Inside Search Event

Uploaded by Google on 14.06.2011

[whimsical music]
Singhal: Thank you, Gabriel.
And thanks, everyone, for coming today.
It's my pleasure to share
the numerous exciting things
that are happening inside Search at Google.
So let's get started.
Today we will talk a lot
about how Search is breaking barriers
between you and the knowledge you seek.
And in that light, we will talk about
all the innovations that are happening
in the Mobile Search space
and in the Desktop Search space.
And we will talk about what's next
for Google Instant.
So let's first talk about knowledge.
What is knowledge?
Knowledge is defined as the sum total
of what is known,
facts and information
in a given field or in general.
Knowledge emerges from you having facts
and information you need.
Every time you are missing a fact
or a piece of information,
your train of thought is derailed
and you are slower to acquire
the knowledge you want.
With Search, we strive to make sure
that there are no derailments
in your train of thought.
And every time you are not at your desk,
you can't type a query,
you can't think of the right query,
or the Internet is slow,
your train of thought will get derailed.
Search is all about removing these barriers
between you and the knowledge you seek
and getting to you
all the knowledge instantaneously.
Now let's talk about one of the biggest barriers
that we face in our busy lives today
between you and your quest for knowledge.
It is the fact
that you're not in front of a computer.
However, thanks to the modern mobile devices,
this is not a problem anymore.
Your quest for knowledge does not stop
just because you stepped away
from your computer.
Even when you are out and about
doing things that you love--
playing with your kids, spending time with friends--
you need knowledge all the time.
And our data shows this.
What I am about to show you
is our data telling me
that your quest for knowledge
is eternal.
And our data reinforces my belief in this--
humanity's fundamental need to learn.
What you see here
is our traffic graph over a week.
We start out strong
with a lot of Desktop Searches on Monday.
This is a Desktop Search graph.
And the traffic stays strong
throughout the week.
However, as the weekend approaches on Friday,
people start doing other things outside of work.
They are not at their desk as much,
and our Desktop Search traffic
goes down a little bit.
And you know what,
it goes down further on Saturday
because you are spending time with your friends and family
or doing other things that you like.
And on Sunday, it's still low,
but people start preparing for work again.
Now let me show you the data that says
that even though Friday, Saturday, and Sunday arrives,
your quest for knowledge did not take a break.
What you see here in the red graph
is the same week shown for Mobile Searches.
And you see it--we start strong on Monday,
and we keep going strong for Mobile Search traffic
Monday, Tuesday, Wednesday, and Thursday;
however, Friday arrives,
and you see nothing's going down.
You are out and about.
You are doing things that you want to do.
At the same time, you are pulling out your mobile devices
and searching in your quest for knowledge.
And you know what, on Saturday,
you are out and about even more
and you're searching even more on mobile.
And the trend stays strong on Sunday.
Let me dissect this graph
for a 24-hour period.
Here is the blue graph again.
This is desktop traffic
dissected over a 24-hour period.
And you see, you get up.
People are not searching that much
at 5:00 in the morning.
Then you get to work
between 7:00 and 9:00 a.m.,
and yes, you start searching.
You need knowledge to get your work done every day.
And then lunchtime arrives. You get hungry.
You walk away from your desktop
and you see that tiny dip around noon.
And you come back after lunch
and with, you know,
with food and everything you have all the energy
and you start searching again
in your quest for knowledge.
And then evening arrives, and you go home,
and our traffic slows down.
You do observe that people are working later and later
these days, and our traffic stays strong
almost until 9:00 p.m. these days.
So it looks like we're all turning into workaholics.
However, let's dissect the same graph
for Mobile Searches.
And what you observe here
is that, yeah, you wake up in the morning
and you search on mobile.
However, when you get home
and at-- starting at 9:00 p.m.,
the Mobile Search traffic
for the day peaks.
You know what, you went home.
You are not searching on desktop
but you keep searching on mobile devices.
And I truly hope after 9:00 people--
after 9:00 p.m., people are being productive
in the right sense of the word.
Not only that,
you observe this wonderful phenomena
that, around noon,
when the desktop traffic went down,
the mobile traffic actually sees a tiny bump.
And you know what happens?
You go out to a restaurant.
You ordered the food,
and who wants to talk to your friends?
You all pull out your mobile devices
sitting on the same table
and you start playing with your mobile devices
and searching.
Then your food arrives
and you start eating with your left hand
and you keep searching with your right hand.
And that's what we observe in our data,
that your quest for knowledge
does not slow down
every minute you are awake.
And this is very exciting for us
because we see all this mobile traffic
growing on top of our desktop traffic.
So let's talk about this mobile explosion.
What I am showing you here
is early Google traffic growth from desktops.
What you see here is a three-year graph
of early Google Desktop Search traffic growth.
I was there, and it was like--
seeing this traffic grow,
it was like a child was born
and the child starts growing up
at an amazing pace.
Then comes first Christmas
and you see the first dip.
People go on vacation.
They don't search as much on desktop.
They come back in January and start searching again.
Then comes next Christmas
and you see a valley in the graph again.
And people come back in January
and they start searching again.
And then you see some flattening out
because now the child is big enough,
the traffic is plentiful,
and summer vacations start registering
on our traffic graph.
We call it "the summer slump."
