Real Time Search Event


Uploaded by Google on 08.12.2009

Transcript:
>>
MAYER: Good morning. Thank you for joining us here at the Computer History Museum. We're
so excited to be here within so many and surrounded by so many wonderful pieces of computing history.
And it's an especially perfect backdrop for today's search event, which focuses on the
future of search and the innovation that's happening in search. At Google, when we think
about the future of search, we think about four key elements: modes, media, language,
and personalization. What do I mean about each of these? Modes really refers to modalities.
How do people search? Today, the predominant mode is that people type keywords into desktop
computers. When we look at the future of search, we think they'll be many more modalities.
What happens with mobile phones? What happens if you can talk to the search engine? What
happens if you could give concepts or pictures that could cue off of--the search results
could cue off of? We're really excited about the fact that in the future there'll be many
different ways of searching. And when you look at how search will grow in the future
and change, we think these many different modalities are really what will drive search
forward and really grow with overall adoptions to even greater numbers than we see today.
On the flipside, there is media. Media is what is appearing within the search results.
And the Web, of course, has gotten very, very rich; books, movies, news, local, maps, all
on the Web. And so, our search results have to mirror that same richness that we feel
on the Web. So we're constantly looking at how can we make Google more comprehensive
and more relevant with regard to media. So it's not just about 10 blue links; it's about
the best answers. Media also leads the next piece, which is language. We really are working
at Google to try and branch across languages, break down the language barrier because this
focus on language in translation is what unlocks the Web. Today, we are able to translate from
51 different languages into all known pairs of those 51 languages. And we have 173 local
domains, because we really foresee a world in the future where you can search and find
your answer wherever they exist in the world, whatever language it's written in. And the
final component that we see in the future of search is personalization. We don't know
a lot about what the search engine will look like 30 years from now, 50 years from now,
and it's because search changes so quickly that we--it's hard to actually pinpoint what
that future looks like. But one thing we do know is that the results would be more personalized.
They'll know what you are an expert in. They'll know who your friends with and where you're
located. And, ultimately, our results will become more rich and more relevant to our
users because of that personalization. And there is a fifth component to the future of
search, which is the rate of progress. There has to be a consistent rate of innovation
pushing us towards this future. And we've always been a company that likes to launch
early and often and iterate. And we've done lots and lots of search features over the
years. And in October, we released a new blog series called "This We Can Search." And "This
We Can Search" chronicles each week all of the new user visible features that appear
in search. So, our users get to see the latest and greatest, and they also get a sense of
this pace that's driving us towards the future of search. And since October 2nd, in the past
67 days, we've actually launched 33 different search innovations. So that's one innovation
every two days. And if you look at the innovations, they include things like the new homepage,
Flu Shot Finder, personalized search on mobile, and they all fit neatly into these four categories
of media, of personalization, modes, and languages. And we're really excited about this overall
focus that we have on search, this rate of innovation. And today's event really drives
us forward even further increasing that rate of innovation and new features that we're
launching. Today's event will focus mostly on modes and media. And so, without further
ado, I want to introduce our master of modalities who's driving our Mobile Innovation, vice
president of engineering, Vic Gundotra. >> GUNDOTRA: Well, thank you very much, Marissa.
You know, I'm very excited to have a chance today to speak to you about innovations in
mobile search. But before I dive into those, I'd like you to join me and just take a step
back for a moment. Think about the technological innovation that was so profound that it changed
mankind. What comes to your mind? Maybe you think of Gutenberg's printing press. Maybe
you think of the steam engine. Maybe you think of electricity. And it's true; those innovations
really change the course of human history. But what's interesting is that at their outset,
the full impact of those innovations was not understood. You know, Gutenberg was broke
within a few years. The first mass-produced newspapers that came from Gutenberg's printing
press happened many, many decades later, or who could have predicted that the steam engine
would lead to the Industrial Revolution or that the invention of electricity would one
day lead to the Internet and to microwaves. At Google, we argue that that same dynamic
may be happening in the personal computer space. You know, PCs are, depending on how
you count, 27 to about 33 old. In other words, we are just in the third decade of the personal
computer revolution. And it may be that only now have our eyes become open to what the
possibilities may be. In fact at Google, we see three major trends converging in combination
with mobile phones that enable new scenarios that absolutely excite us. Let me talk to
you about those trends then I'll get in and show you some of those scenarios. Now, the
first trend is Moore's Law or computing. Now, I am in my 40s. All I've ever known in my
life is Moore's Law. Ever since I was a kid and I had electronic toys to the computers
I used as a teenager, I knew one thing for sure; that next year whatever device I have
would be better, faster, cheaper. It's the ubiquitous, pervasive law we know as Moore's
Law. And it's a powerful trend in computing. That's trend number one, computing. The second
trend is far more recent. That is the trend of connectivity. You know, think back just
as recently as a decade and a half ago. Think back to 1995. If someone came to you in 1995
and said, "Look, there's a time that's coming that billions of devices, every device would
be connected to every other device." Would you have believed that? You know, the more
technical you are, the more cynical you likely would have been. And, you know, if you think
in 1995, the best file servers of the day, things like Novell NetWare, you know, they
connected a few thousand simultaneous connections and the idea that you have billions of connections
seemed a little bit absurd. Today, of course, we take it for granted. The Internet has happened.
You don't even bat an eye when someone next to you takes out their cell phone and controls
their TiVo at home. It is an amazing change, this change of connectivity that has swept
the world. Computing, connectivity--and the third trend is the most recent, and that is
the emergence of powerful clouds. Now, when I say cloud, I mean huge amounts of compute
resources that programmers have at their disposal, data centers the sizes of football fields
that host massive amounts of computational power and the ability to manipulate huge data
models. When you combine these three things: computing, connectivity, and the cloud, and
then you think about what's happening with mobile devices, you get something very, very
interesting. Let's talk about mobile devices. Let me go grab a mobile device here. This
is an Android device, common Smartphone. And you think about something like this, it's
got built-in sensors. It's got a camera. It's got a GPS chip. It's got a speaker. It's got
maybe an accelerometer. You know, by themselves, these sensors are not that extraordinary.
