Uploaded by
Google on 08.09.2010
>> STRICKER: Good morning. Hi. Wow, you like that entertainment. Goodness. I'm Gabriel
Stricker, Director of Global Communications and Public Affairs at Google. I just have
a couple of quick housekeeping items to go through before we begin. First, is everyone
able to access the Wi-Fi in here? I'm actually--I'm getting some yes and no. So, the networks
are Search On 1 through Search On 5, so not the SFMOMA network. And the password is--is
isearchongoog. The letter "I" search on goog is the password. So, Search On 1 through Search
On 5. isearchongoog is the password. For those of you who are not here, who are viewing this
via our streaming, we're going to have a Q&A at the end of the presentation, and you guys
can submit questions for us via the email alias search2010@google.com. So that's search
2-0-1-0 at google.com. And then for those of you who are here, we're going to have some
demos at the--at the end of today's presentation as well, so just stick around and you guys
will be able to walk through a lot of what we're going to be announcing today. So with
that, I just wanted to give you a quick sort of update on why it is that we're here. So,
periodically we do these State of the Unions really for two reasons. First is we hear from
a lot of you that with the kind breakneck pace of innovation that we go through at Google,
it's nice for us to kind of let you catch your breath and give you a sense of where
we are and where we're going, and so we'll do some of that. And then secondly, we hear
that it's helpful from time to time to kind of take a look under the hood and get a sense
of the anatomy of what we do and why and how, and so you'll certainly hear a lot about that
as well. So, here we are at the SFMOMA, and, you know, what we do at Google, and in Search
in particular, is really one part art and one part science. We're here at the SFMOMA
today in recognition of the art part of that equation. The science, well, you're going
to hear from our Search rocket scientists in a second who will hold your hand through
the latest and greatest of what we're up to. But before we go there, I just wanted to flag
one thing, and some of you may have heard some of--of the remarks from Eric Schmidt,
Google's CEO, yesterday in Berlin. And there he said, "Never underestimate the importance
of fast." Today, we're going to be talking a lot about speed and the importance of fast.
And in recognition of that and in an attempt to kind of keep things moving on time, I'm
going to hand it over to our Vice President of Search Products and User Experience, Marissa
Mayer. Thank you. [pause] >> MAYER: Thank you, Gabriel. Good morning
and welcome, and thanks to all of you for coming to our Search event this morning. Obviously,
we have a really big announcement. But before we get to the announcement, I thought I would
offer some insight into just how far Search has already come in 2010, starting with an
exciting stat. In the past few months, we've crossed the "One Billion Users On Google's
Site Each Week" mark. We're really excited about this. We couldn't have done this without
our users, and we're profoundly grateful to them. And we want to really renew our commitment
to constantly be improving our quality and constantly improving our user experience.
This statistic makes Google Search one of the most used services. Not only internet
services, but services all over the world, and we're really excited and honored to have
achieved this milestone. And we couldn't have done it without our users. We also couldn't
have done it without a constant innovation and a constant sense of improvement. Most
of you know that we actually roll out hundreds of changes each year. In 2009, for example,
we rolled out more than 500 UI and ranking changes to Google Search. Already in 2010,
we've already rolled out more than 500 changes to our ranking and to our UI. And so this
stands to be the best year ever for Google Search. We've already had hundreds of improvements
in 2010, and I thought I would highlight just a few of them. First up, we have Caffeine,
our new indexing system, which makes our index 50% fresher. This ultimately means that our
users get much more up-to-date and timely results, and we can find the best information
regardless of where it is and when it was written. It really helps our index to be fast.
We also rolled out Real-Time at the tail end of 2009, but that hasn't stopped us in 2010
from continuing to build advances. We've been working on our triggering, so when Real-Time
shows up in our search results, it's incredibly relevant. We've also just rolled out a whole
slew of features, including timeline views and conversation views. We're really excited
about how to show real-time information, tweets and updates in the context of Search to really
help our users understand what's going on. We've also been making improvements to our
spell corrections. That may seem funny, because we've had spell correction for a number of
years now, but in the past year, we've actually made it much easier for our users to find
the spell corrections, because we now offer them as part of the auto complete. You now
get spell corrections before you even do your search. They show up there right underneath
the search box. We've also made great strides in how to handle proper nouns and really do
complicated spelling corrections. We've also rolled out enhancements to our Questions
& Answers. Two years ago in 2008, at an event much like this, we rolled out a product called
Google Squared, which was a technology demonstration that showed how we could intuit facts from
the data on the web. We've taken the technology behind Google Squared and launched it as part
of our core web search. So today, you can type small questions or facts into Google
Search and get answer right on the top of the Results page. Like the inventor of the
telephone. It just says Alexander Graham Bell. And it also works for real-time events as
well and current events. You can type things like "Iron Man 2 release date" and it'll give
the release date right on top of the Search Results page. And we've also rolled out Stars
in Search. We're constantly working to make Google Search more personalized. And through
our experiments with products like Search Wiki, what we found was that users really
wanted to be able to bookmark results. To say, "This URL I want at the top of this search
in the future and any related search, and I also want a way to be able to get back to
it easily." We now have Stars in Search rolled out on all users who are signed in for all
their searches and with a Star accompanying each search result, so it's always easy to
bookmark a result and be able to come back to it. And finally, in April we rolled out
our redesign. Our redesign introduced a left hand navigation panel that has all kinds of
dynamic tools that'll help our users slice and dice their--their results in entirely
new ways, ultimately helping them find what they're looking for all that much faster.
We also redid the formatting of the main Search Results in order to help our users scan and
find what they were looking for that much faster. And through all of these improvements,
we've also managed to have a little bit of fun. So, in January, we had our first television
commercial, a Super Bowl commercial, no less, featuring Search. Let's go ahead and roll
the video. >> [speaking French]
>> Hello. Bon jour. >> Tu es tres mignon. [pause]
>> Hello. >> [speaking French] [pause]
>> And the fun didn't stop there. We actually renamed ourselves Topeka for April Fool's
in honor of Topeka, Kansas, which named themselves Google for our Fiber-to-the-Home project.
And over the past few days, we've been having fun with our logos. So many of you probably
saw yesterday the bubble logo that was on our web site. How many people played with
it? It's pretty fun. It's almost like a digital Zen garden, which is very relaxing to see--see
the bubbles sort of explode all over the screen. But we really ran this logo because we want
Search to be fun, fast, and interactive. And this doodle was designed to tease our announcement
today. And then last night we lead with yet another doodle that helped illustrate and
give even more of a clue of what we're doing in that it lit up as your typed into the search
box. So this was a little bit of fun to get people in the mood for today's announcement.
And today's announcement does represent what we believe is a fundamental shift to Search.
How people think about Search. How people do Search. The speed and ease with which people
can find information. And as such, and to put today's announcement in perspective, I
thought I should offer some insight into the evolution of how people find information.
We're here today at the MOMA. The MOMA has an amazing permanent collection, and one of
the most important pieces in it is Woman with a Hat by Henri Matisse. Woman with a Hat was
painted in 1905. It actually is Matisse's wife, Amelie. When she posed for the painting,
she was actually dressed entirely in black. A black dress and a black hat. And as Matisse
painted this painting, a new art movement was born called Fauvism, where the focus is
on painterly technique and the use of bright colors even when bright colors don't appear
in the subject. It's a very important piece. It's a very special piece that the MOMA has
here on site. But if you wanted to be able to find out all of the information and trivia
that I just rattled off about--about Woman with a Hat, say, 75 years ago when the MOMA
first opened in 1935, you would've probably had to have spend half a day to a day in a
library. You would've physically had to have gone to the library. You would've had to find
references to the painting. You would've had to stitch together all of these different
facts to really understand what was happening with this painting. If you move forward to
1950, things get a little bit faster, mostly because of the prevalence of telephones. With
the telephone, you could basically skip the step of actually going to the library. You
could call a librarian. You'd still have to verify that you were talking about the right
painting and getting the right information, but you could get the information probably
in, say, about half an hour. Fast forward to 1995. Here personal computers come into
play. There were CD-ROM-based encyclopedias, and as long as your encyclopedia had an entry
for Woman with a Hat, you could find this information probably in the matter--in a matter
of minutes. But the problem with that is that it was static information. It didn't really
tell you is the piece on display right now? You know, is it on tour? Where is the piece?
