Thank you for joining us.

As we were preparing for this event, we were all devastated by

the news coming out of Las Vegas as I'm sure all of you were.

And that is coming off of a challenging past few weeks with

hurricanes Harvey, Irma, and Maria and other events around

the world.

It's been hard to see the suffering but I have been moved

and inspired by everyday heroism, people opening up their

homes and to first responders literally risking their lives to

save other people.

Our hearts and prayers are with the victims and families

impacted by these terrible events.

We are working closely with many relief agencies in affected

areas and we are committed to doing our part.

It is a true privilege to be at the SFJAZZ Center.

It's a great American institution for jazz performance

and education and it is really good to see familiar faces in

the audience.

As always, I want to give a shout out to people joining us

on the livestream globally from around the world.

Since last year and since Google I/O, we've been working hard,

continuing our shift from a mobile first to an AI

first world.

We are rethinking all of our core products and working hard

to solve user problems by applying machine learning

and AI.

Let me give you an example.

Recently, I visited Lagos in Nigeria.

It is a city of twenty-one million people.

It is an incredibly dynamic, vibrant, and ever growing city.

Many people are coming online for the first time.

So it's very exciting unless you happen to be in the Google Maps

team and you have to map this city.

And it is so - it is changing so fast and normally we map a place

by using Street View and doing a lot of stuff automatically but

it's difficult to do that in a place like Lagos because the

city is changing.

You can't always see the signage clearly and there are variable

address conventions.

Things aren't sequential.

So for example, take that house there.

If you squint hard, you can see the street number there.

It is number three to the left of the gate.

That was relatively easy.

Onto a harder problem now.

That house, that is what we see from Street View.

I think as humans, it's probably pretty hard.

Maybe one or two of you can spot it out.

But our computer vision systems, thanks to machine learning, can

pick it out, identify the street number, and start

mapping, mapping the house.

So we approach Lagos completely differently.

We deployed machine learning from the ground up and just in

five months, the team was able to map five thousand kilometers

of new roads, fifty thousand new addresses, and a hundred

thousand businesses.

It's something which makes a real difference for millions of

users there as Google Maps is popular.

And we think this approach is broadly applicable.

Let's come closer to home in parking in San Francisco.

I don't even try it anymore but for those of you who try it, we

again use machine learning.

We understand location data.

We try to understand patterns.

Are cars circling around?

And the color shows the density of parking and we can analyze it

throughout the day and predict parking difficulty and in Google

Maps, give you options.

A simple example but it's the kind of everyday use case for

which we are using machine learning to make a difference.

The best example I can think of, what we have talked before, is

Google Translation.

I literally remember many years ago adding translation in Chrome

and making it automatic so that if you land on a page different

from your language, we do that for you.

Fast forward to today.

With the power of machine learning and our neural machine

translation, we serve over two billion translations in many,

many languages every single day.

To me, it shows the power of staying at a problem, constantly

using computer science to make it better, and seeing users

respond to it at scale.

This is why we are excited about the shift from a mobile first to

an AI first world.

It is not just about applying machine learning in our products

but it's radically rethinking how computing should work.

At a higher level in an AI first world, I believe computers

should adapt to how people live their lives rather than people

having to adapt to computers.

And so we think about four core attributes as part of

this experience.

First, people should be able to interact with computing in a

natural and seamless way.

Mobile took us a step in this direction with multi-touch but

increasingly, it needs to be conversational, sensory.

We need to be able to use our voice, gestures, and vision to

make the experience much more seamless.

Second, it is going to be ambient.

Computing is going to evolve beyond the phone, be there in

many screens around you when you need it, working for you.

Third, we think it needs to be thoughtfully contextual.

Mobile gave us limited context.

You know, with identity, your location, we were able to

improve the experience significantly.

In an AI first world, we can have a lot more context and

apply it thoughtfully.

For example, if you're into fitness and you land in a new

city, we can suggest running routes, maybe gyms nearby, and

healthy eating options.

In my case being a vegetarian and having a weakness for

desserts, maybe suggest the right restaurants for me.

Finally and probably the most important of all, computing

needs to learn and adapt constantly over time.

It just doesn't work that way today.

In mobile, you know, developers write software and constantly

ship updates but you know, let me give a small example.

I use Google Calendar all the time.