Some of you may say, "So what? This camera isn't very special. My Nikon or Canon camera
from 10 years ago surpasses this in quality." And you could be right. Or the microphone--what's
the big deal about the microphone? You know, the microphone I use in my church or synagogue,
much better than this microphone. But not when you compare this in the context of computing,
connectivity, and the cloud. You see, when you take that camera, and you connect it to
the cloud, it becomes an eye. That microphone connected to a cloud becomes an ear. And in
fact, what we're going to show you this morning is innovations that are the combinations of
those three trends and mobile phones that allow you for the first time to do powerful
new things including search by sight, search by location, and search by voice. Let's begin
with search by voice. Now, some of you may remember that it was about a year ago that
we introduced Google--Google Voice Search available for the iPhone in the Google Mobile
App. And during 2009, we worked very, very hard to improve the accuracy and the recognition
rates of that voice recognition. You might be surprised at how good it's gotten. Let
me show you. This is Google Mobile App, and it's running this Voice Recognition Software,
Google Search by Voice. You simply bring it to your ear and speak a query. So, let me
try a query, something like pictures of Barack Obama with the French President at the G8
Summit. You know, unlikely that you would ever type in a query that long. And there
you have it. Isn't that amazing? Now, as impressive as that query is, it becomes even more impressive
when you think about what just happened. What happened was we took your voice file, the
digital representation of your expression, sent it up to the Google Cloud where it was
broken down into phrases, sentences, words, and those were compared against the billions
of daily Google queries we get. A probability score was associated with all the potential
textual representations, the textual expressions of what we thought you said, they were ranked,
and then the best ones were sent back to your phone all within what you saw here in fractions
of a second. Really amazing, amazing work. Now, we've done more than just work on the
improvement on accuracy of Voice Search. We've also been adding languages. You heard Marissa
talked about our focus on languages. Last month, we announced Mandarin Voice Search.
Now, think about our customers in China, our users in China, who have struggled to enter
in with a notoriously difficult Mandarin character set on their mobile phones now being able
use just voice. Let me do a demo for you. Same app, Google Mobile App available from
the App Store, except in this case, I'm going to show you the Mandarin version of this particular
app. Okay. There you go. So you see that it's in Mandarin and I'll try my only Mandarin
query that I'm capable of doing. [SPEAKS FOREIGN LANGUAGE] That was McDonald's in Beijing.
Let's see if we get it. There we go, all the McDonald's in Beijing. Okay. Now, I'm very happy to announce that
today, and joining English and Mandarin, we have a new language we'll be supporting and
that is Japanese. Now, instead of me trying to say McDonald's in Japanese, we thought
we'd have a native Japanese speaker, Toshi--Toshi, please come up on stage--and do a real demonstration
for you. So, Toshi, thank you very much. So, you're going to do a query. What kind of query
are you going to do? >> TOSHI: So, my first query, I want to search
for the pictures of Kyoto's Kiyomizu Temple. >> GUNDOTRA: Okay. Let's see you do this query.
>> TOSHI: Sure. Okay. Here's the screen. [SPEAKS FOREIGN LANGUAGE].
>> GUNDOTRA: Fantastic. Okay. So, that was a great query. But why don't you try a query,
a longer query, a query that you probably wouldn't type? And why don't you tell us what
it is before you do it. >> TOSHI: I want to try Voice Search for my
favorite restaurant around the Google's Tokyo office.
>> GUNDOTRA: Okay. Your favorite restaurant by the Google office in Tokyo. Okay.
>> TOSHI: So, I'm going to search for the By Address. [SPEAKS FOREIGN LANGUAGE].
>> GUNDOTRA: Oh, wow. Fantastic. Thank you. Thank you, Toshi. You can see why I've--when
I practiced trying to learn that, I gave up. So, thank you, Toshi. You know, as I mentioned
at the beginning, we really do get the sense that we are just now beginning to sense the
possibilities. In fact, our dreams at Google go way beyond what you just saw--as Marissa
mentioned earlier, in addition to voice recognition, Google also has massive investments in translation
from one language to another. In fact, we have huge compute infrastructure for translation.
Imagine the following scenario. Imagine that you could speak to your phone in one language
that Google could recognize what you said, take the text of what you said, feed it to
our translation infrastructure, translate it to a different language, and have it then
come back to the device using the text-to-speech engine on the device and have it play back
in a different language. In other words, it would be almost a real-time interpreter. Would
that be great? Let me show you a demo. What we're about to show you is a technological
demonstration of a capability that we hope to deliver in 2010. There is a technological
preview. Let me try something. Hi. My name is Vic. Can you show me where the nearest
hospital is? English up to the cloud, translated back into Spanish.
>> [SPEAKS FOREIGN LANGUAGE]. >> GUNDOTRA: Okay. You know what? I apologize,
Randy. I didn't have this thing plugged in. Let me try this again, okay? Hi. Can you show
me where the nearest hospital is? English up to the cloud, back to Spanish, back down
to the device. >> [SPEAKS FOREIGN LANGUAGE].
>> GUNDOTRA: Okay. Now, that's just an early indication. You can imagine us working to
get latency down even faster. We also demonstrated English, Mandarin, Japanese. In 2010, you
will see us dramatically accelerate our efforts and support many, many more languages. Our
goal at Google is nothing less than being able to support all the major languages of
the world, okay? So that's Search by Voice. Let's talk about location. You know, we were
reviewing some market research from Japan and we found the research quite surprising.
What the research showed was that, in Japan, consumers were holding onto their cell phones
24 hours a day and keeping them as close as one meter away. And we thought, "Wow, those
Japanese, they must really love their cell phones to have that cell phone with them all
the time." Until I went home, fell asleep, rolled over on my bed, looked at my night
stand, and right on my night stand was my phone. In fact, we're not that different than
the Japanese. You know, I suspect, if you look around now, either you have the phone
in your hand, it's in your purse, it's in your bag, you know, the phone, because its
location is likely your location has become the most intimate and the most personal of
all personal computers. At Google, in our engineering teams, we want to make location
a first-class component of all the products we built. Why? Well, it speaks to the personalization
point that Marissa talked about earlier. In addition to personalizing the results we bring
back to you, we can also just deliver them faster. For example, we included My Location
in Google Mobile Maps. Maybe some of you use Google Mobile Maps. Our data shows that, on
average, we save about nine seconds per user because when you open up Google Mobile Maps,
we know your location and render that location immediately. But we want to do so much more
than that. In fact, let me show you some of the things that we're doing to use location
in other properties. How many of you use Google Suggest? Okay. Lots of hands go up. Our data
shows that over 40% of mobile queries that come into Google are queries that result from
a user selecting a suggest item. Now, how could that be improved if we use location?
Let me show you. I have two separate iPhones here that I'm going to do this demonstration
on. Let me just get them started up. And in both cases, I'm going to show you a future
version of the Google.com homepage that supports this suggest. Okay. So that's one. And that's
two. Let me just make sure you can both see them. Now, we've done something here, we've
hard coded, we've hard coded this phone to believe, as you can see that it's in Boston.