This is my favorite piece in the San Francisco MOMA's permanent collection. It's Roy Lichtenstein's
Mirror #1. And I come here often to see it. And I will say that sometimes it's here, and
sometimes it's not. And it's really hard to tell when it's going to be here or not. But
when you look at Search and how it plays out today, this is where the power of the internet
really comes into play, because you can have real-time information. So, for example, if
you search for "Lichtenstein mirror SFMOMA," in the snippets you can actually see that
the mirror in on display in the second floor, and if it's on tour, it notes it on the web
page. This is something that's really wonderful, and it actually has taken search time from
that half a day to a day that we saw in 1935 to about half a minute. But let's go ahead
and take a look at where time is really spent in a search. We know that it takes a user
on average about nine seconds to enter their search into Google. After they hit the Search
button, it comes to us. There's some network time. Spends about 300 milliseconds on our
servers on average. Goes back over the network. And then after it renders on their page, they
spend about 15 seconds on average picking the right result for them. As you can imagine,
because we are so focused on speed--never underestimate fast--we have spent a lot of
time optimizing the 300 milliseconds when the search is on our servers. Making our algorithms
more efficient. Making our computers more efficient. Really trying to makes sure that
we have Search as fast and as efficient as absolutely possible while it's on our servers.
We've also spent some time thinking about network time. This is what our Fiber-to-the-Home
project is about; helping people understand how fast their connection could be, how fast
it should be and really pushing on this. Also we have tools like Google Chrome, which really
don't focus on network time but do focus on rendering time and other overhead in the browser
to really try and make people's experience with the internet as fast and clean as possible.
But despite all that emphasis on optimizing, you might say, "Well, a search takes 25 seconds,
and all of that focus is really on only one of the 25 seconds." Of course, there's 24
other seconds of typing and thinking that a user really spends. And you might say, "Well,
but is it possible to even optimize that?" And in truth, we actually have tried to optimize
it. For entering a query, we've added features to our search engine like auto complete, so
when you're typing a query, popular completions show up right there so you can do less typing
and enter your search faster. On selecting a result, we'll do things like the Google
redesign to help people scan the page and find information even faster because of the
overall layout. But you may realize that at some point we're really up against the fact
that entering a query, there's a physical speed for typing, and in terms of selecting
a result, there's a physical speed for thinking. And this is where we really say, you know,
can we optimize Search even more? Is it possible to make it even faster? And in the past few
months at Google, we've had a key insight. We think it is possible. We think it's possible
to have a system that provides a user with an easier way to enter a query with a lot
of feedback, and ultimately make Search very, very efficient. And we call that product Google
Instant. And that's what we're launching today. So, Google Instant actually gets queries and
gets your search results as you type and brings them to you and streams those results right
to your computer, so the search is entirely interactive the whole time you're typing.
And because it's so interactive, we really do need to show a demo. So I'm going to go
over to the demo computer, and I have Dan here, who's going to take a picture--we have
a picture-in-picture that'll show up here in a second, uh, so you can actually see what
I'm typing and what actually happens onscreen. So, let's go back to the Woman with a Hat.
Let's say we couldn't remember the specific name of this painting, and we just remembered
there's an important painting at the SFMOMA. It has a woman in it. You know, how does Google
Instant work? Well, the first thing you'll see if that Google Instant, it just is--it
is and looks like the main Search Results page--the main home page that you've always
been accustomed to. So if I go ahead and start typing my query and type S-F-M you'll see
I did not hit return, and the results for S-F-M, and actually for SFMOMA, because that's
the predicted--the predicted query, are already on the page. So with just three keystrokes,
we actually already have the result for the San Francisco MOMA. If I say, okay, that's
great. I typed S-F-M. The prediction of SFMOMA is right. I'll go ahead and tab, and then
I'll just type "woman" to try and find the name of the painting. And again, just six
keystrokes later, we already know that, in fact, the painting is called Femme au chapeau,
Woman with a Hat. So it's just that fast and that interactive. The results with each keystroke.
No hitting the Search button. No hitting Return. The results just stream to you based on the
predictions that are most likely, uh, given your--given what you've typed so far. So let's
go ahead and actually do a look up on the formal name of the painting, which is French.
Femme au chapeau. If I type "femme au", what you'll see here is that the most likely completion
is a Picasso painting, Woman with Crossed Arms. But you can also see five other searches
here right below the search box. And I can do those searches by simply using my up and
down arrow keys. So you'll note here that all I do is press down. Does that search.
"Women with chocolate." Uh, and then you can keep going through the search results and
they just update. So as you scroll up and down the overall recommendations and predictions,
you can see the results right there. No more typing. You can explore the whole space of
your query, Femme au chapeau, and all of the related queries. Let's go ahead and take a
lot at Fauvism. Fauvism is the painting movement that actually came into play with Woman with
a Hat. And if we type "Fauvism", what you'll see is--if I type F-A-U-V, it's already figured
out that I'm likely to type Fauvism. So with just four characters, I already have a definition
of Fauvism, pictures and paintings that show examples of it, mm-hmm. And you'll note that
a lot of people think Google Instant is search as you type, but it's actually search before
you type. We didn't do a search for F-A-U-V. We did a search for Fauvism, because that
was the most likely completion. So we're actually predicting what query you're likely to do
and giving you results for that. We think this ultimately results in a much higher quality
experience, but there's even a psychic element of it in that we can actually predict what
you're likely to type and bring you those results in real time. And the other cool thing
is using the predictions, you can build--you can build your query much more easily. So
let's go back to the Lichtenstein mirror. If I type Lichtenstein. L-I-C-H-T. Google's
already figured out that I'm likely to type Lichtenstein. Doesn't know is it the artist
or the country, but I can hit Tab, then type M-I for "mirror", and again, it's already
figured out that I'm likely to type mirror. Hit Tab. Hit S-F. And there's the page on
the Lichtenstein mirror here at the MOMA. We're really excited about what Google Instant
means in terms of Search. It means much faster Search, easier Search, an easier way to explore
all of the related queries around the query that you're doing, and really providing results
in real time before you've even had the--had the opportunity to type your query. We're
really ultimately excited about this. Let's go ahead and go back to slide. This is funny.
This is something that I don't know if many of you have seen, but in 2000, we actually
thought that this idea of being able to search before someone typed was so far out it was
actually our April Fool's Day joke. I was called MentalPlex, and it says, be sure to
remove your hat and glasses. Look into the circle. Think about what you want to search
and present it. And so we thought this was sort of a fun--fun bit of Google history,
and that obviously dovetails with today, because just ten years later, what we're seeing is
it's actually possible for us to greatly ease the entering of a query, greatly ease the
scanning of results, and really speed up the Google Search experience. Google Instant is
going to be--is going to be available later today on a variety of browsers. It'll be available
on Google Chrome, on Firefox, on Safari and on IE8. Mm-hmm. And this will all be rolling
out over the course of today. So starting today, if you're using one of those browsers,
you will be getting Google Instant as part of your core web search experience. We'll
be rolling this out as part of our home page and on google.com in the U.S. But over the
next week, we'll also be rolling this out internationally. In the U.K., in France, in
Italy, in Germany, in Spain, and in Russia for all users who are signed in. So if you're
signed in and you're using one of those country sites, you'll also be able to experience Google
Instant within the week. We're really excited about what Google Instant could mean, in terms
of the evolution of Search and the future of Search and what it means in terms of how
it will change people ultimately searching. In fact, we estimate that Google Instant will
help our users save two to five seconds per query. That may not seem like a lot, but when
you add that to the billion searches and more a day that come--a week that come to Google,
you actually get significant savings. In fact, you actually will be able to save 11 hours
for every passing second. We're really--again, we couldn't be more pleased with how this
product has developed, and I'm really excited to introduce to you the team that built Google
Instant, so they can talk to you about some of the technical challenges and some of the
product challenges that came about in building Google Instant and what it ultimately means.
With that, I'd like to welcome Othar Hansson and Johanna Wright to the stage. [pause]
>> WRIGHT: Imagine if Google were even faster than it was before, if it knew and understood
just what you were trying to say before you finished typing, if it could give you feedback
as you went along to get you better results. Well, that's what we're here to talk about
today. We've been using Google Instant for months, and we want to talk about how it's
gonna revolutionize the way that we interact--all of us interact with a search engine. This
is not a screen-shot story. You truly have to experience Google Instant yourself to understand
its power. We're gonna do the best we can with our demos, but please check out the demo
stations on your way out, and today, please use Google as Instant rolls out to our servers.
