naturally, but doing this for open-ended conversations across the Assistant is a really hard problem. Moving our speech models to run on the device made things faster, but we wanted to push the envelope even more. The breakthrough comes by creating more comprehensive neural networks that run on the Google Tensor chip, which was built to handle on-device machine learning tasks super fast. Let me show you a preview of how this will all come together. For example, I might tap and hold on my Pixel Buds and say, play the new song from-- GOOGLE ASSISTANT: Mhm.
SISSIE HSIAO: Florence and the something. GOOGLE ASSISTANT: Got it. Playing "Free" from Florence and the Machine on Spotify. [MUSIC PLAYING] SISSIE HSIAO: You heard how I stumbled at the beginning. But my Assistant gently encouraged me to complete my thought.
And then even when I messed up the artist name, Google Assistant correctly figured out the song I wanted. It's amazing how these improvements change the way it feels to talk to your Assistant. You can stop worrying about the right way to ask for something and just relax and talk naturally. This is how we're pushing computing forward with natural conversation, letting you easily initiate conversation and making it so you can just speak naturally, all so you can be truly understood.
I'm excited to see how our voices will become a faster, hands-free way to get things done across many types of devices, including a growing Android ecosystem that you'll hear about in a few minutes. Thanks and back to you, Sundar. [APPLAUSE] SUNDAR PICHAI: Thanks, Sissie. We are continually working to advance our conversational capabilities. Conversation and natural language processing are powerful ways to make computers more accessible to everyone. And large language models are key to this.
Last year, we introduced LaMDA, our generative language model for dialogue applications that can converse on any topic. Today, we are excited to announce LaMDA 2, our most advanced conversational AI yet. We are at the beginning of a journey to make models like these useful to people. And we feel a deep responsibility to get it right. And to make progress, we need people to experience the technology and provide feedback.
We opened LaMDA up to thousands of Googlers, who enjoyed testing it and seeing what it was capable of. This yielded significant quality improvements and led to a reduction in inaccurate or offensive responses. That's why we have made AI Test Kitchen. It's a new way to explore AI features with a broader audience. Inside the AI Test Kitchen, there are a few different experiences.
Each is meant to give you a sense of what it might be like to have LaMDA in your hands and use it for things you care about. The first is called Imagine It. This demo tests if the model can take a creative idea you give it and generate imaginative and relevant descriptions. These are not products.
They are quick sketches that allow us to explore what LaMDA can do with you. As you see, the user interfaces are very simple. Say you're writing a story and you need some inspirational ideas. Maybe one of your characters is exploring the deep ocean. You can ask what that might feel like.
Here, LaMDA describes the scene in the Mariana Trench. It even generates follow-up questions for you on the fly. You can ask LaMDA to imagine what kind of creatures might live there. Remember, we didn't hand program the model for specific topics like submarines or bioluminescence. It's synthesized these concepts from its training data. That's why you can ask about almost any topic-- Saturn's rings or even imagine being on a planet made of ice cream.
Staying on topic is a challenge for language models. Say you're building a learning experience. You want it to be open ended enough to allow people to explore where curiosity takes them but stay safely on topic. Our second demo tests how LaMDA does with that. In this demo, we have primed the model to focus on the topic of dogs. It again starts by generating a question to spark conversation.
Have you ever wondered why dogs love to play fetch so much? And if you ask a follow-up question, you get an answer with some relevant details. It's interesting. It thinks it might have something to do with the sense of smell and treasure hunting.
You can take the conversation any way you want. Maybe you are curious about how smell works, and you just want to dive deeper. You'll get a unique response for that, too. No matter what you ask, it'll try to keep the conversation on the topic of dogs. If I start asking about cricket, which I probably would, the model brings the topic back to dogs in a fun way. I do think dogs would make-- [LAUGHTER] Now, the challenge of staying on topic is a tricky one.
It's an important area of research for building useful applications with language models. And this last demo is my favorite, so we are going to do it live. Let me turn it over to Josh. [APPLAUSE] JOSH WOODWARD: Thanks, Sundar. As a team, we've learned a lot on this project. And this will be the first ever live demo of LaMDA from stage.
Are you all ready to see how it works? [APPLAUSE] All right. Here I am in the AI Test Kitchen app. I'm going to open up this demo called List It.