And we've hard coded this phone to believe it's in San Francisco, okay? So, let's start
doing a query. How about RE, what would you expect? Well, there comes the suggest options,
Red Sox, Red Sox schedule, which makes total sense in Boston. Let's do exactly that same
query in San Francisco. And what do you think will show up? RE, REI, one of the most popular
retailers in San Francisco. Isn't that great? Customize suggest based on location. Of course,
there's other things we can do, for example, product search. This is the time people do
lots and lots of shopping. But maybe you're like me. Maybe, in addition to just shopping
online, there are times that you wish you could actually know if that product happened
to be available locally or could we at Google combine your location with inventory feeds
from retailers and tell you if that product was available locally? That's exactly what
a future version of Google product search will do. Let me do a search here for a product.
How about something like a Canon Camera? How about a Canon EOS Camera? I'll do a search
here. Let's wait for those results to come back. Now, you'll know that the top two results
have something very special. You see that blue dot? That blue dot says those two products
are in stock nearby. And if you select in stock nearby in combination with our partners
who are sharing inventory data with us, you can see that Best Buy has this camera 1.3
miles away and it's available at Sears which is about 7 miles away, isn't that great? All
right, let's go back, let's go back and talk about the Google.com homepage again. Have
you ever asked yourself, what's nearby now? Maybe you get to an event a little early.
You have a few minutes to spare; you're in a new place. And you go, I wonder what's nearby?
That simple query is very hard to answer. You know, think about how you might solve
it today. Maybe, you go into your car. Maybe, you'll use your car's navigation system and
go through the rigid categories they provide. Maybe you'll type in an address onto a map;
but query is so difficult because your location is the query. And what we're going to do is
make location a first-class object right within the Google.com homepage. In fact, let me show
you a brand new feature called "Near Me Now" on Google.com mobile. Right there, "Near Me
Now" and I simply select "Near Me Now" and look at that, it knows I'm at the Computer
History Museum and shows me other categories of areas or interests that I might have to
go search for. But we can even do better than that; what we really want to do is take your
location, send it up to the Google cloud, have the Google cloud reverse GO code your
lat-long location, understand the business or place of interest that you're at, look
around you for all other relevant places, rank them appropriately at--for their popularity
and then send those back to you, back down to your phone in a fraction of a second. That
would tell me what's nearby. But watch what will happen; I'm going to click that down
arrow right by explore right here and boom. There you go, all the places right around
the Computer History Museum. In this case, we've not only answered the question of what's
near, nearby now, but we've also, if you look at the ratings, also answered the question
of what's good nearby. Now, of course, we realize that you may do the search on more
than just the Google.com homepage. And I'm very happy to announce that, today, we have
a new version available today in the Android marketplace, a Google Mobile Maps for Android.
And among the many features that are in this new product is also this--is this What's Nearby
feature. You simply go wherever that you want to look at, any place on the map; it doesn't
have to be any particular place. You just long press. So I'll just long press, select,
and then, say what's nearby, and exactly that same feature is available today on Google
Mobile Maps for Android. Now, let me switch and talk about search by site. You know, of
the human senses, sight or vision is considered the most complex. It's estimated that up to
two-thirds of our brain, of the human cortex, is involved in the processing of visual images.
And it's widely believed by scientists that if we could figure out how the human brain
processes images, that would be a monumental step forward in understanding in how the brain
actually works. Well, I'm very happy to announce a new product available in Google Labs today,
a product called Google Labs, I'm sorry, product called Google Goggles which represents our
earliest efforts in the field of computer vision. Google Goggles allows you simply to
take a picture of an item and use that picture, that picture of whatever you take as the query.
Now, before I demonstrate the product, let me tell you a story on how I use the product.
Obviously, we "Googlers" test product before we make them available. And I was doing my
job testing the product. I had a friend, a couple, call me. They were scheduled to come
over for dinner, but they called and said they were running late. They were stuck in
traffic. In fact, they were stuck for traffic for one hour. So, when the doorbell rang and
I opened the door, both the husband and wife had to use the bathroom desperately. And as
I opened the door, they said, "Sorry Vic, we've been stuck in traffic, we need to go
use the restroom." I very happily pointed to where the restroom was and the wife handed
me the gift that they have brought, a bottle of wine. And she said, "While we're in the
bathroom, open this bottle of wine." Well, I did what you would probably do. I pulled
out Google Goggles and took a picture of the bottle of wine. Maybe, you've ever wondered,
is this bottle of wine any good? Oh, you know, I got a result sent back; it said that the
wine has hints of apricot and hibiscus blossom, just then the door opened and my friend came
out of the restroom. Of course, she's not a "Googler," so I took the confidential product
we were testing and I put it away. She said, "Please, please pour the wine," and so, I
poured the wine. And she said, "It's my favorite wine, you know, what do you think of it?"
And I tasted it and I said, "It's got hints of apricot and hibiscus blossom." She was
blown away. They thought I had a great wine palate. I'll be honest with you, I don't even
know what hibiscus blossom is. You know, let me show you a demonstration of that. I happen
to have that bottle of wine, not that exact one, but the same bottle. Let's try it, let's
take a picture. Let's launch Google Goggles and let's see what happens. Now, the lighting
conditions are less than optimal. I'm getting a huge amount of glare, so I don't know how
this is going to work, but we will try. So, let's see here, you--let me take an image,
a picture of this bottle and then we'll bring it over here. Okay, oops, well, it helps if
I actually get an image, so sorry, I moved that. Let's try that again. One more time,
and I will try to hold my hand steady; surprising, holding my hand steady wasn't a problem during
rehearsal. Okay, and then you see, it's scanning, attempting to recognize what it sees here
and, in this case, it has gotten some of the bristling items. And, of course, as you use
it and as we get more and more data, it'll become more and more accurate; pretty exciting
work. Now, you may think, "Well, Vic, I'm not going to take a picture of a wine bottle,
if that's all the Google Goggles does." Well, it does a lot more. It recognizes things like
CD covers, movie posters, books, barcodes, lots and lots of categories. You know, for
example, imagine that you're traveling in Japan and you come across this landmark. Now,
you don't speak Japanese, but you do know that's a famous landmark. How would you go
about and getting any information about that? Well, using Google Goggles, you could just
take a picture of it, and ask Google to help you. Let's try that. Let me come over here
and let me try to take a picture of that landmark. I'll pretend I was there. Okay, I got a picture.
Let's come back over here and show you the results. It's analyzing and there we go. How
about that? It accurately recognizes that landmark. Now, once again, it's incredibly
impressive when you understand what's going on. In this case, those images are being screened
to the Google or send to the Google Service, the Google clouds. There, our vision algorithms
are analyzing the image and looking for objects that it can detect. Those objects have signatures
that are created and then those signatures are matched up against an index that has over
a billion images in it. The best matches are ranked and then sent back down to your device
all in a fraction of a second, showing you really the power of devices that are connected
to the cloud. Now, some of you may ask, "Why is this in labs? Why is this product in Google
Labs and available today?" Well, we put it in Labs for two reasons; one, because of the
nascent nature of computer vision. We really are just at the beginning here, and the technology
is just getting under way. The second reason is because of the scope of our ambitions.