I'm Johanna Wright. I'm the Director of Product Management on Search and working on Google
Instant. >> HANSSON: And I'm Othar, the engineering
lead on Google Instant. Today we're making Google Search faster and easier than ever
before, and Johanna and I are excited to show you the three main features that you'll start
noticing today. >> WRIGHT: And you can think of these three
features like gears in a system. One small change in one gear can have vast leverage
over the full system, making your search much more efficient. In Google Instant, the first
gear is Instant Results. The second is our set of Predictions. And the third is what
we call "Scroll to Search." Let's jump in, and let's start by taking a look at Instant
Results. Immediately when you start typing, results display on the screen instantly. There's
no need to finish typing. There's no need to hit Enter. Um, I'm actually a little sad
that the summer is over. It's hard to believe that Labor Day Weekend was just last weekend,
and, um, now it's fall. And, Othar, we're here in San Francisco. Why don't you use Google
to check out the weather? >> HANSSON: Sure. This is a great search for
Google Instant, because with exactly one key press, on W, I can get a weather forecast
for San Francisco. Whoo! 60 degrees. So, uh, my career as a hand model means we can't show
my hand, but in slow motion, I'll use one key press. Weather. I did not press the Search
button or press Enter. My fingers never left my hands.
>> WRIGHT: So even in slow motion, Othar was fast. And you can tell that he found out the
weather was 60 degrees out-- that's right, 60 degrees out by typing in just one keystroke.
So now that we've talked about the first gear, Instant Results, let's move on to the second--
Predictions. This summer, I, and maybe many of you in this room, read the book "The Girl
with the Dragon Tattoo." I don't think Othar read it. He's making fun of me there.
>> HANSSON: I was a little busy. Sorry. >> WRIGHT: Well, what if we wanted to learn
more about this book? >> HANSSON: Sure. Book titles and movie titles
and song lyrics--they're all great searches for Google, because people half remember these
things, get to a search box, and they do their best, and we do our best with our ranking
and other tools to give them the results they need. Google Instant just gives you even more
feedback. So to prove my point, I've already forgotten the title, but I have faith in Google
Instant, so I'll just start typing "the girl," and there you go. This look like a book. I
scan down to the results. It's a book that's been made into a movie. Is this the one you
had in mind? >> WRIGHT: It sure is.
>> HANSSON: That's great. So what you'll notice is the black text says "the G-I." Those are
the only letters I typed. They're in black text. The rest of the search is in gray text.
We call this the Prediction. And the way--people intuitively focus on the Search box while
they're typing. So the gray text is a great cue as to what Google thinks you're searching
for. And in this case, this looked enough like a book title that it was worth my while
to scan down to the results and see if it was, in fact, the book title you had in mind.
>> WRIGHT: The real power is in tying that gray prediction to the Instant Results. Google
becomes Search before you type, and it's as if it knew exactly what Othar was trying to
say. >> HANSSON: Exactly. It's not quite psychic,
but it is very clever. >> WRIGHT: Our third feature today is what
we call "Scroll to Search." And this summer when I was planning my summer vacation, Othar
told me about a trip he and his wife had taken to Yosemite.
>> HANSSON: Oh, yeah. >> WRIGHT: And this search, Yosemite, actually
shows off Scroll to Search quite nicely. >> HANSSON: And it has great sentimental value--first
trip my wife and I took. So I'll just start typing Y-O-S-E, and there I get Yosemite.
I just learned from Marissa that I can use Tab to complete. And then if I use the arrow
keys on my keyboard, I can get to lodging. I can get Map.
>> WRIGHT: Wow. Those showed up right away. >> HANSSON: I can get camping information.
And you'll notice that the page is the full results page that you're used to, right? All
the results and special features, all the ads, et cetera, all being served to you and
even things like weather. It's warmer but raining in Yosemite. Sorry.
>> WRIGHT: Well, thank God we're in San Francisco. So what you saw here is Othar used the down-arrow
key to scroll through each of the predictions, and as he did that, the results showed up
instantly. This is a great way to explore, um, aspects very quickly of things that you're
searching for, and we call this third feature "Scroll to Search." Now that we've talked
about the three components of Google Instant, our next topic is feedback. The real power
is when these three components-- Instant Results, Predictions, and Scroll to Search--come together.
This turns Google into a system that gives you feedback as you go along. So one thing
that happens to me today is when I'm searching, I'll type something in, and I'll hit Enter,
and I'll see if I got the right results. And sometimes I do, sometimes I don't. And if
I didn't, I'll just do it again, and I'll do it again until, finally, I get the right
answer. >> HANSSON: This happens to everyone. It's
happened to me. I've recently been planning a trip back home to New York, and I want to
take my mom and her grandkids to a musical, get them some culture, and me too. Johanna,
you seem cultured. Do you have any musicals to recommend on Broadway?
>> WRIGHT: I actually hear that they made a musical out of my favorite movie, um, "The
Addams Family." >> HANSSON: Okay, that sounded promising until
the title, but... Uh... So let's just search for "Addams Family" and see what we get. We
get an ad for the musical. That's great. The results seem to be mostly about the TV show
and the movies, so I'll just tab and type space M, and now I get the musical, and now
I'm feeling like I'm on a roll, so I'll just see if I can get tickets by adding space T.
Yes, I can. I can buy tickets for today's performance, if only we could get there in
time. >> WRIGHT: So let's decompose what happened
here. Othar typed in "Addams Family," and then the Instant Results gave him feedback
that he needed to type more, because he was looking for the musical. He wasn't looking
for the movie. So he went ahead and added the word "musical" to his search. Then since
he again was thinking he wanted to buy tickets, in a split second and in his mind's eye, he
said, "No, I need to change my search. I'll add the word 'tickets.'" So now what had previously
been three searches became one search, and this sped Othar up tremendously. [pause] So
that was our discussion of feedback, and we've done a number of demos. And at this point
in our day, you guys might be thinking that the Search button is feeling a bit lonely.
What the heck is the Search button there for, anyway, at this point? Well, why don't we,
um--Well, what happens, actually, if we press the Search button? Why don't we go ahead and
show you? Well, what we do is we give you the same great results we always have. I'll
give it away. But why don't we go ahead and do a search for New York?
>> HANSSON: Sure. So you're right. The Search button--And the Enter key is the cleanest
key on my keyboard now. But if I search for New York by typing "NY," the best prediction
is actually that you want the "New York Times." I'm sure someone in the audience is happy
to hear that. That's what the odds are. But if I just want NY and I type "NY" and then
either press Enter or press the Search button--once I found my mouse--then we do what we've always
done for 12 years. If you type NY and press the Search button, we'll search for NY. That's
the natural thing to do. >> WRIGHT: Yeah, so Google works seamlessly
as it always has. All it does is it speeds you up.
>> HANSSON: Exactly. So whenever you improve a great product, you want to make sure that
everyone who's already using the product gets a benefit out of the new version immediately.
Everything they know how to do applies to the new version. I had this experience recently
with bicycles. I had this great bike that I ride to work as often as I can, which is
not often enough. And I finally got a raise--hint, hint--and bought myself a better bike, in
the same line, just a better version. I went to the bike store, and they have this little
hill you can climb behind the bike store. And so I hopped on this bike. I knew how to
use it immediately--the great thing about bikes--and I climbed up the hill. And the
only thing that was different is I got to the top of the hill in about half the time,
because the bike was faster and lighter and more responsive. And that's what we're shooting
for with Google Instant. So on that day, I got myself a better bike, and today we're
all getting a better Google. >> WRIGHT: So if you know how to use Google
Search right now, you know how to use Google Instant. It's just like riding a bicycle.
Thanks for laughing. Okay, so let's do one last demo to bring it all together. This last
demo's gonna come from another story between Othar and I. So I want to get a dog. Othar
has a dog, so I asked Othar to tell me about his dog.
>> HANSSON: Well, in fact, during this project, we scaled up everything at Google, so I went
from one dog to two. But we do have two tiny dogs. And they're tiny, but they have a big
name--Cavalier King Charles Spaniel. And something like that is a great search example. I'll
meet people in the park, and the little kids will say, "Daddy, Daddy, I want a dog just
like that." "I thought you want a pony." "No, I want a dog just like that." And the dad
will ask me what's the name, and I'll tell them Cavalier King Charles Spaniel. And I
know that an hour later they'll be at home, and they'll have no hope of remembering the
exact name. So half-remembered queries are great examples for us, 'cause let's say they
go home, and they start typing "King Charles." There you go.