And then people come back in fall
after their summer vacation
and they start searching at an amazing pace.
And then you see the huge Christmas valley
third year in a row.
Now, fast-forward a few years
and our second child, Mobile Search, is born.
And let me show you
the growth of our second child
a few years later.
And the second child's born.
It's growing strong as expected
like the first child,
but here is the beauty of this.
This one doesn't slow down over Christmas.
If anything, this one searches more over Christmas.
There is no summer slump or Christmas calm
in mobile traffic.
It just keeps growing at an amazing pace.
Over the last two years,
we have seen 5X growth
in our mobile traffic.
And it's incredibly satisfying for me--
me who was there
when our desktop traffic was growing at this pace--
to see my second child, Mobile Search,
grow at the same amazing pace.
What is fueling this success
for Google in mobile?
The principal step we put in place
when we were building our Desktop Search
have unwittingly become the foundation
for Google's success on mobile.
From the early days of Google,
we were completely focused on the users
and completely focused on getting them
the most relevant result as the first result.
We say at Google,
"Our job is to get a hole in one every time."
It's like playing a billion rounds of golf
and striving to get a hole in one every time
and getting it right more often than not.
Imagine how hard that is.
And our focus on users and relevance
has become that foundation for success on mobile,
because on mobile,
the screens are smaller.
Users are going to see just one result or two
It's even more critical
that you get the first result right.
And we laid the foundation
for our success on mobile unknowingly
while working on Desktop Search a decade earlier.
And in this willing foundation,
there are multiple ingredients.
There is, of course, relevance, that I talked about.
It's Google's relevance algorithms
power the top results that are ever so more critical
on mobile.
And you add to that
the simplicity of Google's interface--
a simple front page
and very simple, understandable result page.
On a mobile device, where you have to touch things,
this simplicity becomes even more critical
than ever before.
And you know that at Google,
we are obsessed with speed.
At Google, we often say,
"Speed is still the killer app."
And on mobile devices
where the networks are still somewhat slow,
this focus on speed
becomes the third pillar
of our successful foundation for mobile.
Now, to show you what goes
into building a successful mobile search system
like Google's,
let me invite Scott Huffman,
Director of Engineering for mobile,
and Steve Chen,
Director of Product Management for mobile,
onto the stage.
[applause] Huffman: Thank you.
Thanks, Amit.
Well, Amit talked about how mobile search
is breaking down barriers to knowledge.
If you went back a few years and looked at the mobile searches
that we were getting then, you might say,
"Gee, Amit, that's kind of a grandiose statement."
Besides not getting very many mobile searches then,
the mobile searches that we did get,
honestly, were kind of boring, right?
People did pretty simple things.
They would look up a stock price.
They would look up the weather, things like that.
We saw people just doing
very simple search tasks on mobile phones.
An amazing thing happened when the mobile phone
became a powerful computer in my pocket,
with the advent of devices
like the Android and iPhone and others.
And sort of overnight, we saw our mobile search stream
really transformed to something very interesting,
where all of a sudden, we were still seeing people
doing some simple things on mobile,
but we were also seeing them do very complex search tasks,
things like planning a trip, doing research,
things that required complicated searches
to get done.
So, given that today we see that whole range
of interesting search tasks,
from simple to complicated, on mobile,
how do we think about designing
a great user experience for Search on mobile?
Well, at Google, we always start from the basic search process.
And if you think about it, it's pretty simple.
First of all, you need to tell us
what it is you're looking for.
Whether it's simple, whether it's complicated,
whether it's one word, many words,
whether you want to type it,
whether you want to just say it to your phone,
you've got to enter that query and let us know.
And in mobile, we're constantly thinking about,
"How can we make that process of entering a query easier?"
And then after you give us the query,
you get back your results,
and you need to use those results
to get the information that you were looking for.
And so, again, we're constantly thinking about,
"How can we make the process of getting those results
"and using them to solve your problem,
get your answer easier?"
So that sounds simple.
But what Steve and I wanted to do
is go through a few examples
of very different kinds of queries that we see on mobile
and point out some of the things that we've done to make it
easier to enter your search and easier to use your results.
So let's jump in.
So the first demo we're gonna do is a local demo.
After we give our demos here
and get lots of good questions from all of you,
I think Steve and I are probably gonna be hungry,
and we're not familiar with this area,
so we're gonna need to find a restaurant for lunch.
What Steve's showing here is our refreshed home page
for mobile.
What launched today is this set of icons at the bottom
that give you instant access into local queries.
So let's see how hard it is
for Steve to enter our lunch query.
There, he just did it.
Pretty easy to enter that query.
And his query is more
than just the word "restaurants," of course.
We've also passed through our location.
What we see now is our new search interface
for mobile local searches, again, launching today.
And you see at the top, we have a map
that shows the cluster of restaurants around here,
and the blue dot is us.
So, again, trying to make it easy to see
what you're probably looking for with this query.
As Steve begins to scroll through the results,
you'll see that the map actually changes in place
to give me the context of where each restaurant is
in relation to me.
And you see the listing
lists some other information about the restaurant.
And let's pick this restaurant here.
As Steve taps through, I want to point out
how quickly that result came up on the screen
to give me more information
about this particular restaurant.
And that happened because we're Ajax
and the power of the modern browsers on these devices
to make sure that all the information
that a user might need as they go through these local results
is available right away.