Google Goggles today works very well on certain types of objects in certain categories. But
it is our goal to be able to visually identify any image over time. Today, you have to frame
a picture and snap a photo; but in the future, you will simply be able to point to it, as
simple, as easy as pointing your finger at an object. And we'll be able to treat it like
a mouse pointer for the real world. Of course, we're a long way from that today. But today
marks the beginning of that visual search journey. And we strongly encourage those of
you with Android phones to download that product, give us feedback and help us really grow.
So let me wrap up here. You know, we really are at the beginning of the beginning. You
know, if you think about the mobile announcements that we talked about today, everything from
Japanese voice search to a new version of Google Mobile Maps that allows you to search
nearby or the upcoming changes to the Google.com homepage or even something like Google Goggles,
all of these are powerful demonstrations of what happens when you take a sensor-rich device
and you connect it to the cloud. Yes, it could be that we are really at the cusp of an entire
new computing era, an era where devices will help us explore the world around us that devices
can understand our own speech or help us understand others, devices that may even augment our
own sense of sight by helping us see further. We hope you're as excited as we are about
these mobile announcements. While it is just the beginning, the possibilities ahead inspire
us. So, thank you. Marissa, please. >> MAYER: So, you can see, search engines
that understand where you are in the world, search engines that understand you when you
talk to them, even search engines with eyes. These are the things that are going to change
the interface for search fundamentally, as we move forward. The other thing that will
fundamentally change the interface for search is media. Remember, media is what we refer
to in terms of the richness of the Web. And the way that richness needs to be reflected
on the results page. And it's not obvious, but media is really fundamentally a relevance
problem. Can you find the right information? Can you rank it the right way? And if you
look at the way media has evolved inside Google Search over the past 11 years, it's pretty
interesting. We started with just Web sites, URLs, a list of 10 new URLs that you needed
to be ranked. Then we evolved towards universal search bringing books, news, video, local
information, images, products, blogs, all onto the results page. But there again, there's
all kinds of interesting relevance questions: When should those genres appear at all? Where
should they appear? Which one from those genres? Which item from those genres should be surfaced
on the results page? It's a huge relevance challenge. And then think about the Web today;
the Web is hugely participatory. It's hugely fresh. I heard this morning that in Tahoe,
people woke up to two feet of snow. And I've heard that some of you were caught on it on
your way home; I'm jealous. But what's interesting is was it two feet, was it one foot, was the
snow blowing or not? And there are some official sources for that, but they don't always get
it right. Yet, there are user updates out there that do get it right. But how do you
find them? And how you cut through all of the updates that aren't relevant? That's why
we have one of our foremost relevancy experts here in the company to talk about some of
the challenges with media and relevancy. Please help me welcome Amit Singhal, Google fellow.
>> SINGHAL: Thank you very much, Marissa. So, we are here at this wonderful Computer
History Museum today. And before I get to today's big announcement, it's just fitting
that we take a moment and talk about the history of information flow. Now, I've worked in the
field of search for almost 20 years now. And what we are going to announce today is one
of the most exciting things I have seen in my career. But let me first take a moment
and talk about how information has flown over century. So thousands of years back, people
got most of their information by word of mouth. Sitting around campfires, from their tribe,
from their village, kids would walk up to the village elders who have all the knowledge
and say, "Hey, grandpa, should I eat those beans?" And the grandpa would say, "No, no,
no, them are poisonous, okay?" And the kids who listen eventually became grandpas and
pass that knowledge along generations. And it took generations for knowledge to get from
one point to another point geographically. And that was clearly not fast enough. Then
Guttenberg invented the movable-type printing press and this process of information dissemination
was parallelized. An author could write a book, thousands of books can be--books can
be printed. And then, they were sent on horsebacks and by boats to around the world. Village
elders around the world now had the power of that knowledge. And believe it not, some--up
to some 20, 30 years back, that was the primary mode of information transfer. And even though
great strides were made in the physical limits--in the printing technology and transportation
technology, the physical limits of printing technology and transportation technology still
made it so that for information to get from an author to their--to the consumers took
weeks, months and sometimes even years and that was clearly not fast enough. And then
came the Internet. And what I'm showing you here is one of the early Google servers that
are displayed here at the museum. And, now, suddenly, the world changed because billions
and billions of documents were available to millions and millions of users just through
the Internet, through search engines. And that was a great revolution. In the early
days of Google when I got here nine years back, we used to call that information every
month, and we would put up a new index every month. People will call it Google Dance. And,
clearly, a month was not fast enough. And then, the Web world, and then we started calling
every few days then every day, and then, every few hours to now when we actually can call
every few minutes. But, clearly, in today's world, that's not fast enough. In today's
world, the world is producing information from around the globe every second by tweeting,
by posting other updates, by creating Web pages, by writing blogs; you name it. Information
is being created at a pace I have never seen before. And in this information environment,
seconds matter. Let me just use a metaphor to explain my point, the old metaphor of a
library. Imagine a library with billions and billions of books. The librarian, just when
the librarian understood the billions and billions of books in his library, the librarian
realizes that there are a hundred million new books coming in every day. Now, the librarian
finally figures out how to deal with a hundred million new books arriving at his library
every day so that he can tell the patrons what they should look for. And just when he
mastered that process, the librarian realizes that there are a hundred million of people
running around in his library, adding pages, deleting pages, writing notes, adding things
to books and guess what? If they didn't find the book on what they were looking for, they
wrote 140-character note and handed it to the librarian saying, "Not in your library."
And that's what, that's the information environment today. And imagine in this library, a patron
walks up to the librarian and says, "Hey, I need to learn about President Obama." And
the librarian says, "You need that book, that book, that article, that image, that speech,
that video. And, oh, by the way, 17 people just left me a note that President Obama has
been talking to the Democrats in a special session on Sunday about health care and, by
the way, he'll probably be in Oslo on Thursday receiving Nobel Prize," and so on and so forth.
Now, imagine that librarian does all that in under half a second without breaking a
sweat. As of today, that's what Google can do. We are here today to announce Google Real
Time Search.
Google Real Time Search is Google's relevance technology meeting the real-time Web. Now,
I can't emphasize this enough--Marissa has said this right now--relevance is the foundation
of this product. It's relevance, relevance, relevance. There's so much information being
generated out there that getting to you relevant information is the key to success of a product
like this and that's where we, as Google, come in because for 11 years, that's what
we have done. Rather than talking about this more, let me just show you, I have with me
Dillon Casey, the product manager for this product and Dillon, what's happening out there?