>> WRIGHT: Aw. >> HANSSON: So with the Instant Results, you
immediately recognize, well, this is a dog, right? The name sounds vaguely like what I'm
looking for. And if I scan down or keep, uh, searching, I see, oh, it's the Cavalier King
Charles Spaniel. So that's, in fact, what I was looking for, even though I started searching
in, you know, not an exactly correct way. The other great thing about a search like
this is ambiguity, right? So dog people call these dogs "Cavaliers." Now, there are other
things in the world called Cavaliers, right? Now, we could have a little game here where
I ask you to bet which one is the most common, but basically, people can't know this, right?
So if I start typing "cavaliers," it turns out that the basketball team, despite LeBron
James, still outranks my dog, and so I realize that I have to keep typing. And if I just
type "cavalier k," the feedback from the Instant Results and the Prediction show me that, okay,
now I'm on the dog. So I see the results for the dog, but, Johanna, back to your original
point that you need a dog. I think that you really need to get puppies to match the size
of your kids. And if you can, I think you should get the dogs that are the color that
mine are--tricolored dogs. >> WRIGHT: Those are pretty cute.
>> HANSSON: So if you get two and I bring my two to work, we can have playdates all
the time, never get any work done. That'll be awesome.
>> WRIGHT: So just with all the demos, what happened here is Othar started typing in "cavaliers,"
and then he got the feedback from the Instant Results, telling him that he had to go ahead
and type a little more and add the word "king." This gave him the right results. He used the
Predictions to understand that Google had guessed correctly, and then he went ahead
and he used Scroll to Search to down-arrow and see the cute pictures of puppies. So those
are, um--those conclude our demos of what we're launching today, but we also have a
sneak preview of something that will be coming out this fall. Do you want to go ahead and
show that off? >> HANSSON: Yes. So at Google, we normally
don't speak in the future tense, but we are working very hard to get this experience to
you on Mobile. When you're on your phone, first of all, you're on the go. You're probably
in a hurry, so Google Instant can help you more. But more importantly, typing on a phone,
no matter what kind of phone you have, is just slower. So the feedback from Google Instant
is even more useful. So let's start searching for, uh, one of these art queries that Marissa
was doing. We'll search for "Roy Lichtenstein." Now I'm halfway through, and I don't know
how to spell Lichtenstein, but I assume it starts with an L, and the feedback shows me
that I'm getting Roy Lichtenstein, and the Image results show me that I've got the artist.
So now I can scroll down to check out more information about Roy Lichtenstein. So you
can check this out later this fall, and you can also check it out later today at the demo
stations outside. >> WRIGHT: Our group at Google is focused
on making Search faster and easier for everybody to use. As part of our process, we test all
of our features on a number of users and gain feedback and improve the products as we go
along. We were lucky enough today that a handful of our testers have been willing to share
their thoughts with you all and have given us some testimonials. So let's go ahead and
take a look at what they had to say. >> As I was typing, things were already showing
up. >> Oh, wow.
>> I don't even have to scroll. It's right there.
>> It's guessing for you. >> Oh, look at that. I like that.
>> It's not an extra clunky add-on. It's just completely integrated in the Search experience.
>> The Autocomplete kind of shows up in gray. >> I didn't have to really press Enter.
>> As you're typing it, it's coming up with the results below, and you can sort of backtrack
three letters, start retyping, and see the results instantly.
>> Seeing the results as I was typing really helped, because I can scan with my eyes exactly
what I want to find. >> It helps you kind of pinpoint what to type
in to find the answer. >> I felt that I was on track to get what
I wanted to get quickly. >> It took a lot less time to find the answers
that I was looking for. It was pretty fabulous. I liked it.
>> Well, I always try new things, and I appreciate what you folks are doing, as far as making
it easy for us old guys to work on the computer. >> HANSSON: That's great. It's always great
to come to work and listen to our users and hear them say things like, "Oh, wow, that's
fabulous. This is gonna help me a lot. And when is this launching?" Well, the answer
is today. Under the hood, obviously, there's a huge amount of engineering work that has
gone into this project, and so Ben Gomes is gonna join me now to talk about the engineering
challenges. [pause] >> GOMES: So when we first began to demo this
product and this idea around the company, we got several responses. Some people said
that we just could not do this. It was just gonna be too expensive. Other people said
we must not do this, because it was gonna be too complex a change. But then we showed
it to users, and then we showed it to some important people at the company, and they
said we must do it, and we must do it now. So we all came together to figure out how
we could make this vision a reality. And the challenges fall into three areas. The first
challenge is the user interface. How do we create a user interface that's really simple
to use and that's familiar to people? The second challenge is how do we make it work
on your computer, on your browser, with your connection. And the third challenge, which
is probably the biggest challenge of all, is how do we possibly do this without melting
down our data centers. So let me start with the first challenge. How do we come up with
the user interface? We started out with this rough idea. We want to give you feedback as
you type your query. It seemed that that would help you formulate a better query. In fact,
somebody at Google had done this demo way back in the mists of time, and engineer named
Amit Patel. And what it did was very literally search as you type. So I'm searching for bike
helmets. I type in "B-I-K-E H." Now what do we search for--"bike h." Now what are the
chances that you were searching for H? It's almost zero. So what happens is that we end
up doing a ton of work that's entirely useless with these useless queries. You end up seeing
results that are entirely useless to you. So it's a lose-lose proposition. And with
no offense to Amit, this was not something that could work. We knew we needed something
different. We needed to work with those partial queries.
>> HANSSON: Exactly. So the best way to make this useful was to take your partial query
and complete it to a full search. Thankfully, in the meantime, between ancient history and
today, we launched Autocomplete. So Autocomplete filled that role of completing partial searches.
Now we just have to build a user interface around this. And as you've already seen in
demos, we hit upon this idea pretty early on. We built prototype after prototype and
showed them around the company--this idea of showing gray test to indicate the prediction.
So you're typing, Google is guessing what you're searching for, and we show that to
you as gray text. So that UI seemed to work really well. The next question was, would
it be too distracting to show you results as you're typing? And in fact, it was too
distracting in early prototypes. Part of it was timing. So we got the timing right and
the responsiveness right so that this felt a lot more seamless. Once we had a prototype
that the team was happy with, we went through the process called dogfooding. So we eat our
own dog food at Google. I'm always glad that I don't actually work in the dog-food industry.
And dogfooding was very positive. We had--you know, we had to bring the demos down occasionally
for maintenance, and we get people complaining that they had to use regular Google today.
It just seemed broken that they didn't get Search results as they typed. When is it gonna
come back up? That was very gratifying to see. And then we eventually did live experiments
with millions of users. And it was very gratifying to see, you know, just like in that video
of our trusted testers, that people learn how to use this product very quickly, right?
All the things they know about how to search still work, and we saw that in our live experiments.
I wanted to point out, uh...by skipping some slides, the, um... One of the techniques we
use in the lab--we use an eye-tracking experiment. We'll invite people into Google, and we'll
sit them in front of a computer, and we'll show them that around the monitor are sensors
that detect where they're looking on the screen, and we'll show them that there's a red dot
that indicates where they're looking, and we'll have them fix their glasses and clean
their glasses and whatnot so that this works. And then we'll give them some tasks or let
them do some of their own searches, and we'll watch where they're looking on the screen
and how well they interact with the user interface. So I'm gonna show a video in a few minutes,
but the things to notice are, as people type their search, they are focused on the search
box. They're focused on the gray text. Sometimes they'll dip down to the Autocomplete. And
then once the gray text is something that they're interested in, they look at the results.
In fact, in this video, you'll see that the person has done a couple searches before this,
and by the time the gray text is correct, they look at the first result a split second
before it arrives, right? So they've been trained to know that it's coming. So let's
take a look at the video. So they're searching for parasailing in San Diego. We initially
predict San Francisco. The odds are just better of that. But once they get to San Diego, you
see that they actually look at the first result before it arrives, right? So Search is fast.
Google Instant is even faster, so let's take another look at that just so you can see what's
going on. So parasailing, San Diego is the Search. Once the gray text is what they want,
they're looking at the first result, right? And then they're gonna click on it, obviously.
So we saw this time and time again with, you know, over a hundred users that we tested
in our usability lab. Obviously, some people look at their keys and look up eventually.
That's another important class of people. But, uh, for people that actually look at
the screen, they are focused on the Search box. So we knew we had a user interface that
worked and, in retrospect, it seemed like such a simple design. I was reminded of how
my dad used to tease me when I told him that I was gonna go work at Google and I was gonna
go work on user interfaces. And he said, well, that's got to be the easier job in the world.