So there's a local example.
Easy to enter that query.
Easy to consume just the kinds of information
that I want from it.
So next, let's do a simple search task.
Maybe while I'm at lunch, I'll do a query that I do a lot.
I want to see how the market's doing,
so let's do "S&P 500."
So let's see how hard this one is to enter.
Steve enters S, and boom.
So what happened there?
Steve only had to type one character,
and he got the query that we were looking for.
What happened there is that we noticed a while back
that people do do repeated queries on mobile.
Through the day, they are doing things like checking the market,
checking the sports scores, those kinds of things.
And so what we have done is to integrate
query suggestions from your history
with our normal kinds of query suggestions.
And so, when Steve typed S, because in my history,
I have the S&P 500 query, that's what came up.
With Google Instant on mobile,
Steve didn't even have to hit return,
and the results came up.
But the real reason I wanted to do this query
was to point out the results here.
If Steve just scrolls a little bit,
you'll see that, of course,
we have all the usual kinds of results.
You can drill into any of these.
But for this kind of query with a strong intent,
we've returned at the top here
an interactive widget, basically,
that gives me the information
that probably I'm looking for about the S&P 500.
And you can see, with a tap, for example,
Steve can go and look at how this index
has been doing over the year.
If he slides, right, we'll see some news for the index,
market overview, and some other things.
This is something we launched back in March,
trying to make, again, this kind of strong-intent query
very easy to consume the information that you want.
So we've done a local one. We've done a simple one.
Let's do a little more complicated search.
And for this one, I'll just set this up this way.
I have a feeling that I'm gonna be making a trip to Russia soon.
The reason is, while we were backstage,
Amit mentioned to me that if the demos don't go well,
he's sending me to Siberia.
So I think it's going pretty well.
So I'm hoping that he's only gonna send me
to our Moscow office,
and I'm gonna need a hotel while I'm there.
So let's do "Hilton Hotel, Moscow, Russia."
So let's see what happens as Steve starts to enter this.
As he types H-I, not surprisingly,
the whole query "Hilton Hotel, Moscow, Russia"
is not one of our top-three things.
But notice that along the side, we've added these pluses.
This, again, is a new feature launching today,
something that we've had in the Android search widget
and the iPhone search app for a while now.
We call it a "query-building capability."
So what Steve can do is tap on the plus,
and he goes to the next piece of the query
and gets suggestions from there.
So he's done "Hilton Hotels."
He types M-O-S, and there's "Hilton Hotels, Moscow, Russia."
So we counted. This query has 26 characters.
And Steve did, actually, eight taps to enter that query.
So, again, making it easy
to enter these more complicated searches.
Let's do one more like that.
While I'm in Moscow,
I'm gonna want to hopefully get a chance to do some sightseeing.
So let's do "Moscow top attractions."
So, again, Steve's gonna use the query builder
to build this query.
Okay, so there it is.
What I wanted to point out about this query is,
for a query like this,
you get a lot of different kinds of results,
and the textual snippets are trying to help me
kind of choose what I want, but what Steve can do here
is to actually tap on the spyglass icon
that's along the side.
This, again, is something we launched a couple months ago,
and what this does is take me instantly
into a visual search mode,
where what we're seeing here is previews,
screenshots of each of the results.
And so, by looking at those,
I can see kind of what kind of result it is.
Uh, and so, as Steve scrolls along,
this is the kind I was really looking for.
Something with pictures of all the attractions
and little descriptions of each one.
And so then, of course, Steve can tap
and get to that.
So for these more complicated queries,
we've used the Query Builder to make it easier
to enter your query.
And then things like this feature
which we call Mobile Instant Previews,
to make it easy to use the results
and get to the result that I want.
So I'm gonna show one more.
Uh, and for this one let's just say
that I'm in the airport,
and I'm waiting for my flight to Moscow.
And for this one, I'm gonna pull out my tablet.
In that list of attractions, one that I saw
that was pretty interesting to me
was St. Basil's Cathedral.
And so I--it looked very beautiful,
so I want to see some more.
What we're showing here, this is kind of
a sneak preview of a refreshed tablet search UI.
This'll be launching in a few weeks.
And what I want to point out here
is just that we've used the screen real estate
a little better, made the results bigger,
make all the tap targets easier.
Just an easier-to-use search experience
for the tablet.
When Steve taps through to Image Search,
This again is our new Image Search UI for tablet.
And what you'll see here is that
the images are bigger
and taking advantage of the beautiful screens
that these tablets have.
The images are actually aligned
so that it's very easy for me to kind of scan through them,
and as Steve begins to scroll--
we actually have infinite scroll here,
so that Steve can just keep going,
look at as many of these beautiful pictures as he wants.
And this is really recognizing that, on these tablet devices,
people really do like to consume a lot of information.
And so we're making that easy here in Image Search.
All right, so I've shown four different examples
where, for each one,
we've done pretty different things
to make it easy to enter your search
and easy to consume your results.
Now, I actually lied before when I said I was done.
I want to show one more example,
because I'm worried that everyone will say,
"Well, Scott, that was too easy."
Right? "You know, all those queries, they had words."
Right? "They were in English."
And the reality is, when I get to Moscow,
I might see something like this.
I'm gonna walk down the street.