>> CASEY: Well, as you mentioned, we weren't the only one's working yesterday, so one of
the great things about this new product is we can actually see what people are talking
about in regards to Obama right now. >> SINGHAL: So, Dillon types Obama into Google--Google
Research Page comes in and, wow, look at that, results just came in Real-Time, this page
has come to life. Do you see that?
This is results coming into Google's Results Page. As they are produced on the Real-Time
Web out there, okay, our users will get their results on the results page as they are produced
in Real-Time out there. This is the first time ever any search engine has integrated
the Real Time Web into the results page. So, let us show you a little more of this, Dillon,
why don't we click on the latest result links out there. When Dillon clicks on the latest
results link out there, he's taken to a full page of Real-Time Results as they are produced
on the Web out there coming into this page, you see there's a twit that just came in.
Here's another Real-Time page that we called from anwerstatyahoo.com seconds ago, another
twit that just came in seconds ago and--hey, Matt--man--what are you twitting out there?
Our good friend Matt just twitted and, guess what, it just showed up in Real-Time as he
twitted. This is the power of Real-Time Search. Now, let me just take three examples to demonstrate
what you have seen out here. So we have been testing this product internally with our Googlers
and as we have been testing this product, I have received good examples from my friends
about how they are experiencing this product. One time, one Googler had heard that GM's
car sales were finally stabilizing, so she typed GM into Google and, of course, our results
were right there--the biggest news of the day was indeed that GM's car sales have stabilized,
however, she noticed this latest results for GM section while she was searching and read
that GM CEO Fritz Henderson had just stepped down seconds ago, now this was the information
she needed right then and this is the power of Real-Time Search. Now, on the results page,
as I said, this is the first time we are presenting Real-Time Web on the results page--what you
see in this Real-Time section is a scrollbar to the right, so that if a Real-Time result
just scrolled past you, you can go back, go forward, in addition, what you see here is--seconds
ago, we had called an article from Business Insider and that was presented to the Googler
and there was also an update from Twitter.com, a twit from Twitter.com talking about Mr.
Henderson stepping down. Now, what you observe here is this is the whole Real-Time Web; this
is comprehensive Real-Time Web, with twits, with news articles, with blogs and so on and
so forth. Let me take another example. One of the Googlers was searching for Bank of
America, and, of course, our results were very relevant but the key part there was that
the Real Time Web was buzzing about how Bank of America had decided to re-pay its TARP
loan and all those results started showing up in this latest results section that I've
been talking about; but if you click on the latest result for Bank of America, as Dillon
did for the query Obama to take you to the second page, you are taken to this full page
of Real-Time results and what you notice here is this special new link that we are launching
today under our search options. We are very happy to launch this new search option called
the latest results, which are available to you by clicking through with the latest result
section or opening the search options on the top left of the result page and then clicking
at this new feature. And once you click on this new feature, you are given information
from the Real-Time Web, the comprehensive Real-Time Web from Twitter, from Wall Street
Journal, from a Website, the Loop21.com, and so on and so forth--comprehensive Real-Time
Web results coming to Google's results page in Real-Time. Now, let me take a third important
example, one of the Googlers was visiting home for Thanksgiving break in Maryland and
wanted to get H1N1 vaccine and heard--had heard that his whole high school was administering
H1N1 vaccine, so he typed H1N1 vaccine Arundel, the high school's name, and was taken to the
result page, which were very relevant results saying "Free vaccine to be distributed at
schools." So, very fresh results around Thanksgiving telling the Googler that vaccines will be
free and available at this high school but the Googler already knew that. He wanted to
know how long the lines were, what else is happening. So the Googler clicked on this
show options link that I just talked about and, having clicked on that, got the new options
panel and on this options panel, we are very happy to announce today that we are adding
a new update's link and by click on--by clicking on this--this updates link, you will get all
the twits and other updates coming into Google system in Real-Time. In this particular case,
when the Googler clicked at this update's link, the Googler got one very, very relevant
twit saying that the high school had run out of vaccines--we saved him a trip and he was
totally impressed. Now, for such hyper-local queries, we are--maybe one person is twitting
on their cell phone or very few people are saying something, Real-Time Search becomes
incredibly powerful because it shows you exactly what you need in your geography when you need
it. This was one single twit and it became available to that Googler right on Google's
page. So why I don't show you this page live. Hey, Dillon, what's happening out there man?
>> CASEY: Well, actually I hate to admit, I've been up here kind of surfing around while
you were talking but, you know, I don't know about you but I'm really excited about Google
Goggles and I've been watching what people are saying about it, it's--it's so cool. So
I just put the query in and I hit search and there's people talking about it right now,
in fact... >> SINGHAL: Wow, look at that. Vic, dude,
when did you announce it? How many minutes back? And here we are on Google's Result Page
Real-Time Web brought to our users for something you heard from Vic right now. Now, that's
incredibly exciting and, as you can see, we can go into all the full Real-Time page and
you can see the entire Real-Time Web for Google Goggles being brought to our users right at
Google's results page and--and the page that we have been talking about after that. Hey,
Dillon--hey, man, what are you doing? We are in the middle of the most important launch
of the year and you're playing with your cell phone, man? Grow up.
>> CASEY: Right, sorry. >> SINGHAL: What's happening?
>> CASEY: Sorry, Amit, a little guilty pleasure. You know, I've been kind of following the
Tiger story and it turns out this also works on mobile. I'm getting updates from the Apps.
>> SINGHAL: What? It does? Then why don't you share it with all of us?
>> CASEY: Okay, forgive me but you can see right here.
>> SINGHAL: Wow, look at that. This is Google's Real Time Search on mobile phones. So we are
very happy to announce today that Google's Real-Time Search is available on iPhone and
Android and anywhere you are--you need your information now, just pull out your Smartphone,
type a query into Google.com and you will get the Real-Time Web in your palm right away.
And that's the power of mobile Real-Time Search. At this point, I am also happy to announce
that our Google Trends page is coming out of labs--it's leaving labs, it's graduating
from labs. We are very happy. And we have added to it this new hot topic section that
Dillon is showing you right now. On the hot topic section, you will see what the Real-Time
Web is generating right now, what information is coming into from the Real-Time Web into
Google systems and, by clicking on one of those queries, you will of course see Google's
Real-Time Search Results. In addition, we have added a window down there where you can
type your query and see Real-Time Search Results. Now, before we go there, let me just say one
thing, we are rolling the Google Real-Time Search product over the next couple of days
and over the next couple of days, some users may not have access to this product as we
roll the binary out, however, you can always go to this new google.com/trends page and
by clicking on one of the hot topics, you will get to see Google's Real-Time Results
or you can type your own query into more hot topics. So Bernanke's speech is what's happening?