There's a box that you type into and a button that you press. So first of all, which half
are you working on? And then he said, well, okay, so your job is basically to make it
easier to to type into the box and then easier to press the button. And in retrospect, he
was both clever and psychic, because with Google Instant, we made it easier to type
in the box by giving you feedback with the gray text and the results. And we made it
easy to press the Search button by not making you have to press it as often. I've only pressed
it once today, and only at Johanna's command. >> GOMES: So the key insight there was the
predictive text that showed you the kind of--the query you were going to get. So the next problem
is how do we take this experience and make it work for you? Search has so far been a
simple HTML page. We are turning it into an AJAX application with all the challenges that
entails on different browsers and on different computers. Google has done this before. Google
has done this with maps. Google has done this with Gmail. But now we are doing it with Search.
And this is not a simple problem. You will notice the fact, with these demos, in fact,
that the--the chrome of the page, the part that stays constant, never refreshes--only
the part, the results that are coming in refresh. And that's a sign that we have turned this
into an AJAX application. But there are many challenges in doing this.
>> HANSSON: Exactly. So we'll just describe the basic HTTP request flow. I'm sure people
will dissect this anyway, so might as well explain how it works. I've always wanted to
use this analogy, so let's pretend I'm the client and my boss is the server. So I'm the
client. I'm typing in my browser, and I've typed half a query. We send an Autocomplete
Request to the server... >> GOMES: And I send you the Search Predictions.
>> HANSSON: Right. The Autocomplete Predictions come back. This is one of the fastest things
our servers do. So they're on the browser being displayed very quickly. Meanwhile...
>> GOMES: In the backend, I take the first Search Prediction and I send it off to the
servers to get Search Results and send it back to the client.
>> HANSSON: Right. So now I can show the instant results. We saw in the little video that the
person anticipated that they'd be there. Um, we showed them as fast as we can to the user.
In the meantime, lots of things could have happened in the browser, right? We might have
clicked on a result from a previous, you know, set of results. I might have typed in some
unpredictable direction, etcetera. We send other Autocomplete Requests to the servers,
and the servers figure it out amongst themselves whether the search they just finished is still
relevant to the user. >> GOMES: And cancel some searches.
>> HANSSON: Yeah. So that's one optimization we did. We also had to do all kinds of JavaScript
optimizations in different browsers. This experience doesn't work if the browser spends
more time rendering results than the, you know, time between results arriving. So we
had to optimize our JavaScript and work around browser performance issues. The other thing
we had to do in our JavaScript was actually detect when the experience was broken. You
might go from one cafe to another, and the internet speed, you know, drops dramatically.
So we wanna detect that and proactively opt you out of the experience and give you a little
error message, giving you a hint as to what's happened. When you go back to the other cafe,
it should turn back on. Um...so that's a rough summary of the application.
>> GOMES: So now we come to the third part and hardest and most interesting part of all
of this: how do we possibly do this without melting down our data centers? So think about
the problem. We want to give you search results with every letter. An average query's about
20 letters long. So if we give you search results with every character, that's 20 queries
for every search you do. So we got this--I got this response every time we demoed this
initially-- Ben, are you crazy? There's no plan in the world where we want--we're already
serving billions of queries a day. There's no plan in the world for us to be serving
10 or 20 times that amount of traffic, like unless people from other planets suddenly
wanted to start using Google. Why would we plan for like 20 times--20--a factor of 20
growth in Search traffic? So this was one of the most interesting challenges in getting
this down to something practical that we could actually do.
>> HANSSON: Thankfully, Google has thousands of engineers who love challenges like this,
many of them in our Search infrastructure group. And they came to us at first skeptical.
They would say things like, "You're never gonna launch this." But I realized last night
while running a load test that if you did this and that, you could save 27%, and here's
a simulator that, you know, shows you how to do it. And then someone else would say,
I realize that if you just made this optimization here, you could probably save another 12%.
And if enough of these skeptical, helpful people come together, you get enough optimizations
to get to the finish line. So let's just look at a couple of these optimizations. So the
first was realizing that some searches are more important than others, where more confidence
given, you know, like a single letter that--in some predictions than in others. So we wanna
make sure we serve those results to you as fast as possible. And that, you know, by prioritizing
searches differently, we can reduce the average cost per search and the number of searches
we do. The next optimization is this thing we alluded to before in the client/server
aspect--that if the servers can keep track of what the browser either already has or
what's already in progress for them on some other server, then we don't have to do needless
work. We can, uh, reduce the number of searches that way. A final and pretty sophisticated
improvement we made was in our caching system. So caching is always a tension for Google
because we want our results to be as fresh as possible as we crawl and index the web.
But we also want them to be cached so we can serve them to users as quickly as possible.
So this was one of, you know, the most sophisticated things we built. There was already a team
working on it, and we accelerated that effort tremendously. So we have, you know, very sophisticated
caching in all our backends and many more optimizations to get us to the finish line
and be able to launch today. >> GOMES: So...I'd like to take a moment to
actually thank and acknowledge the team that worked on this. It was a really phenomenal
team that came across--together from across the company. A few of them are here, and I'd
like them to stand up, actually. But many of them are in our offices across the country
and our office in Haifa in Israel that does the Autocomplete. And it started with one
engineer a year ago and doubled every six weeks, approximately, until the team you see
today. I'd like to give them a round of applause, actually. [pause] We work on Search for many
reasons. We work on Search because it's an absolutely fascinating problem that goes to
the heart of our human curiosity of our need for information. We also work on Search because
we're engineers, and the sheer scale of the problem makes it fascinating. There's billions
of queries, there's tens and hundreds of billions of documents. We need to get you from your
query to the right set of documents in a fraction of a second, and that's just fun. But most
importantly, we work on Search because we believe it matters to people. It matters to
us. Getting people to the information they need quickly and efficiently matters in their
lives. It helps them make better decisions. It helps them enrich their lives...from the
patient looking to understand his treatment to the teacher looking for materials to inspire
her students. To the journalist looking to investigate a new story--it matters. So we're
constantly working to improve Search. This last year alone, we launched over 500 changes
to Search. Most of them just quietly improved your search experience...from a much bigger
and fresher index to better understanding of documents with synonyms and so on, to real-time
results and hundreds of changes like that. But occasionally, we launch a change that
changes the way in which you interact with a search engine, a change like spelling or
Autocomplete. Today we are launching the biggest change of that sort in Google Instant. Having
used it for a few months, from early prototypes to the version you see today, I feel sure
that this is the future direction of Search. Not only does it make Search faster--that's
the obvious benefit--but it makes Search much more fun, fluid, and interactive. It enables
you to formulate better queries with the results you see as you type, and in doing so, you
learn more, and it enriches your life. So Search is always on this journey. It's a journey
towards the ideal Search Engine that knows all the information in the world, it knows
exactly what you want and can take you from what you want to the best information in the
world. Google Instant was a change that touched virtually every part of our serving systems.
It was a massive change. But I believe that Google Instant is one of those changes that
will seem so obvious in retrospect that you'll wonder how Search was ever any other way.
[pause] >> MAYER: Like Ben and Othar, I'm very optimistic
about the future of Search. And I'm also very optimistic about our search at the speed of
thought--Google Instant. When we look at the future of Search, it really moves forward
along three key axes. One is interactions...how do people search? And when you look at how
Google Instant is changing that, how Google Goggles has changed that, how Google Voice
Search has changed that, it ultimately really transforms the landscape of Search in terms
of how people search, how much they search, and what types of things they search for.
But this is a big component of what the future of Search looks like. Comprehensiveness is
the next axis of the future, being able to bring more and more offline content online,
being able to find answers to queries now that we couldn't have found just a year ago
or just a month ago. By having more and more results and more and more content in our index,
we're able to really achieve the type of comprehensiveness our users have come to expect. And the final
component is understanding. Both understanding our users, like in the case of Google Instant,
where we're understanding their intent, or understanding the web at large, like with
Google Squared and our Question and Answer service. We really think that by understanding
the web and our users better, that's how we'll achieve quantum leaps forward in Search. And
obviously, we feel that Google Instant is such a quantum leap forward in Search. It
has all three components of these different pieces of the future of Search all wrapped
into one-- a different way of interacting, results with every letter, more comprehensiveness.