And so the question is,
how can Google help me with this?
Well, today we're happy to announce
that we're launching Google Goggles with Translate
with the ability to translate in Russian.
And so what's Steve's gonna do here is pull out his phone,
take a picture of this...
and up comes the characters.
Steve can tap the translate button.
And here, I can see that it's a list of cold drinks
with Coca-Cola, soda, mineral water,
orange juice, and so on.
So...I've talked about how we've made it easy
to enter queries,
easy to consume results on Google.
There is one key way that we've made it easier
to enter queries on Goo-- on mobile
that we haven't talked about,
and that's searching by voice.
To talk about that, I'd like to invite Mike Cohen,
who leads our speech efforts at Google,
up to the stage.
Cohen: Thank you, Scott.
Amit earlier today was talking about
breaking down all barriers to knowledge.
And Scott just showed you some really, uh,
concrete ways in which we're making progress
in that direction.
Things that make it faster, easier,
more seamless to enter your query.
So I'm gonna be talking about the same problem,
but a different dimension of that problem,
which is allowing users to just say what what they want,
to use their voice.
Now, arguably, speaking is the most natural way
that we learn to express our needs.
Certainly, we learn to speak before we learn to type.
Um, when it comes to mobile,
speech input is particularly important.
Um, you know, when you're mobile, you're on the go.
You're--you may be walking down the street,
riding your bike,
driving your car.
So scenarios that allow you to talk
can be especially important.
So we've been working on enabling speech interactions,
and we're doing this now for Search
and many of the other things that you do
when you're mobile.
And we're seeing tremendous growth.
So this stuff is being used a lot.
So this shows the growth just in the past year
for mobile speech inputs.
And what we see is,
in the past year,
the volume of mobile speech inputs
has grown by a factor of six in one year.
To give you an idea of the volume of speech,
every day these days,
roughly two years of nonstop speech
comes into our system.
Okay? Two years of nonstop speech every day.
Now, to me, that's very, very gratifying.
I came to Google seven years ago
to start up a speech effort,
but actually, I'm a lot older than that.
I've been working on nothing but speech technology
for the last 27 years.
And the user value that we're seeing here now--
Back 27 years ago,
this was just a very long, distant, future dream.
Now we're getting there.
But still, this is just the beginning.
So if this is just the beginning,
what does it really take to succeed
with speech interfaces?
So my definition is,
any time a user is interacting with the Web,
interacting with a device,
and they prefer to be speaking,
I want them to be able to speak
and get what they want.
In order to accomplish that,
there are two foundational capabilities
we need to offer that make that possible.
The first one is accuracy.
The accuracy needs to be as close to perfect
as we can possibly make it.
We don't want the--
we don't want the choice of the modality of interaction
to add any friction to that interaction.
We want it to be seamless.
And the second part of it is ubiquity.
And by that what I mean is,
it needs to be available everywhere.
Has to be available for every application,
every language, every platform,
every device.
So what we're really trying to do
is change the user's mental model from today,
where people may say, "Oh, let's see.
Can I use speech for this application?"
To making it a basic habit.
People feel like talking, they just assume they can talk,
and be understood, and get what they want.
So I'm gonna mention progress
towards these foundational capabilities.
Um, first, accuracy.
So we're constantly working to improve our accuracy.
What we see here is basically telling--
something telling us that the higher our accuracy,
the more repeat usage we get.
So we offer higher accuracy,
users use it-- speak to the system once,
and they keep coming back,
and they become repeat users of spoken input.
So we're constantly working to improve our accuracy.
And we're making a lot of progress.
To give you an idea of what do we do
to improve accuracy,
we actually feed our speech recognition system
massive amounts of data.
And from that, the system actually learns.
It learns about
all of the different kinds of pronunciations
that may come into the system.
And it learns about all the different ways
people may put together words into phrases and sentences
to express what they want.
So just to give you an example
of the scale of this training,
the scale of this learning that the system is doing,
just for U.S. English,
just to train the part of our model
that learns about the phrases
and how people may put their words together
to express their needs,
we feed the system
roughly 230 billion words' worth of data
from real queries.
Now, that learning process takes many CPU decades
of, um, processing
just to do that learning.
Now, I'm a really impatient kind of guy,
and I don't like to wait many decades
for the result of each experiment.
So luckily, we get to run
many, many thousands of computers.
And we come in the next morning and we have a result.
So the other foundational capability
is ubiquity.
Being available everywhere.
One aspect of that
is being available in every application.
So a little over a year ago,
we released something called voice input.
And voice input is basically, um--
In Android, whenever the keypad pops up,
you also have this microphone button.
And that means whenever the keypad is there,
if you prefer to speak, hit the microphone button
and you can speak rather than type.
So that means for all applications,
for all Web pages,
any time you're filling out a form,
when you prefer to speak, you can.
The developer didn't have to do anything
to make that happen.
And in many cases, the developers
may not even know that such a capability exists.
They may not even realize
that their application is speech enabled.
But by virtue of their application being on Android,
it's speech enabled.
Another aspect of being ubiquitous
is being in all the world's languages.
When we first released Voice Search
over two years ago, it was available
in the United States and Canada,
5% or 6% of the world's population.
Today we're available
in roughly 27 languages and dialects around the world.
And we estimate we cover something like
2/3 of the world's population,
close to 5 billion people.