On Sundays, I try to get a weekly view of how my week looks

like for once the work week starts, say on a Monday or a

Tuesday, I'm trying to get a view into what the next few

hours looks like.

I have to constantly toggle back and forth.

Google Calendar should automatically understand my

context and show me the right view.

It's a very simple example but software needs to fundamentally

change how it works.

It needs to learn and adapt and that applies to important things

like security and privacy as well.

Today, a lot of us deal with security and privacy by putting

the onus back on users.

We give them many settings and toggles to improve those.

But in an AI first world, we can learn and adapt and do it

thoughtfully for our users.

For example, if it is a notification for your doctor's

appointment, we need to treat it sensitively and differently than

just telling you when you need to start driving to work.

So we are really excited by the shift and that is why we are

here today.

We have been working on software and hardware together because

that is the best way to drive the shifts in computing forward.

But we think we are at a unique moment in time where we can

bring a combination of AI and software and hardware to bring a

different perspective to solving problems for users.

We are very confident about our approach here because we are at

the forefront of driving the shifts with AI.

Three months ago at Google I/O, our Google AI teams announced a

new approach called AutoML.

AutoML is just our machines automatically generating machine

learning models.

Today, these are handcrafted by machine learning scientists and

literally only a few thousands of scientists around the world

can do this, design the number of layers, weight and connect

the neurons appropriately.

It's very hard to do.

We want to democratize this.

We want to bring this to more people.

We want to enable hundreds of thousands of developers to be

able to do it.

So we have been working on this technology called AutoML and

just in the past month for a standard task like image

classification, understanding images, our AutoML models are

now not only more accurate than the best human generated models,

but they are more resource efficient.

So it is pretty amazing to see.

We are now taking it a step further.

Let me talk about another use case, object detection.

When we say object detection, it's just a fancy name for

computers trying to delineate and understand images, being

able to draw bounding boxes and distinguish between all of the

vehicles there, scooters, mopeds, motorcycles, and even

pick out the bike in front.

It has a lot of practical use cases.

The Street View example for Lagos works based on

object detection.

Google Lens, which you will hear about later, as well as our

photography in Pixel, uses object detection.

This is really hard to do.

The best human generated models we have only have a 39% accuracy

but our AutoML models as of the past couple of weeks have

reached around 43% accuracy and they are constantly

getting better.

So the rate at which we are seeing progress with AI is

amazing which is why we are really excited about combining

it with our software and hardware to bring it together

for our users.

Let me give you a concrete example.

I was recently very inspired by this tweet from a journalist who

was attending the Little League World Series.

As you can see, there are two Little Leaguers here, one from

Dominican Republic and the other from South Dakota, and they are

talking to each other using Google Translate.

It's a great example but I looked at it and I feel like we

could do a whole lot better.

Computing needs to evolve where this happens in a more natural,

seamless, and a conversational way.

Later today you will get an early glimpse of how we can push

this experience further by thinking about AI, software, and

hardware together.

I couldn't be more excited by the direction our hardware team

is taking in approaching their work here.

So to give you a lot more insight let me welcome Rick onto

the stage. Thank you.

>> RICK OSTERLOH: Hi everyone.

I want to welcome everyone here today, Google fans, the media.

In the front row we have some of our top sales reps from Verizon

and other partners.

It's great to have you all here.

Yes. Welcome. Welcome.

We've got a lot to show you today, including a

few surprises.

So I thought I would quickly set the scene and explain what we've

been working toward and just how it ties into the vision that

Sundar outlined.

A year ago, we stood on a slightly smaller stage and

introduced the world to Google Pixel, Home, and Wi-Fi,

Chromecast Ultra, and Daydream View.

This was the first generation of hardware made by Google.

So much has happened since then and we have made tremendous

progress in our first year.

For starters, our team is getting a lot bigger.

We're bringing on two thousand very talented engineers for HTC

along with some important IP.

It's an amazing team that has created a series of industry

first innovations and by working more closely together, we'll be

able to better integrate Google hardware and software.

And our products have built up a lot of momentum going into our

second year.

Let's take a look.

>> SPEAKER: Okay Google.

>> SPEAKER: Okay Google.

>> SPEAKER: Okay Google.

>> TY BURRELL: Okay Google.

Play the Soundtrack to Spaghetti for Pepe.

>> SPEAKER: Google set to hold a product event in San Francisco.

There are a ton of new announcements.

>> SPEAKER: Google is showing off new hardware.