You can actually explore all of the related queries easily with just a keystroke. And
also understanding, really understanding our users' intent, being able to predict our user's
intent. We really are very proud of what Google--what Google Instant represents in terms of a step
forward into the future of search. It also represents a large time savings. So you may
wonder what these counters were on the walls as we've been presenting. But these counters
are counting the hours saved by Google Instant. So during this presentation, Google Instant
will have saved 36,000 hours of our users' time. And over the course of a year, we anticipate
it will save somewhere around the order of 350 million hours of our users' time. We obviously
are very inspired by Google Instant, as was our creative lab. So in closing, they created
a video. We gave them the Dogfood of Google Instant and let their imaginations run wild,
and they were inspired by Bob Dylan. In 1965, Bob Dylan gave up the acoustic guitar and
took up the electric guitar, and his first song was "Subterranean Homesick Blues." It
was also his first song to break the top 40. They felt that Google Instant had a lot of
the same properties, taking something tried and true and making it even better. We obviously
hope that it's as big a hit, and so with that, let's go ahead and look and see how Bob Dylan
inspired our creative lab. [pause] Johnny's in the basement, mixing up the medicine, I'm
on the pavement, thinking about the government, the man in the trench coat, badge out, laid
off, says he's got a bad cough, wants to get it paid off, look out, kid, it's something
you did, God knows when, but you're doing it again, you better duck down an alleyway,
lookin' for a new friend, the man in a coon-skin cap in the big pen wants 11 dollar bills,
but you only got ten. [pause] >> STRICKER: Thanks so much. So as promised,
we're gonna have a time here for some Q&A. There are gonna be some runners here on the
sides. You can see them here. They have mikes. Just give us one--one quick second, and we'll
get set up. But if you raise your hand, we will try to stare through the lights' glare
to be able to see you. Then I also just wanted to give a quick reminder, uh, that for those
of you who are viewing the webcast, you can send an email of your question to us at search2010.
That's search2-0-1-0@google.com. So, um, why don't we get started right here? Um...are
there really enough--really enough seats for us? Okay. That's amazing. Fantastic. Go right
ahead. >> SCOBLE: I'm Robert Scoble. When will this
be in the browsers in Google Chrome or whatnot? >> MAYER: Uh, this is something that we're
working on, so we anticipate sometime over the next few months this will be something
that can be activated from the Search boxes. [pause]
>> STRICKER: I see Greg here. >> GREG: Impact on AdWords-- lot of impressions
potentially going on by. Are those counted as impressions, uh, in terms of ad quality
and click-throughs and all of that? A lot of Search marketers want to know. Inquiring
Search marketers want to know. >> WRIGHT: So there's no change in how we
serve or rank our ads and, um...in terms of counting impressions within AdWords and within
other people's mechanisms, what we do is we do our best to understand...well, I guess
the way to think of this is that Google Instant changes the way that people search. And so
we're doing the best to calculate the way, um, what a search is. So the other addition
to impressions is we've added a three-second pause as an impression.
>> HANSSON: But...the other main cue that something's an impression is someone clicked
on a result or clicked on an ad or clicked anywhere else on the page.
>> WRIGHT: Or hits the Search button, as always. >> GOMES: And that just stays the same as
before. >> MAYER: And it's important to note that
because we are operating on a CPC basis, obviously, you pay for the clicks and the clicks are
largely preserved, despite the change in overall interaction. So basically, while your click-through
rate may change, the overall number of clicks accrued to your site is likely to remain very
constant. >> So, historically, you've said about 20%
of queries are unique; you've seen them for the first time. Um, how is this going to affect
that? I assume that it will channel them into the more obvious and common queries instead.
>> GOMES: I think that--that'll await expert, actual data, because I think some of this
will be learning effects and product--the product changes the way you search. So I think
we'll have better answers than that in a few weeks time as people get used to the product.
>> HANSSON: The other thing is if you make search easier and faster, people will search
more and they'll search more, uh, honestly, they'll give us more difficult problems to
solve. >> STRICKER: Yes? Can you just wait one second
for the mic? >> SINGLE: Ryan Single from "Wired." Could
you give me a sense of how this is going to fit in with, um, users', uh, search history,
and what of the results are tied to people's web history and search history, and has that
changed at all from the current results? >> MAYER: Uh, so this operates largely as
you'd expect. So in web history, queries where you've done a click will be shown, as they
always have been. We also will show you the impressions where you pause for more than
three seconds, since we're using that as part of our impression count. And Personalized
Search continues to operate the same way based on the searches you've done and the results
you've clicked on. The inputs into that mechanism remain unchanged.
>> SLUTSKY: Hi, Irina from "Ad Age." Ahem, it looks like you have a blacklist of words.
My last name is actually one of them. It's Slutsky. And there is the-- Can you talk about
that? >> WRIGHT: So we, um--
>> BRIN: Have you been Googling yourself while we were here?
>> AU: Everyone around me is actually Googling me right now.
>> STRICKER: Oh, I see. >> AU: And, as we know, there's my great uncle,
Eugene Slutsky. The Slutsky equation doesn't come up either.
>> WRIGHT: So we care a lot about child safety and these kinds of issues. And so we had to
think a lot about, um, Autocomplete and how that works as you're getting the result as
you go along. So we applied the same policies that we actually always have, whereby we,
um, we filter for both violence, hate, and pornography. And as a result, if you're typing--and
apologies on your name--but as a result, if you're typing something that, um, may not
be appropriate for, um, certain people to see as you go along, we won't show the results
as you go along until you go ahead and press enter.
>> HANSSON: And obviously this isn't perfect, but something we're--we're-- [indistinct questions]
>> WRIGHT: We'll--we'll look into it afterwards. Thanks a lot for pointing it out.
>> STRICKER: We have, uh, some demos, actually, for those of you who are here afterwards,
and we're gonna actually troubleshoot for you. Just give us one second. Go right ahead.
>> CHAPMAN: Oh, just real quick. Glenn Chapman with "Agence France Press." To play off on
a little bit more of an expansion of what was asked earlier, how much personal data
is needed to make this faster? Like when you pressed "W" for weather, it was San Francisco,
but if you're in Chicago and you press "W," you're gonna get Chicago. So how much does
the system need to know about the user to make this fast? And then just really nuts
and bolts-ish, you said you reached optimization. But how much more pressure does this put on
your data centers, I mean... Basically, do you have to build new ones? What's going on?
>> MAYER: Uh, well, I'll take on the first one. In terms of personal information that's
required, it's unchanged. So today on Google Search, for example, we already factor in
your IP address location, so just typing "weather" ultimately would bring you the results for
weather in San Francisco or weather in Chicago, if you were there. Uh, and it's just because,
when you type weather, we're predict--when you type "W," we're predicting it's weather.
We're running that query as if it was weather from a San Francisco IP address. So there's
no personal information exchanged in there beyond what already exists on Google today.
>> HANSSON: The feature's unchanged, just cooler, so to speak.
>> STRICKER: Just give us one second; we actually have a--a--
>> WRIGHT: Can I take the data center question? >> STRICKER: Oh, please, sorry. Go right ahead.
>> WRIGHT: Um, so the other thing to know about our data center is that we've been--we're
constantly investing in Search, and this just is a representation of one of many investments
in Search. And the cost of Search has been growing over time, and this is just totally
in line, um, with that growth. And it's what's to be expected with our--with our growth curve.
Um, but the thing that you should also note is all of the work you saw from Ben and Othar
about, um, the magic that our engineering team has done to make us use many fewer servers
than we would have expected when the first, um, when the inception of the project was
done. >> STRICKER: So I want to take an online question.
And this is for Sergey from Alexis Ibarra from "El Mecurio," and it's, um, question
is: Interface strategies seem to be evolving at an astounding rate. Where is this taking
us? What are your insights on the future of the machine/human relationship?
>> BRIN: Um, I'm embarrassed to admit that, uh, a sort of a phrase I was toying with--we
want to make Google the third half of your brain--we just had--that came up in conversation,
uh, the other day. But, uh, I am also astounded by the rate of innovation in user interfaces,
and I think this, uh, been caused by a whole bunch of factors. Many, many companies, you
know, that are contributing to it looking at the things that Apple is doing or even
Amazon with the Kindle, or many of the websites out there, um, I do think it's a little bit
of a new dawn in computing in the sense that things were I would say fairly stagnant with
respect to the desktop, uh, for a decade or so in there. Kind of throughout my graduate
career and shortly thereafter. And the web was really a big deal for the past 15 years
or so. Uh, but I'd say, over the past several years, based on the capability of both devices,
browsers, um, the technology just coming of age. And I think people being open-minded
to trying out new things. You can see lots of different platforms out there, lots of
non-traditional ones. There's a lot of exciting work going on. Um, you know, we definitely
are proud to contribute some of it. But I think this is just a piece of really changing
landscape in computing. And I think the things that you'll see come out over the next decade
from Google and from other companies are going to really change the way that you interact
with computing devices. >> STRICKER: Another question here? I'm sorry,
does anyone have the mic right now? I'm gonna take that as a no, so you gentle--gentleman
right here has been waiting very patiently. >> PARR: Thanks. Uh, Ben Parr, "Mashable."