And we're constantly releasing new languages
to cover more and more of the world's population.
So those are a few of the important steps
that we've been taking
in order to get speech interfaces
available everywhere and useful.
So to see where we're going next,
with speech and with other innovations,
let's invite to the stage Johanna Wright,
the Director of Product Management
for Search.
Wright: Thank you, Mike.
Mobile has opened a world of possibilities.
Mike and Scott talked about using these innovations
on the phone to break down barriers to knowledge.
They talked about things like Voice Search,
and searching by image using Google Goggles,
where you can just take a picture,
search for these restaurant results,
and see them right there on your phone.
But we look at our phone,
and then we look at our computer,
and we say, why is it
that we can't take these same innovations
and bring them back to the computer,
and use them to break down barriers to knowledge
there as well?
And that's what I'm going to talk about today.
I'm going to talk about bringing mobile innovations
back to the desktop.
Mike also had a really interesting point
about ubiquity.
And to really understand this, I have a family story
that brought it home to me.
So here's the story.
In my family, we're all trying to learn Spanish.
I actually think I am the family expert in Spanish,
but my son's in Spanish kindergarten.
My husband doesn't know Spanish.
But we're all trying to learn.
So the other day,
my husband and I were out on a walk.
And my husband looks at me and he said,
"Johanna, how do you say 'squirrel' in Spanish?"
So I said, "You say 'ardilla'."
And what'd he say?
He said, "Well, no, that doesn't sound right."
What does he know?
But how do you convince your husband
that you know something?
So he has a Droid.
And I said,
"Take out your phone and just say,
'Translate squirrel to Spanish.'"
So what'd he do?
He said, "That's never gonna work."
And I said, "No, just do it."
So he took out his phone, and he said,
"Translate squirrel to Spanish."
And as you can see,
the way you say squirrel in Spanish is "ardilla."
And I was right.
[audience laughs]
So this had two cool points for me.
First, I was right and he was wrong.
The second cool point was, in fact,
Voice Search really did work,
and you could just use it,
even when you wouldn't think to.
And this is why Mike's point about ubiquity
matters so much.
The reason my husband didn't think
that he could just use his phone
and ask this question
was because Voice Search isn't everywhere.
He's not used to seeing it.
He's not used to just speaking his searches
and having Google get him the results.
And that's why, today,
we're announcing Voice Search on the desktop.
[cheers and applause]
I have with me here Bill Byrne,
our speech interface engineer,
who's going to help us demonstrate how this works.
Byrne: Thanks, Johanna.
So let's get right to an example
to demonstrate how this works on the desktop.
So Johanna and I were talking the other day
about our kids' school projects,
and Google searches are always
a big part of school projects.
We talked about solar systems
and astronomy and things like that.
So why don't we try something like that right now?
I'll just click on the microphone
that you see now in the search box,
and say my search out loud.
Pictures of nebula.
And as you see...
Wright: Pretty cool.
Byrne: Yeah, just as if you'd typed it.
Wright: Why don't you try something harder?
Can you try
"recipe for spaghetti with Bolognese sauce"?
Byrne: Wow, that's a long one. Let's see what happens.
And again, I'm just speaking
right to this laptop's built-in mic.
Let's try it.
Recipe for spaghetti with Bolognese sauce.
Nailed it.
Wright: Okay. Try "Worcester, Mass."
Byrne: Worcester, Mass.
Well, I'm glad I don't have to spell that.
Let's see.
Worcester, Mass.
Is that right? Yep.
Wright: Pretty good.
Byrne: You know, Worcester, Mass. reminds me
of another thing that kind of sounds the same.
Not quite the same.
I don't think it's spelled that way, either.
It's Wooster College. Let's try that.
Wooster College.
Yep, got that too.
Wright: So what you're seeing here
is an example of the big data that Mike talked about.
And that is, as we hear more and more searches,
our technology gets better and better.
And you'll be seeing this over time.
Voice Search will just get better and better.
Bill, can we give one last example?
Byrne: Sure, and why don't we combine again
two technologies at Google
that make things so much nicer for the user.
We can combine Translate with speech recognition.
And this is something maybe your husband
would like to do as he's learning Spanish
with your family.
So let's try a longer Translate example.
And again, we'll see what happens
if we can get this right.
Translate to Spanish,
"Where can I buy a hamburger in this neighborhood?"
Oh, I think it got it.
Wright: Pretty cool. Thank you, Bill.
Byrne: Sure.
Wright: We're really excited about Voice Search
and how it can take these really hard queries
and just search for them for you.
But sometimes, you don't have the words.
Sometimes you don't have the words
to describe what you're trying to search for.
Sometimes, like Mike talked about--
I mean, sorry, what Scott talked about--
something like Goggles is a better technique.
So let's take this picture as an example
that we have here.
How would I-- this is a family photo.
It doesn't have any place on the Internet.
It doesn't have any tags.
Doesn't have a geotag.
It doesn't have a date associated with it.
But I'm trying to remember where I was on vacation.
How could I do that?
How could I ask Google that question?
Well, if I try to describe it in words,
what would I say?
I would say there's a guy.
He's on a path on an island
with a beach behind him.
A rocky cliff.
Well, it's pretty hard to describe this
in a way that Google could understand.
But I have the picture.