Now, List It it explores if LaMDA can take a complex goal or topic and break it down into relevant subtasks. It can help me figure out what I'm trying to do and generate useful ideas I might not have thought of. If you love to-do lists like I do, this is a dream come true. I'm going to tap Start.
And this is a project I've been thinking a lot about lately-- plant a vegetable garden. I'll send this off to LaMDA. And there it is. On the fly, it's come up with these different steps and broken it down into this list of subtasks.
I can see things like make a list of what I want to grow, the location. I can also regenerate a list on the fly to get even more ideas. Now, what's interesting about these is I can quickly drop into one of them. Let's say this one, like what might grow in the area.
And you can see it will give me further suggestions. I can keep going, breaking this down, where eventually it gives me a list of what I might want to plant, like tomatoes, or lettuce, or garlic. We'll keep garlic out of it this time. One of the other things lambda does is not just break down lists, but you can generate a tip. So here when I tap generate a tip-- oh, it's never seen this one before, actually. It's telling me, if I have a small yard or patio, it gives me different vegetables I might be able to grow.
Now, when we think about products like this and experiences like this, it's much more than just coming up with a list of vegetables to grow. If I scroll back up, you can see all the different pathways that LaMDA is helping me think through and giving me tips along the way. And just like that, this whole task feels a lot less daunting. Back to you, Sundar. SUNDAR PICHAI: Thanks, Josh.
Just like the other demos, you can input all kinds of goals, whether it's moving to a new city or learning an instrument. These experiences show the potential of language models to one day help us with things like planning, learning about the world, and more. Of course, there are significant challenges to solve before these models can truly be useful. While we have improved safety, the model might still generate inaccurate, inappropriate, or offensive responses. That's why we are inviting feedback in the app, so people can help report problems. And we'll be doing all of this work in accordance with our AI principles.
Our process will be iterative, opening up access over the coming months, and carefully assessing feedback with a broad range of stakeholders, from AI researchers and social scientists to human rights experts. We'll incorporate this feedback into future versions of LaMDA and share our findings as we go. Over time, we intend to continue adding other emerging areas of AI into our AI Test Kitchen. And you can learn more here. As you just saw, LaMDA 2 has incredible conversational capabilities. To explore other aspects of natural language processing and AI, we recently announced a new model.
It's called Pathways Language Model, or PaLM for short. It's our largest model to date and trained on 540 billion parameters. PaLM demonstrates breakthrough performance on many natural language processing tasks, such as generating code from text, answering a math word problem, or even explaining a joke. It achieves this through greater scale. And when we combine that scale with a new technique called Chain-of-Thought Prompting, the results are promising.
Chain-of-Thought Prompting allows us to describe multi-step problems as a series of intermediate steps. Let's take an example of a math word problem that requires reasoning. Normally, how you use a model is you prompt it with the question and an answer.
And then you start asking questions. In this case, how many hours are in the month of May? So as you can see, the model didn't quite get it right. So in Chain-of-Thought prompting, we give the model a question-answer pair. But this time, we explain how the answer was derived, kind of like when your teacher gives you a step-by-step example to help you understand how to solve a problem. Now, if you ask the model again, how many hours are in the month of May, or other related question, it actually answers correctly. And it even shows its work.
[APPLAUSE] Chain-of-Thought Prompting increases accuracy by a large margin. This leads to state of the art performance across several reasoning benchmarks, including math word problems. And we can do it all without ever changing how the model is trained. PaLM is highly capable and can do so much more. For example, you might be someone who speaks a language that's not well represented on the web today, which makes it hard to find information.
Even more frustrating because the answer you're looking for is probably out there. PaLM offers a new approach that holds enormous promise for making knowledge more accessible for everyone. Let me show you an example in which we can help answer questions in a language like Bengali, spoken by a quarter billion people. Just like before, we prompt the model with two examples of questions in Bengali with both Bengali and English answers.
That said, now we can start asking questions in Bengali. What's the national song of Bangladesh? The answer, by the way, is "Amar Sonar Bangla." And PaLM got it right, too. This is not that surprising because you would expect that content to exist in Bengali.
But you can also try something that is less likely to have related information in Bengali, such as what are popular pizza toppings in New York City? And the model again answers correctly in Bengali, though it probably just stirred up a debate amongst New Yorkers about how correct that answer really is. What is so impressive is PaLM has never seen parallel sentences between Bengali and English. It was never explicitly taught to answer questions or translate at all. The model brought all of its capabilities together to answer questions correctly in Bengali.