>> CASEY: Yeah, yeah. While you were giving your speech, Bernanke just presented a huge,
gigantic speech and the stock market went up.
>> SINGHAL: Wow, it happened right now? >> CASEY: Yeah.
>> SINGHAL: Man, I should call my broker. What's happening to my money now? Anyone knows?
My broker is not here. So what you observed here is this new google.com/trends page which
will take you to our new feature right away. Okay, so, we are all technologists here. I
have been working in the field of search for almost 20 years. We all love technology, you
love technology and I would be cheating you if I didn't tell you what went into building
a product like this. Let me just tell you, we literally had to develop dozens of new
technologies to make Google Real-Time Search as relevant as it is; technologies like language
models, we had to model whether a certain twit was genuine information-carrying twit
or was it just a weather buoy sitting out there twitting automatic twits. We had to
develop query fluctuation models. If queries fluctuate, the volumes fluctuate at a certain
rate, then something becomes eligible for you to see on your results page. We had to--we
had to develop the Real-Time content generation fluctuation model. If there's suddenly a lot
of content about Bernanke's speech or the stock market, something just happened. Now,
this is--these are some of the most exciting technologies that we have developed to build
a product like this. Today, we are processing over a billion documents a day generated by
the Real-Time Web out there and within seconds, we have analyzed the documents and twits and
updates and we have understood the documents and twits and updates and within seconds we
have seen a user query, which may have--which we may have never seen before. We don't--we
haven't seen one-third of the queries that we will see today ever before. So you take
documents that you've never seen before, you take queries that you have never seen before
and you merge them together and filter for relevance and bring it to the users within
seconds and that's what Real-Time Search is all about. Now, at Google, we talked about
the four pillars of search: comprehensiveness, relevance, user experience and speed. And
let me tell you one thing; I've worked in this field for a long time--I've worked at
Google for nine years and as the information world has exploded, as the amount of information
at the level out there has exploded, we are getting hundreds of millions of new items
every hour, the importance of relevance has gone through the roof. Everything is important.
Comprehensiveness is important. User experience is important. Speed is important. Indeed,
relevance is important but as the amount of information out there has grown, as much as
it has much and it's growing at the pace at which it is growing, relevance has become
the critical factor in building products like this. Now, let me just recap of what we just
talked about. So today we are very proud to announce Real-Time Search, a new latest search
option in the Google search options, a new update search option in the Google search
options, a mobile version of our Real-Time Search and the new trends page living labs
with a new hot topic section that would give you Real-Time Results. And I'm incredibly
proud of what we have built. As I study this--information is now getting to you instead of--from in
generations, instead of in years, instead of in months, instead of in days, instead
of in hours, in minutes, it's getting to you within seconds and I'm incredibly proud of
what we have built but at Google, we are never satisfied. It takes about one-tenth of a second
for lights to go around the world and, at Google, we will not be satisfied until that
is the only barrier between you and your information. Thank you. Let me hand it back to Marissa.
Thank you, Dillon. >> MAYER: Thanks. So the first time that updates
have been integrated into the search results and we have actually the most comprehensive
set of updates, too. We're so excited about this product. With that said, we didn't want
to rest on those laurels. We actually have two new exciting partner announcements. The
first of those announcements is with Facebook. Facebook will be providing us with the feed
of updates from their public profile pages, also known as Facebook Pages and these will
be appearing in Google's Real-Time Search. The second new partner, we have to announce
today is MySpace. MySpace will be providing us with feed of updates from all of their
users on any updates that are public and these updates will also be appearing in Google's
Real-Time Search. We have support from our partners here today. I want to thank Biz Stone
and the--and the team from Twitter for being here today and also Jason Hirschhorn from
MySpace, the chief product officer from MySpace and his team. Thank you very much for coming
and supporting the launch today. Last year on--in conclusion, last year on our tenth
anniversary, we published a blog on the future of search and that laid out the vision for
modes, and media, and language, and personalization. And I think when you look at today's announcement--search
engines that have eyes, search engines that can understand you when you talk, search engines
that know where you are and search engines that know what is happening anywhere in the
world and can bring it to you in Real-Time. It's amazing to see how far we've come on
realizing that vision in just one short year. And with that, I'd like to welcome Amit and
Vic back to the stage and we'll take some questions. Thanks.
>> BENNET: And we're also going to take some questions here in a second from online as
well while we set up. And if folks have questions here, feel free to step up. We have microphones
just right here. But, we're going to start with an online question first that's coming
from The Guardian in UK. This is for you Vic. We got to put you on the hot seat first. It
says, "Given Google's acquisition of Neven, to what extent can Goggles recognize faces?"
>> GUNDOTRA: It's a great question. Hartmut Neven is in the room, I believe. Hartmut,
are you here? There he is. Hartmut Neven had a company that we acquired and he is the leader
of this particular project concurrently at Google. His previous company did some pretty
amazing work around voice--around face recognition. And the technology that we built with Google
Goggles is very general. Of the billions of images in the index that we do recognize,
"faces" is one of those objects. However, for this particular product, we made the deliver
product decision not to do facial recognition. At Google, we deeply respect the user's privacy,
and we still want to work through issues around user opt-in and control. And so while we have
the underlying computer science technology to do facial recognition, we decided to delay
that until we have more safeguards in place. >> BENNET: That's great. And again, if folks
have questions here, just go forward, put up your hand. And if you could just introduce
yourself, so that people who are listening in online can know where the question is coming
from. >> TENAKA: Hi, this is Akito Tanaka from the
NIKKEI. I had a question regarding the Real-Time Search. What is the advertisement opportunity
in that area? >> SINGHAL: So, right now, we are concentrating
on bringing the most value to our users with all the wonderful partnerships that Marissa
just announced, and our partnership with Twitter. And I believe that this phase is very, very
young. As time goes on, new models would develop, and all the companies that we are talking
about are experimenting with multiple models of how to generate revenue from all this wonderful
real-time information that the world is producing. I think all the companies like Twitter and
others have added tremendous value to the world, because we can figure out what these
key conditions are in Tahoe right now. And I can figure out how the traffic is like in
Bangalore right now. And you name it, right? There's so much information out there. And
as long as the product brings value to users, I think new models are recognizing and various
of the revenue streams emerge. And you will see a revolution in that space over the next
few years. >> WATERS: Rich Waters in the Financial Times.