And if I have the picture,
why not just use it?
And that's why today, we're also announcing
Search by Image on desktop.
I have with me here Peter Linsley,
our lead Product Manager on Image Search,
who's gonna help us see some demos of how this works.
Linsley: Okay. Thanks, Johanna.
So let's just pull up that image we were looking at.
And I want to give you a little bit
of the back story behind this image.
This actually came from a software engineer at Google.
His name is Gabe.
And he was on holiday in Greece ten years ago,
island hopping.
And he took a whole lot of digital photographs.
And fast-forward ten years,
he was going through his collection,
and he couldn't quite recall where this image was taken.
Now, how would you normally try and solve this problem?
You'd probably show it to a bunch of friends,
much like I might ask you guys right now,
"Does anybody recognize this place?"
But what is the metaphor when you have that query
for a search engine?
Perhaps what you want to do
is to be able to pick up the image,
drop it in the search box,
and get results.
Pretty cool.
So let's take a look
at exactly what's going on here.
As Johanna mentioned earlier, this is an image
that hasn't been published online.
We do not have this image in our index.
Yet we were somehow able to figure out
that this is in fact Nea Kameni,
island in Greece.
We've returned authoritative web results.
There's Wikipedia there,
so you can go off and learn more about it.
And we've also returned visually similar images
which appear to be the same island,
but take--photographs
presumably taken by different people.
Wright: Pretty neat.
Can you show us another example?
Linsley: Sure.
So we have another good example
that came from internal testing.
This rather blurry-looking logo
was picked up by Jack, a sales engineer at Google,
at a conference that he attended several years ago.
And there was something about the logo
that caught his eye.
He'd long wondered about the origin.
So again, during the internal testing of Search by Image,
he was able to pick it up,
drop it in the search box,
and there's his answer right there.
Turns out this is the logo
from the Hong Kong Monetary Authority.
What's really cool is you can click
on these results here,
and verify that, in fact, that is the case.
Right from the horse's mouth.
Wright: So Peter, these examples you've shown
have been interesting, but I'm wondering
if this technology can help me with a problem
that I have.
So at Google and at work,
there's all these pictures floating around.
They have cute little animals on them,
and they have these block white letters
with, um, with captions on them.
And I'm trying to figure out what's going on here.
Sometimes I feel like an old lady at work,
and I just want to know what this is all about.
Linsley: Okay, well,
I think I know what you're talking about.
And fortunately, because I anticipated this question,
I prepared this for you earlier today.
Now, some of you may recognize this guy.
Some of you don't.
I get forwarded these images all the time.
So we'll go ahead--
I have to admit, you know, it's very difficult
to stay on top of all these latest internet memes.
But let's go ahead and drag and drop this
into the query box.
So next time, Johanna,
you come across an image like this,
just drop it into Search by Image
and you'll get results.
So by looking through the results page,
we can see that this is the Y U NO Guy.
And you can click on Know Your Meme
and find out more about it.
More about the origin.
Why it went popular.
How it went popular.
And I want you to look
at the visually similar images here.
This is interesting, because this is where
you kind of get the crux of it.
That this is an image, an internet meme,
that gets reused, where people change the text
across the top and the bottom
to add their own kind of funny captions.
But what's really important to note here
is that, despite this being-- the meme that I gave was,
you know, directed at Johanna,
but the similar images are, in fact, different variations
of the same meme.
It's almost like we were agnostic
to the text that you see there.
Wright: So Amit talked
about breaking down barriers to knowledge.
But what seems really important here
is that Google can also make me cooler.
Peter, can you help us-- Mike talked
about the technology behind search by voice.
Can you help us understand the technology
behind Search by Image?
Linsley: Sure. So Search by Image uses
a lot of the same features you see in Google Goggles
on the phone,
except it enhances that for the Web scale.
Here's how it works. We take an image.
We break it down into fundamental features,
whether it be points or lines, shapes, texture,
and so on and so forth.
This then formulates a query,
which we send to our back end
against billions of images we have,
as we found on the Web.
This then returns results, which we can rank to the end user--
something like this.
Now, just to wrap up, I want to tell you a little bit
about how this feature is getting rolled out.
So it's rolling out globally on
over the next few days.
And when you see a little camera icon in your search box,
that means you have access to Search by Image.
And there are four ways--
Wright: Could you amplify?
Linsley: Yeah, so there are four ways
you can access this feature.
You can copy and paste an image URL.
You can upload images directly from your desktop.
You can also drag and drop images
right into the search box, as I did in the demo.
And finally, we're also announcing
Chrome and Firefox extensions,
which mean that Search by Image is one click away from images
as you discover them on the Web.
Wright: Thank you, Peter.
Every picture has a story,
and we want to help you discover that story.
A search should not be limited
to just one way of expressing yourself.
If you're finding it challenging to type,
you should just be able to say your search.
If a picture tells what you're trying to search for better,
just use that picture.
In Google Search, we're committed to getting you
the best results in the fashion
that's as effortless as possible for you.
One important component of effortless search at Google
is speed.
And I'm gonna invite Amit Singhal back up on stage
to talk about what's next for Google Instant.
Singhal: Thank you, Johanna.
This is very exciting.
We are breaking barriers to you getting
what you want on mobile with easy query mechanisms
and on Desktop when speaking is the most natural thing to do.