And we can extend the technique to more languages on other complex tasks. We are so optimistic about the potential for language models. One day, we hope we can answer questions on more topics in any language you speak, making knowledge even more accessible in Search and across all of Google.
All the advances we have shared today are possible only because of our continued innovation in our infrastructure. Recently, we announced plans to invest $9 and 1/2 billion in data centers and offices across the US. One of our state of the art data centers is in Mayes County, Oklahoma.
I'm excited to announce that there we are launching the world's largest publicly available machine learning hub for all our Google Cloud customers. This machine learning hub has eight Cloud TPU V4 parts custom built on the same networking infrastructure that powers Google's largest neural models. They provide nearly nine exaflops of computing power in aggregate, bringing our customers an unprecedented ability to run complex models and workloads. We hope this will fuel innovation across many fields, from medicine to logistics, sustainability, and more. And speaking of sustainability, this hub is already operating at 90% carbon-free energy.
[APPLAUSE] This is helping us make progress on our goal to become the first major company to operate all our data centers and campuses globally on 24/7 carbon-free energy by 2030. [APPLAUSE] Even as we invest in our data centers, we are also working to innovate on our mobile platforms so more processing can actually happen locally on device. Google Tensor, our custom system on a chip, was an important step in this direction. It's already running on Pixel 6 and Pixel 6 Pro and brings our AI capabilities, including the best speech recognition we've ever deployed, right to your phone. It's also a big step forward in making those devices more secure.
Combined with Android's Private Compute core, it can run data powered features directly on device so that it's private to you. To share more about how we are making computing safer with Google, let me turn it over to Jen. [APPLAUSE] JEN FITZPATRICK: Every day, people turn to our products for help in moments big and small. Core to making this possible is protecting your private information every step of the way.
Even as technology grows increasingly complex, we keep more people safe online than anyone else in the world with products that are secure by default, private by design, and put you in control. Today, I'm proud to share with you our latest advancements that make every day safer with Google. Widespread cyber attacks, like Colonial Pipeline and the recent Log4J vulnerability, threaten to put people's private information at risk, disrupt critical services like energy grids and telecommunications networks, and weaken global democracies.
To prevent future attacks, we're raising the bar for the entire industry by pioneering advanced cybersecurity technology, alerting others to security risks within their own systems, and open sourcing solutions that make the whole internet safer. Specialized teams like Threat Analysis Group and Project Zero counter serious threat actors and detect vulnerabilities across the internet. Last year, our Threat Analysis Group detected that over 40,000 users were being targeted by government-backed actors. We automatically alerted everyone, and increased protections, and blocked attacks. [APPLAUSE] Most recently with the war in Ukraine, we observed a surge of distributed denial of service attacks against websites providing critical information, like current news and evacuation resources.
We expanded our free DDoS defense program, Project Shield, to defend more than 180 Ukrainian websites, including those that belong to the Ukrainian government, news, and human rights groups. And because much of the world's technology infrastructure is dangerously outdated, we're now investing $10 billion to modernize vulnerable systems and infrastructures, secure the software supply chain, and train 100,000 Americans in digital skills, including data privacy and cybersecurity through the Google Career Certificate Program. In addition to keeping companies and organizations safe around the world, we build advanced security into everything we make to protect individual users. In the last few years, phishing scams have risen substantially. And they're responsible for 90% of recent cyber attacks.
Our built-in protections intercept these attempts before they ever reach you. For example, every day, Gmail and Messages by Google block more than 15 billion spam and phishing messages. Google Play now scans 125 billion installed apps for malware, making the entire app ecosystem safer. And our safe browsing technology built into Chrome and other major browsers now protects 5 billion devices from risky sites.
Detecting and blocking threats at this scale every day makes our AI powered protection second to none and also enables our teams to identify new areas to safeguard, which is why we're now scaling our proven phishing protections to Google Docs, Sheets, and Slides. Soon, if you're working in a shared doc that contains a suspicious link, we'll automatically alert you and take you back to safety. [APPLAUSE] No matter where they occur, all phishing attempts share a single goal-- to compromise your account, often using it as a tool to spread the attack to your network. Protecting your account starts with building the most advanced authentication technologies everywhere that you sign in. Cybersecurity experts say the single most important way to protect your account and help prevent cyber attacks is to use multifactor authentication.