Um, there's already discussion across the web about Instant Search's impact on SEO,
Search Engine Optimization. Um, what do you think is the impact on that, and do you think,
um, I guess Search Engine Optimization marketers and the like will have to do or change in
order to adapt to Instant Search? >> GOMES: I think the first point to note
is that, basically, ranking stays the same. Right? So people are going to adapt to a new
search interface over time, but the fundamental things of ranking stay the same with this
change. So in that sense, I don't think that there's a big change for search, you know,
for people who are trying to adapt their results to our--to our--to the search engine. On the
other hand, behavior and the kind of searches we see may change over time because of the
way in which you interact with the search engine may change, the kinds of queries you
do and how often you query, and so on. So I think that's a longer term effect, and you'll
see that much more over time and understand them much better over time.
>> STRICKER: Go right ahead. Sorry I can't see you so well behind the cameras there.
>> CLAYBURN: Hi, I'm Tom Clayburn, "InformationWeek." Um, given that the web search suggestions
are blocked in China, do you anticipate that this, uh, the Google Instant will be--eventually
be able to be made available throughout the world, or are there certain locales where
that kind of suggestion is going to be problematic? >> MAYER: Um, our goal is to actually roll
this out in as many different platforms--as you can see, mobile--and as many different
geographies as possible, and as quickly as possible. Uh, and at least my understanding
is that we should be able to, uh, have our--our predictions running on Google.com.hk, the
Hong Kong site. >> STRICKER: We're gonna take one more online
question here while we get the mics here sorted. This is from "Wired" Italy. "What do you think
are the concrete benefits for an individual user? Do you believe users want faster search?"
[laughter] >> GOMES: Yes.
>> WRIGHT: Yeah. >> GOMES: Yes.
>> WRIGHT: Absolutely, I mean, I think we talked a lot about the benefits. But it speeds
you up, gives you feedback as you go along, helps--it just makes things a lot easier for
you. Um, and it takes a lot of the effort out of searching. It helps search before you
finish typing. >> HANSSON: I think there's also a threshold
at which you speed things up and they become just, uh, there's a quantum leap in how much
interaction you do with them. That's what happened with word processors, drawing programs,
et cetera. So I think people are actually gonna be interacting with the search box for
the first time. >> MAYER: And I think that what we saw in
the Dogfood was really interesting. Uh, you know, we actually have likened Google Instant
a lot to, say, power steering or power braking. Once you get used to it, it's very hard to
go back to the old way of doing things. So when we would have--the few times when you
had to do maintenance and take our--our Dogfood down, we'd have people literally sending emails
and begging and saying, "Like, I don't even know how to search anymore, now that I don't
have this feedback and have this functionality." >> STRICKER: Yes, sir?
>> BRANDT: Hi, I'm Richard Brandt. I'm the author of "Inside Larry and Sergey's Brain."
[laughter] So I'd like to ask Sergey what the third half of your brain is--as we develop
technology like this that is just more intuitive and more predictive, there are always questions
about, um, privacy and how much we're giving up to Google in order to get all this technology.
Is that something you're concerned about at all?
>> BRIN: Uh, I think that...it's something that--privacy is something that we think a
lot about as a company, and, uh, and certainly users place a lot of trust with us. Uh, if
you look even starting with probably the more sensitive information--most sensitive information
is likely in your email, uh, and, you know, there are very important information like
your email that we really have to store and be great stewards of. So that's something
we spend a lot of time thinking about, a lot of time making sure we keep it secure, making
sure we have solid policies around it. Uh, and, uh, and I don't think Google Instant
is any different in that respect. >> STRICKER: Go ahead, Maggie.
>> SHIELS: I'm Maggie Shiels, BBC. Um, Ben, you said that you reckoned with Google Instant
that behavior and the kinds of searches that people are going to make may change over time.
Can you expand on that? How will behavior change? How will the kinds of searches change?
>> GOMES: Yeah, I think what happens, you find as you use it more, is you begin to explore
the area around the topic of your interest, right? Whereas before, you may have done one
query and looked at some results, now you may do multiple queries, right? Because you
see, it's very easy to go to the neighboring queries and suggest. Like you did with the
King Charles spaniel or with the musical in New York, right? That you can go and find
tickets and you can explore reviews. You can do all those things very quickly because it's
such a fluid experience, right? This is in the way when maps became infractive. You could
just explore a space. It was no longer like this click and the whole page updates. It
was just drag and it updates, right? And so, I think that's what's gonna happen. People
are going to just sort of explore their topic of interest much more than they do today.
>> STRICKER: Yes, sir. >> METZ: Cade Metz with "The Register." A
couple questions. One, during testing, there must have been a certain percentage of users
who didn't want this, and can you say how big that percentage was and what their specific
complaints were? And two, how does the project relate to the rollout of Caffeine? Was it
dependent on the rollout? Were they related in any way or were they completely independent
projects? >> MAYER: In terms of the testing, there were
some users who ultimately decided to turn off Google Instant, and we actually have a
switch that allows you to turn it off next to the search box. Just to the right of it,
it says, "instant is on." If you click it, it says, "instant is off." And so we did have
some users who chose to do that, usually for connection speed reasons, and it was a very
small percentage, so we were overall really happy with the number of people who kept instant
on and really enjoyed that. And I will let someone else answer that.
>> GOMES: So with respect to Caffeine, it's, the two projects are not directly tied, but
there are two in that Caffeine makes Index much fresher, so it makes caching strategy
a lot more challenging, and so we had to come up with a really great caching strategy that
basically updates the cache as you're crawling the web. That was a part that launches as
a part of this project. And other than that, the two projects, and Index is also bigger
and so there's, you know, more pressure in terms of the total amount of computation needed.
So it's an indirect connection, but there was some.
>> STRICKER: We'll take two more questions online. Two quick ones. One is, how will Google
Instant affect paid search? And the other is, does Google Instant mean the end of SEO?
>> WRIGHT: So again, the way that we serve and rank our ads is the same as always, and
as Ben said about SEO, I think the ranking of our search results is also the same, and
that's not, you know, these things, the user behavior may change over time, but I'm sure
our SEOs are smart and can catch up with us. >> STRICKER: Let's do one more quick one for
Sergey. Sergey, did you ever think--this is coming from Hector Henry in Bogota, Colombia.
Did you ever think that you would reach this point?
>> BRIN: That you can search as you type single characters?
>> STRICKER: I'm assuming so. >> BRIN: Well, I think it's definitely, uh,
this is, um, it's pretty amazing, the kinds of things that we can do today with benefiting
from the Moore's law of computation, the amount of compute power that we have with us, and
you guys saw all those optimizations helping us launch this product, but it didn't come
down to a factor of one. It wasn't--we did have to spend a lot more compute power on
it, and it's really thanks to all the advances in semiconductors and CPUs and so forth that
have been coming out, the multi-core processors that really help us do this. So even though
philosophically I might have certainly understood that we could, you know, in X years' time,
we'll have that much more compute power and will be able to do a search for every character
for X billion people per day. It's quite another thing to actually see it happen, and so it's
very exciting for me to see those hours click by up there. I think it's a tribute to people
who work on computing everywhere. >> STRICKER: Yes, sir?
>> MAN: So Google has done an extremely good job with actually, uh, I mean, helping 99%
of the people use the content that is being generated by 1% of the population, and there's
always this thing with user-generated content where you want to change that distribution
to maybe 95% of users with 5% contributing more. So what can Google actually do? Can
you play a bigger role in actually making more people contribute more content, because
the place where you have a little bit of weakness is where people like Yelp and Kora and stuff,
where they're getting more people to contribute content, and with location, you have people
recommending stuff and which is data that Google doesn't have. Can you do more to actually
get more people to contribute? >> BRIN: That's a great question. That's been
one of our guiding principles for a long time that in order to do a great job search, we
actually need lots of great content out there. Now, one of the things that we developed early
on to address that is AdSense, which actually pays for a huge amount of the content that
you see across the web and creates that incentive. For more user-generated content, we've worked
hard on products like Blogger which, you know, you can see tens of millions or perhaps hundreds
of millions. I need to check the stat of blogs out there contributed by all different people
around the world. And also content authoring tools, if you look at Google Docs and so forth.