Or maybe just uploading a picture.
But what is the biggest barrier that you face
to getting the knowledge you want?
It's time.
Our lives are getting busier and busier by the day.
We have more things to do than we possibly can,
and time has become the most precious commodity
in our modern lives.
Last year we launched Google Instant
to save you time when you search.
Let me just quickly recap.
Google Instant search is the technology
which gives you results as you start typing your query.
You can type a letter or two.
We would predict the query that you are probably
going to ask.
And not only that--
we bring the results of that query
to your result page right away.
Google Instant saves all of us a lot of time.
In a repeated measurement, Google Instant saves,
on an average, two to five seconds.
And in the last year, we have been focused
on bringing Instant everywhere.
We want every user in the world
to have the benefit of our Google Instant technology.
And today I am happy to announce
that Instant is available in 32 languages on 69 domains,
including 16 Latin-American domains today--
on Desktop, mobile devices,
tablets, Chrome,
and, of course, toolbars.
Now, we want to make sure
that you get the joy of Instant
no matter what you're looking for.
And that's why I am also happy to announce today
availability of Google Instant in the coming weeks
on Google Image Search.
This will be rolling out in the coming weeks,
but let me just give you a fun sneak peek.
[upbeat melody plays]
Isn't this beautiful?
Now, let's talk about speed some more.
I keep saying we at Google obsess about speed.
And when I tell my friends that the Web is as slow
as molasses in January,
they say, "What are you talking about?
I've got broadband. Things are fast."
Until I remind them that flipping channels on their TV
is much faster than loading a Web page.
And you can easily flip through pages of a magazine,
and the Web doesn't feel like that.
And we at Google are never happy
with how fast this Web can be.
Until we can make the Web as fast an experience
as it is to flip pages through a magazine,
we will not be satisfied.
We at Google think about the entire search process,
and we measure every millisecond.
Where does this millisecond go?
And over repeated measurements, we have found out
that it is hard for people to find the right query
and enter a query.
On an average, that process takes about nine seconds.
The time it takes Google servers to return the results
is negligible compared to the time
it takes our users to pick up the right result,
which, on an average, is 15 seconds.
With numerous innovations that we have launched in Search
over the last few years,
we have saved the time it takes users
to type a query using our Autocomplete technology.
And with last year's launch of Instant,
we saved users the time it takes
to select the right result,
because the results are right there in front of you
as soon as you hit a key.
Now, we saved users a lot of time
both with Autocomplete and Google Instant,
but you're not done yet.
As soon as even when you select the result
that you want, the page is not there.
You don't have the information you need.
It would take another five seconds, on an average,
for that page to arrive at your browser.
Yes, it takes an average Web page
five seconds to load,
based on repeated measurements that we have done.
And you're asking, "So what can you do about that?
"It's my computer going to the open Web
"and fetching a page from a third-party Website.
What can Google do about that?"
Which brings me to the next announcement of the day.
Today I am very happy to announce
Google Instant Pages.
With Instant Pages,
sometimes when you'd click on a result,
the page would be just there, instantaneously.
And rather than telling me--
Rather than me telling you all about this,
let me just show you.
I have with me Alex Komoroske, our Product Manager for Chrome,
who will demo Instant Pages for us.
Alex? Komoroske: Thanks, Amit.
Now, before I show you Instant Pages,
I just want to take a moment to remind ourselves
what the Web is like without Instant Pages.
So I'm originally from the D.C. area,
so I'm a big fan of the "The Washington Post."
So I go to Google. I type my search.
I scan for the result I want.
I click, and I wait.
Even on this really fast Internet connection,
it takes time for the text and images to load.
Until now, it's just been a fact of life.
But now let me show you Instant Pages.
What I have here is two identical versions
of Google Chrome running side by side.
The one on the left has Instant Pages disabled.
What I'll call "old Google."
The one on the right has Instant Pages enabled.
Now, I also have an extension installed
that will mirror my actions from one browser to the other.
If you watch closely, you'll see there is a slight bit of lag,
but because I'm controlling the browser on the left,
I'll actually be giving old Google a head start.
I'll also go ahead and clear my cache on both browsers,
just to show that I don't have anything pre-cached.
So let's go ahead and do that same search from before.
Now, the first thing to notice is that there really
isn't anything to notice.
The search results page looks exactly the same.
But the difference will come when I click.
And what you'll see is that the browser on the right
will load the page much more quickly
than the browser on the left.
[cheers and applause]
Singhal: Wow. That was fast.
Komoroske: Yeah.
As you can see, the browser on the right with Instant Pages
loads the page practically instantly.
There's no waiting for text or images to download.
They're just there.
At the bottom of the page, we have a small timer installed
that shows how long it takes for the page load event to fire.
Now, you can see, on the left,
that it takes about 3 1/2 seconds
on this very fast connection for that page to load,
whereas on the right with Instant Pages,
it was effectively instant.
Instant Pages is enabled by new pre-rendering technology
in Chrome being intelligently triggered by Google Web Search
when we're very confident that we know
what result you're gonna want.
But, of course, Instant Pages
doesn't just work with news sites.
As an example, I'm going back home to D.C. later this summer,
June 30th, and while I'm there,
I want to see if one of my favorite summertime activities,
the Folklife Festival, will be running.