However, it sometimes gets a bad rap for creating extra friction. That's why we've made our 2-step verification as easy as it gets. Whether you're on Android or iOS, just one tap on your phone, and you're in.
No six digit codes. [APPLAUSE] Over 10 years ago, we were the first consumer technology company to offer 2-step verification. And we're now the first to turn it on by default.
Last year alone, we enrolled an additional 150 million accounts in 2-step verification. [APPLAUSE] And we're optimizing the sign-in flow and account recovery experience so that we can turn on this additional layer of protection for everyone. To extend the world class security of your Google account to all your online accounts, we built Sign In with Google. Every day, 400 million people use it for secure one-click access to everything from travel sites to grocery apps.
Building an authentication system that's secure and easy to use is a massive challenge. I mean, how many times have we all had to click Forgot Password? Thanks to years of engineering investment in our password manager, 2-step verification, security keys, and, most recently, Passkeys, we've laid the path for a future without passwords. [APPLAUSE] And we're now leading an industry-wide effort to enable passwordless sign-in across every device, website, and application on any platform. I'm really proud of the work we've done to make secure authentication accessible for everyone everywhere. As we all do more and more of our shopping online, keeping your payment information safe and secure is critically important.
Today, I'm excited to announce the launch of virtual cards on Chrome and Android. Now, when you use autofill to complete your payment details at checkout, we'll replace your card number with a distinct virtual number, reducing the risk of fraud and identity theft. To give more people access to a safer way to pay online, we've worked closely with Visa, American Express, Capital One, and Mastercard to roll out virtual cards starting this summer. [APPLAUSE] We're constantly monitoring the security of your Google account to give you peace of mind.
We're now adding a safety status on your profile picture. So if anything needs your attention, we'll let you know and then guide you through simple steps to ensure your account is secure. We're relentless about protecting your personal information with the most advanced security in the world, because if it's not secure, it's not private. But protecting your privacy requires us to be equally rigorous in building products that are private by design. And I'm excited to tell you about our latest advancements in this area.
Today, computing is no longer happening just on a computer or on a phone, but across your home, in your car, on your wrist, and in the Cloud. Unlocking personalized helpful experiences while protecting user privacy in this increasingly complex environment presents new technical challenges. Building on deep research and advances in AI, hardware, and Cloud computing, we've engineered a new technical approach we call Protected Computing. At its core, Protected Computing is a growing toolkit of technologies that transforms how, when, and where data is processed to technically ensure the privacy and safety of your data. Our approach focuses on three areas.
First, we minimize your data footprint. Our focus here is on shrinking the amount of personally identifiable data altogether-- collecting less and deleting more, using techniques like edge processing and ephemerality. If the data doesn't exist, it can't be hacked. Second, we de-identify the data. From blurring and randomizing identifiable signals to adding statistical noise, we use a range of anonymization techniques to strip your identity from your data so it's no longer linked to you.
And, third, we restrict access through technologies like end-to-end encryption and secure enclaves. This is about making it technically impossible for anyone, including Google, to access your sensitive data. This toolkit of diverse techniques enables us to deploy the safest, smartest solution possible, and often stack multiple techniques to provide layered protections. Today, protected computing enables Android to suggest the next phrase in your text while keeping your conversation completely private.
It helps Pixel know when to keep your screen awake while continuously deleting ambient signals as they're processed. And it allows Chrome to alert you to compromised passwords without knowing a single one. [APPLAUSE] Later today, you'll hear more about how protected computing is enabling new ambient experiences across your Android and smart home devices while keeping your information private. In addition to providing helpful experiences, Protected Computing is essential to unlocking the potential of data to benefit society more broadly in a way that preserves everyone's privacy. For example, our Environmental Insights team is exploring how, with protected computing, we can provide local governments with de-identified and aggregated location and movement data within their cities to help reduce their carbon footprint.
Protected computing represents our deep commitment to innovating new technical solutions that make technology more helpful, safe, and private everywhere that computing happens. No one else is deploying such a multifaceted approach at our scale. And I'm excited to see all the ways our teams will apply protected computing to ensure that every day you're safer with Google. [APPLAUSE] At the end of the day, we believe that privacy is personal. That's why we continue to build powerful controls that let you choose what's right for you.