But I should mention, there are many other great places that people generate content,
and we certainly don't expect people to generate all their content on Google. That's why the
sites that you mention, it's great to have those be part of the web community and to
be searchable, and also that's why we're really excited to provide the funding for many of
those sites via our advertising programs. >> MAN: Do you think that, for example, you
could exploit the traffic that you already have to improve local search? Because, I mean,
when I look at the map and I see a bunch of locations, I'm gonna find a few of those.
Can you incentivize me at that point to add reviews right there? Because it could improve
a lot more for as many as that you could provide. >> BRIN: That's a great question. So, specifically,
with respect to the local reviews, I think you can, today, provide reviews but it's a
little bit buried. But stay tuned for improvements in that area. Thanks.
>> STRICKER: Let's take one more online from Media Gearhead. I'm glad you guys are sitting
down for this one. Are there plans to make this even faster in the future or do you feel
this is, there's a technical limitation at present? Catch your breath, Ben.
>> GOMES: I think we should take, I think today, we've got to enjoy the speed of this
as it is, but tomorrow, I'm sure we're gonna go back and actually figure out ways to try
and make it be even faster. That's what we always do.
>> I would say simply, you ain't seen nothing yet.
>> STRICKER: Sure. >> IRINA: Irina from "Ad Age." So obviously,
you know, I'm interested in the bigger brands that are going to be showing up in the results,
because when you type "B," you get "Best Buy." When you type "T," you get "Target." When
you type, what else? "W," we get WalMart. So these are just kind of a natural, you know,
promotional traffic thing. So what about, you know, how does that affect smaller brands?
And the other thing also is, it seems that, um, the lower half of the page doesn't appear
to matter anymore, and page two almost doesn't matter anymore. So, I know we talked about
that the rankings are still the same, but it does seem to affect, because I'm not gonna
go to page two anymore, whereas I used to very much when I was interested in an, um,
you know, very small search that was kind of obscure.
>> GOMES: I think one part of it is that, I don't think the user intent changes, because
I typed in "W," I got WalMart. If I was interested in Wittgenstein, I'm still gonna continue
to type in "Wittgenstein," right? So, it's not, it's not that we are changing your user
intent. If you're interested in a smaller brand, you'll still type in the name of that.
And so we're showing you the thing you're most likely to see.
>> WRIGHT: And if you use page two less, I mean, that means that we're getting you your
answer much quicker, and you're not having to kind of scroll and wade through a lot more
information. >> IRINA: [indistinct]
>> WRIGHT: I mean, I think the thing to remember here is, this is a user focus launch, which
is really the way that our development process is done in search, is that we are totally
focused on our users and we really believe that that's gonna be good for advertisers,
and we have seen that to be true in the past number of years that we've been working on
search. >> IRINA: Does it search, in the beginning,
that the number of ads served will be a little bit less?
>> WRIGHT: No. >> STRICKER: The question was whether the
number of ads served will be any less. >> BRIN: No.
>> WRIGHT: No, that isn't what we've seen. >> IRINA: There doesn't seem to be more.
>> WRIGHT: I mean, the way that we serve and that we rank ads is totally unchanged, so...
>> STRICKER: We have a question up top from Richard, it looks like.
>> WATERS: Thanks, yeah, Richard Waters at "The Financial Times." Actually, I wanted
to pull together a few strands, including that last question. If you expect this to
change the way people behave with search and actually conduct more searches, then do you
think that actually people will click on more ads? In other words, you know, how will their
interaction with advertising change over time? Have you conducted any experiments into that
at this point, in terms of exactly, you know, how often they stop and click and what the
frequency of those clicks is? >> GOMES: I think one of our co-prints was,
at least working in search, is, think of the user and all else will follow, so in developing
this product, we have really been focused on the user experience, and we believe that
the other effects will follow via satisfying the user need more.
>> WRIGHT: And our ads are a great part of our user experience.
>> STRICKER: We've got one more question there. Yeah.
>> WOMAN: Thank you very much for the very nice features. I think this is really good
for mobile cell phones and such. Is that for a mobile market? Is that in the sight for
the more computer and such for mobile? And my second question is, when will Google release
these new search engine features in Japan and China? Because Kanji and those characters
must be very difficult, but [indistinct] does Japanese, so if you tell, I will be very glad.
Thank you very much. >> MAYER: The answer to both questions is
sometime over the next few months, so basically, we will be releasing the mobile version. That's
why we demonstrated it here. We do think it's a great thing for our users, because it's
often hard to type on a cell phone or while you're on the go. And in Japan, we do anticipate
being able to roll out this feature some time in the coming months.
>> STRICKER: Marissa, I got one more question for you online from Alex Chitu. "In the past
two years, Google has constantly changed the user interface for search results pages. Is
this part of a bigger plan?" >> MAYER: Well, I think, I think we're always
learning. I think our users also are changing. You know, we often reflect in search and look
at the types of searches people did, say, eight years ago or ten years ago versus the
types of searches they do today and their expectation around search results, and it's
constantly changing. And because their expectations for how good the results should be, how specific
the query needs to be, all of those things are constantly in flux, it makes sense that
the presentation layer also is. >> GOMES: I also think it is part of a bigger
plan. We're always making search better. That's the plan.
>> STRICKER: Mm-hmm. That's it? Another question here.
>> MAN: I have a question about the return of the situations. There seems to be always
five in a demo. Can you reduce it? Or can you expand it? Or it always has to be five?
>> WRIGHT: Right now, it's five, and that's, um, that's the UI that we think works best,
but as we've said here, we're always, um, testing and evolving UI.
>> STRICKER: So we're gonna be available right after this. We're right at 11:00 right now,
and I think we'd like to start the demo part, that all these folks will be free. Danny,
you're, you're in a hurry to get one in. Go for it. Well, we can squeeze you in.
>> DANNY: [indistinct] >> STRICKER: Hold on one second, Danny.
>> DANNY: The change puts more of an emphasis on the very, very top result, I mean, it seems
like even more than we've ever had before. Sometimes those results aren't right, and
so, how is that gonna impact things? I mean, the example I ran was, I typed in, "SEA."
I got a suggestion for search engines. The results that come up for search engines, the
top thing that you're listing is DogPile, which is, you know, an okay search engine,
but Google Instant is supposed to be so great and wonderful, I'd kind of expect to see Google
there. So what are you gonna do about improving the top results, especially since people are
gonna see them even more? >> STRICKER: Danny, why are you doing a search
for a search engine on Google? >> DANNY: Google Instant told me to do it.
>> STRICKER: Oh, okay. Fair enough. Touche. >> HANSSON: So, speaking of the top result,
we did see one very interesting thing in our usability studies. People would actually look
at the results, you know, that were just visible above the fold, and they would keep typing
to pull them up, essentially. So, you know, let's say they saw something on Search Engine
Land, right? At the bottom. If they just type, " SE," they hope that Search Engine Land
pulls to the top. People actually learn how to pull things up to the top even from page
two, to get to Irina's question. And obviously, you know, ranking is a challenge that we work
on all the time, so--except for that DogPile question.
>> WRIGHT: And I just want to highlight how important search quality and ranking is to
us. Ben pulled up a slide here of all the engineers working on Google Instant, but you
can imagine, we have a slide just as big and bigger of people devoted to search quality,
and they're just really tremendous engineers. It's a joy to work with them and we, as you,
we make about 100 ranking improvements every quarter, so we're, we're very focused on this
and really investing in search quality. >> STRICKER: You know, I would make one last
point to this. Ahmet Single is sitting right here in the front row. He did a blog post,
I think it was earlier this summer, called "This Stuff Is Tough." We get this stuff wrong.
It happens. Search quality is very difficult. It will continue to be. We'll continue to
work on it. But it isn't always perfect, and here we are. So, for those of you who want
to ask sort of more formal questions, we will be available after this. And as I said, there's
demos that we're gonna be doing. For those of you who would like to interact maybe a
little bit less formally with us, we're gonna be having a meet-up tonight at 7:00 p.m. at
111 Minna, where folks will be able to meet the team who invent this magic. So, thank
you so much for being here. We'll be available after this. And have a great day.