Can you tell us more about in Real-Time Search how many sources you're crawling? How often
you're crawling? Are you taking feeds from Twitter and Facebook, and other places? And
longer term, you know, how much real-time information you can be able to run?
>> SINGHAL: That's a great question. So, we are crawling a lot of content. As I said,
right about a billion pages a day. We are crawling everyday. We are crawling from many,
many sources. We are going out to the web, and we are crawling all those sources, all
the good sources out there. Definitely, all the new sources, but not just the new sources.
If a company announces a new product into the press release, yes we will get it. If
a blogger writes a blog about something, yes we will get it. And we will do that within
seconds. So, we are crawling, we are casting a very wide net. The key here is comprehensiveness
of real-time information, and integration with Google's search results. Those are the
two keys--those are the two key design principles behind this product. And indeed, we are taking
feeds from our partners, Twitter, and going forward very soon from MySpace and Facebook.
And we would like to get as much information as there is out there via feeds or any other
mechanism, because our objective is to build the most comprehensive Real-Time Search out
there. >> MAYER: And I should also add that on our
latest mode, we actually do have other update providers, including FriendFeed, Jaiku, Identi.ca,
and TWiT Army. And as Amit said, we'll be working to bring Facebook and MySpace into
that functionality over the next few weeks. >> BENNET: Well, we have further questions
from here. Oh yeah, go ahead. >> HELEN: Hi, Helen Malbay [PH] with the Financial
Times Germany. What about availability of Real-Time Search on your non-US site?
>> SINGHAL: That's another great question. So, we at Google strongly believe in all of
our products becoming international rapidly. This first launch is available in all English
speaking locales. That would be the UK, Canada, India, Australia, and New Zealand, and so
on and so forth. And very soon, some time in Q1, we are planning on launching many new
languages. So, this is one of our top priorities in this project. Our first priority was launching
it, stabilizing it, making of our infrastructure and relevance work, because that, as you can
imagine, has been a hard challenge. And as I showed you with all the new technologies
that we have had to develop, that was our first focus: building a product that would
bring value to the users. And going forward very soon, we are going to rapidly internationalize
this product. >> BENNET: So, we're going to take one more
online question here. This is another one for you, Amit, about relevance. How do you
prevent spammers from taking advantage of the Real-Time search results? This is from
Steven Bivins [PH]. >> SINGHAL: This is, you know, this is something
that I know something about having run Google search for about nine years. We have the best
systems in place to prevent gaming of the system, okay? Our spam lead out here sitting
with us, Matt Cutts. And Matt runs the best spam prevention team that there is out there.
And we have had experience with this for so long, we have developed algorithms inside
that can see what's going to happen next and counteract almost before it happens. Matt's
team has developed some of those algorithms. And real-time is moving--for us, real-time
is moving from minutes to seconds, and we are already in the game of running the system
that's minutes fresh. And we do a great job. You find Google results very useful that we
called a few minutes back. And Matt and his team and the rest of the team at Google are
experts in this area. We know many things about it, and that's how our real-time product
is already so relevant. >> CURTIS: David Curtis. A quick question
on the customization, are you going to be able to integrate social memory counts so
I can customize the prioritization of Real-Time search results by the people who I care about?
So in other words, using Facebook connect or LinkedIn, or Twitter a lot to sort of prioritize
or even let me selectively choose, like, the real-time results that I wanted to deal with.
>> SINGHAL: You are just picking questions out of my mind. That's so wonderful. We are
very excited about what's happening. And the key thing that we are excited about is we
are just getting started. You mix with this real-time search all the key principles that
Marissa talked about, and Vic talked about. Localization, personalization, and now you
have a real-time product that everyone would like to use it. In addition, recently you
have noticed we have been experimenting with social search in our labs. Marissa launched
Social Search a few weeks back. And that is an angle of personalization where you see
results from your social circle alongside the most relevant results from Google. Now,
you can just imagine when you merge these multiple technologies that we have developed
here from Real-Time Search to social search to job location-based search, what you will
get is a product that you would love, and that's coming very soon to a Google near you.
>> BENNET: See, Glenn up here in the front? >> GLENN: I'm really curious [indistinct].
Could you further expand on the distinction between Facebook's going to give you the public
feeds, which the stuff that members are already designated, is okay to be seen by anyone,
and then MySpace is giving you all the feeds, I mean instead of...
>> MEYER: Right. >> GLENN: So, that's anything that a MySpace
member will get, please. >> MAYER: So, Facebook has a product called
Facebook pages, which are special public profiles for specific entities, their feed that they're
providing us will have the updates that come from those pages. For MySpace, it covers all
of their user's profile pages and any update that's designated as public. So if I were
a user on MySpace and I saw that my updates could be public, those will all be coming
to us in the feed. >> GLENN: [INDISTINCT]
>> MEYER: That's right. So that users can decide what they'd like to see offered by
this feed to Google and then search and how broad that they really want to share it on
the social network, in general, by using the privacy controls available on each of those
networks. >> MCCRACKEN: Harry McCracken with Technologizer.
You talked about how you have these partnerships with a bunch of major sources of real-time
content. You need those relationships and those feeds to do this or what would happen
if there was something that people are excited about that you did not have a relationship
with? >> MAYER: I think that overall our goal is
always comprehensiveness. But our mission is the world's information making it universally
accessible and useful. And we really do mean the world's information. As Amit can attest
to, you get better results when you have more items to choose from, when you can analyze
them, understand how they relate to each other. So the more comprehensive we can be, the better
we can serve our users. And that's why we've had such a focus on making our real-time results
that we're launching today already the most comprehensive available. And we're taking
that even further with the MySpace and Facebook partnerships. In the future, if there was
something else, obviously we'd want to partner with and include those sources as well.
>> BENNET: We have a question from Danny up in the front here.
>> DANNY: Can you go back and clarify with Twitter what financial deals are there, if
there are any, and then the same for Facebook and MySpace. Are there ad deals that you're
paying for this stuff or is it just the goodness of their hearts or what?
>> MAYER: We cannot disclose the financial details of any of the deals.
>> DANNY: [INDISTINCT] >> MAYER: I can't. I'm sorry.
>> DANNY: [INDISTINCT]. If we go back to MySpace, we've got Murdoch saying that he wants you
guys to pay him to carry his news results, right? But then, with MySpace, either they've
just decided to give you this stuff because they think it make sense and that's his company,
and apparently that information is free for them to hand out, or you're actually paying
for it. So, it seems reasonable to ask whether or not somebody is getting something out of
this financially. Even if you can't do the details, it's either they're doing it for
free or they're not doing it for free. Can't you give us that?
>> MAYER: Yeah, I'm sorry, we can't confirm. >> SINGHAL: We don't exclude any source, any
source of real-time information we would really like to have integrate it with our system.