To share more about the newest ways we're putting you in control of your data, let me pass it to Danielle. DANIELLE ROMAIN: We feel privileged that billions of people trust products like Search, Chrome, Maps, and Android to help them every day. And we work hard to earn that trust by providing tools that put you in control of your privacy and that help you control your online experience. As Jen mentioned, protecting your privacy is central to making every day safer with Google. An important way that we do this is by helping you take more control over your data, including the data used to show you ads. We never sell your personal information to anyone or use the content you store in apps like Gmail, Google Photos, or Drive for advertising purposes.
We also never use sensitive information like health, race, religion, or sexual orientation for personalized ads-- period. We believe that the best ads are helpful, relevant, and safe. So later this year, we'll launch My Ad Center to give you even more control over the ads you see across YouTube, Search, and Discover.
[APPLAUSE] We're expanding on our existing ads privacy settings. So now through My Ad Center, you can directly control the data used to personalize your ads. For the first time, you can choose to see more ads from the categories or brands you like. For example, I'm interested in hybrid vehicles, so I choose to see ads from that category. You can also opt to see fewer ads from categories or brands that you're not interested in.
You'll be able to access My Ad Center through your Google account or directly from the ad. But it's not just about managing your data. We also know people want to have more control over their online presence to feel safer. So I have a question for you all.
How many of you have searched for your name on Google? OK. I do the occasional search for myself. And judging by the reaction, I can tell I'm not alone in checking to see what others may find about me online. At Google, while we strongly believe in open access to information, we also have a deep commitment to protecting people online. This is why we have policies for removing certain types of personally identifiable information from search results that we know people may prefer to keep private, such as their bank account or credit card numbers that could be used for financial fraud.
Now we're building on this by introducing a new tool to accompany updated removal policies that allows you to take even more control of your online presence. [APPLAUSE] Soon if you find search results that contain your contact details, such as your phone number, home address, or email address that you want taken down, you can easily request their removal from Google Search. [APPLAUSE] Of course, removing this information from Google Search doesn't remove it from the web.
But this is an important step in helping to protect people online. This feature will be available in the coming months in the Google app, and you will also be able to access it by clicking the three dots next to individual Google Search results. Another part of being safer online is having access to reliable information. Google Search is built from the ground up to deliver high quality information.
This has set Google apart from day one. And it's something we relentlessly invest in. And we also give you tools to evaluate the reliability of the information you come across.
One of the tools we launched last year, called About This Result, has now been used more than 1.6 billion times. This tool is available on individual search results, helping you see important context about a website before you even visit. But we want to ensure you feel in control of the information you're consuming wherever you are online. So we're making this helpful context more accessible as you explore the web beyond Search. So imagine your research and conservation efforts, and you find yourself on an unfamiliar website of a rainforest protection organization.
Before you decide to donate, you'd like to understand if it's reliable. And with just a tap, the Google app will soon surface relevant context about the website, including the site description, what they say about themselves, and what others say about them, helping you explore with confidence. You'll be able to see contexts like this on any website, coming soon to the Google app on both iOS and Android.
[APPLAUSE] As we've talked about today, at Google, we keep more people safe online than anyone else in the world with products that are secure by default, private by design, and that put you in control. Everything we make is designed with safety at the core, including the platforms with billions of users, like search and Android. And speaking of Android, the team has been building cool experiences across the many devices in your life.
And up next, Sameer will be here to tell you more about it. But first, let's say hello to our watch party in London. [MUSIC PLAYING] SAMEER SAMAT: Hi, everyone. It's great to be back at Shoreline for Google I/O. [APPLAUSE] Over the years, Android has grown into the most popular OS in the world, delivering access, connectivity, and information to people everywhere on their smartphones.
Last year alone, consumers activated 1 billion new Android phones. And as Prabhakar showed us earlier, with advances in machine learning, these supercomputers in our pocket can help us get more done than ever before. While the phone is still the most popular form of computing, people are increasingly adding all kinds of connected technologies to their lives, like TVs, cars, watches, and more. And even if those devices come from different manufacturers, people expect everything to work together without the hassle. It's got to be simple. Android has always been about people choosing the technology that works best for them.
In a multi-device world, we believe this openness is even more essential for creating the best consumer experience. So let's talk about what's new in Android 13 to bring all the benefits of this multi-device future to everyone. There are three big themes we're focused on. First, enabling you to do more with your phone at the center. Second, extending beyond the phone to all the forms of computing people love, like watches, tablets, and more. And third, making all your device