So I'll do my search-- "dc folklife festival."
Oh, yes, it's the Smithsonian Folklife Festival.
That's the result that looks best to me.
Singhal: Done.
Komoroske: And again...
And again, with Instant Pages,
it loads effectively instantly.
In fact, it was so fast that if I looked pretty quickly,
I could have answered my query
that, yes, it is gonna be running June 30th when I'm home,
before the other browser had even loaded the page.
I've also heard that there's a great new museum
called the Spy Museum
that friends tell me I should visit when I'm home,
Singhal: That sounds exciting.
Komoroske: Yeah, so I'm gonna go ahead and research that
and see if it's worth visiting.
So what I have here is, um...
[chuckles] What I have here is results that show
on a map where the Spy Museum is.
Um, but, of course, what I want
is to learn more directly from the source,
so I'll click on their site.
Singhal: Done.
Komoroske: And, again, with Instant Pages,
it's effectively instant,
whereas it took about four seconds with Instant Pages.
In fact, it was so fast
that I could start exploring the site,
browsing, learning more about the museum
before the other browser loaded.
But, of course, it's D.C. It's the summertime.
It's gonna be hot.
I'm sure by the end of the day,
I'm really gonna want a cold treat.
Luckily, I know the best place.
It's called the Dairy Godmother.
It's one of my favorite places in my hometown.
They have amazing frozen custard,
and each day, they have a different flavor.
Singhal: I have to try it next time I'm in D.C.
Komoroske: Yeah, it's great.
So what I'm gonna do is search
for "dairy godmother flavor of the day"
to see which result I have to look forward to.
And so the "Flavor of the Day Forecast,"
that sounds about right.
Singhal: Done.
Komoroske: And you can see that in this case,
the page didn't load instantly.
It took a little while to load,
but even with that, with Instant Pages,
it was about four seconds faster.
And again, I could answer my question--
Swiss chocolate almond
is the flavor I have to look forward to--
before the other page had even loaded.
So that, in a nutshell, is Instant Pages.
Singhal: Great. Thank you, Alex.
This is amazing.
The time it saves us is just amazing.
In a live demo, you saw on an average,
we saved anywhere between three to five seconds
on every page load.
So what's happening here?
We still have to fetch the page and render the page,
and as the Web is advancing,
the Web pages are getting more and more complex.
We have measured that an average Web page these days
is about 700 kilobytes in size.
That's a lot of data.
And an average Web page takes 70 different sub-resources.
Sub-resources like JavaScript files,
images, ads, you name it, to actually build
the entire Web page on your screen.
With 700 kilobytes on an average
and 70 different resources
and numerous domain-name lookups,
no wonder it's slow to load Web pages,
and, yes, it's on an average two to five seconds.
It's on an average five seconds to load a Web page.
What we are doing here is the time it takes you
to scan the search results page--
like Alex was looking at the map of the Spy Museum
and so on and so forth,
the blue 15 seconds that you see up here--
we are taking those five seconds
in cases we are confident
that you are going to click on a result,
and we are basically folding them
into the blue 15 seconds,
thereby you don't experience the delay.
By the time you click, the result is already there.
how did we make Instant Pages possible?
Clearly, there are numerous technologies
that go into building something as great
as Instant Pages.
And it starts with having
Google Instant technology out there.
But this wouldn't be possible
without our Relevance technology,
because we have to predict that users are going to click
on a certain result.
We don't want to waste bandwidth,
and when our Relevance algorithms
that I have been part of building
for the last ten years
are confident that users are going to click
on a certain result,
using Chrome's pre-rendering technology,
we can bring to you Instant Pages.
We start out conservative,
but it's still very amazing.
We have measured time and time again
that Google Instant Search saves you
about two to five seconds when you are searching.
And what our measurements are showing
is that when Instant Pages are active,
you double up on those savings,
and you save an additional two to five seconds.
Now, imagine, for every query you do at Google,
you are now saving an extra four to ten seconds.
That's a lot of seconds saved,
given that we do over a billion searches a day.
All this time goes back to humanity
for them to go back to their quest for knowledge
and search some more.
Instant Pages will be available this week
in Chrome Beta,
and for those of you who are the adventurous type,
it's available today
in Chrome's developer version.
You can download the developer version
and enjoy fast Web today.
And, of course, we started out talking about mobile today.
And in the coming weeks,
we will be launching Instant Pages
for mobile as well,
where speed, as I said, is even more critical.
So let me just summarize this by saying,
today we talked about our momentum in mobile
and how much we are paying attention
to user experience in mobile
and breaking down all barriers
to the knowledge you seek on mobile devices
when you're out and about.
We also talked about how searching by voice
is sometimes the most natural way to query,
and we are bringing searching by voice to desktops.
Not only that, sometimes you don't even have
the right words to search, and it's a picture that it takes.
And today we are very happy to launch Search by Image
on Google Image Search.
And, of course, Instant Pages,
the next big leap for Google Instant,
breaking down big barriers in your quest for knowledge.
At the end of the day, our job is to get you
the information and knowledge you seek
in a blink of an eye.
And we want to make sure that your train of thought
that I talked about,
keeps running at ever faster speed
without any barriers and derailments,
and at Google Search, we are working hard
to make sure that your train of thought
keeps running faster and faster.
Thank you.