And we let our algorithms for relevance decide what updates or tweets, all blogs, all news,
all web pages to serve this to the user. So at Google, we are all about comprehensiveness
and we will accept all sources of real-time information and we'll build the most comprehensive
real-time search. >> But are those--I mean, are you applying
the same relevance algorithms? You write the algorithms, so you know what they can do.
And are you applying the same algorithms you applied to normal web search just on a faster--on
a faster footing or how have you have to change it?
>> SINGHAL: So as I was talking about earlier, we have had to develop at least a dozen new
technologies to make Real-Time Search work as well as it does, because clearly in Real-Time
Search, you need to have models of information fluctuation. As information fluctuates out
there in the real-time world, all as information fluctuates in our query screen you have to
react to that. And clearly, those fluctuations are useful for traditional web search. But
for real-time search, they are just incredibly critical. So, I wouldn't say it's exactly
the same algorithm that we can use, because to work with this amount and this base of
information generation, you have to develop these new technologies. And what we have developed,
some of these are just amazing technologies, right? I've work in this field so many years
and I didn't think we would develop these technologies so fast at the rate that we did.
So, I'm incredibly proud of how much relevance we have brought to the product based on the
technologies we have developed using our experience with relevance.
>> BENNET: So we're going to take one more question from online for Vic. The question
comes from Luke Wilson with more--I don't know if it's the Luke Wilson, but from a Luke
Wilson--with more Android phones and possibly even Google Android hardware, does Google
intend to reduce support for non-Android devices? >> GUNDOTRA: Absolutely not. You know, our
desire is to reach our customers on whatever platform they're on. And today, there are
a variety of smart phone platforms like the iPhone in Apple is a strong partner of ours,
like a BlackBerry, Nokia. At times, we choose different priorities in terms of which one
we do first. But it is our goal to reach as many as possible based on the technical capabilities
of that underlying platform. >> BARAK: Hi, Sylvie Barak from Hexus. I wanted
to know, first of all, do you feel that your Real-Time Search will be the death of journalism?
And second of all, everyone knows that knowledge is power. Does that make Google the most powerful
company in the world? >> SINGHAL: So, let me answer it to the best
of my ability. Your questions are clearly very loaded. So, journalism has its role and
it always will have that role. Information is indeed power, and what you in this room
are doing, are empowering the world, as we speak, with information about what's happening
here right now. So, I can't even think about putting those two words together, death of
journalism based on Real-Time Search, because you bring so much value to the world that
this value has to be brought to the world. And regarding your second question, our goal
at Google has always been to bring timely information to our users. And clearly we are
empowering our users with the information that they need now. We have been in this business
for 11 years. We get the information, we do our special relevance work, and we bring it
to our user. So, I think it's all about user empowerment. I personally have felt empowered
many times when I had the knowledge in my hand through Google and I walk into situations
that I had to. So it's all about user empowerment. And Real-Time Search is the next step in that
direction. >> MAYER: Yeah, I will just add that, you
know, our purpose is really around facilitation and reference. Getting people to do their
search and getting them off of our site and on to where the information exists in its
native form as quickly as possible. And so from that, like I said, you take a little
bit of exception with the question of we have the information, we don't. The web has information
and we want to get the user's back out to the web as efficiently as possible.
>> BENNET: It's probably worth knowing too that Google sends billions of clicks each
month to news, publishers, and this will be yet another channel through which to send
those clicks. Yeah. >> HELLER: Hi, Michael Heller. I'm with the
Google Technology User Group. I just--my question is around the integration of real-time results
with the rest of Google results. I saw, the latest within the web page. And I'm just wondering
what your vision is for long term. From a user perspective, how much does the user thinking
about is something real time result versus some other kind of result and where you're
trying to go in terms of that direction? >> SINGHAL: Another great question. The power
of our universal search is that users don't have to think about whether they should be
searching here or there. They should be searching in Google search box, and all information
that is relevant to them at that moment should surface on the search results. We have made
great strives in universal search with integrating books, videos, images, news, blogs, and so
on and so forth into Google Search. We don't think of it as this search or that search.
We think of it as Google Search. Whatever needs to be seen by the user now should be
integrated on the Google's results page. And Google Real-Time Search is just the newest
feature of Google Universal Search because this is the genre that's very relevant in
today's web and we have just brought it right to our users with our integration with Google
Search. >> MAYER: Yeah, I would just add one example
there, which I think drives on how much of a relevance problem this really is. So for
example, think about when you're searching for a product. So, apparently, a few weeks
ago there was a massive stroller recall, and what's nice is when we're actually dog-fooding
this, this what we call to using it internally, when we're dog-fooding Real-Time Search, we
did search for that stroller and only did you get places where you could go and buy
it, but you also were alerted to the fact that there are been a recall, which I would
argue for users. That is a very important piece of information. If you're about to go
and make that purchase, you want to know that there's been a major news issue with this
or potentially a major safety issue with it. And I think that that shows the power of the
Real-Time Search. And in that case, that Tweet, that news article is incredibly relevant,
and that's why we surface it on the same page as our search results.
>> Marissa, you just mentioned that it's still Google's idea to get users as fast as they
can to other pages on the web. We've seen some changes though from Microsoft and Yahoo
that's in to think that creating pages from information around the web is a better way
to go and gives people answers. You also mentioned answers. Can you give a sense of where Google
could have lay ends on that kind of creating user interface or sort of still having algorithm
being the king? >> MAYER: I think that our view is that we
overall believe that the web thrives on openness. And so, the reason that we have this amazingly
rich set of data to search and provide on our search results is because the web is open
and there's like a huge amount of participation to say, "Oh, you now, we'll develop the most
authoritative page on this." I think it's problematic, because it does make it a much
more enclosed system and we want to be much more inclusive. That said certainly as we
evolve search, there comes a point when you do want to be referring to things potentially
more as entities. Now here's a restaurant. What can you tell me about it? Right? And
we do that by offering heterogeneous search results. So this will really allow you to
see not only the canonical page for that restaurant, but also reviews, et cetera. I think we'll
play with the user interface with that, but again, the point is still that the best information,
the richest information is out there on the web. We want to get people there faster, but
we don't want to hold them on what we would call the most authoritative page or host that
page. >> SHANKLAND: Yeah, this Stephen Shankland
from CNET News. I appreciate your focus on relevance and recency. But in my observation,
a lot of times those things-—they don't necessarily go together. Do you have any way
of putting truth into the equation? There's a lot time where--a lot of situations where
time goes by and the truth seeps out. So, is there a way that you can actually assure
people that they're not getting connected to rumors and things that are potentially
factually wrong very quickly? >> SINGHAL: No, this is a very good question
and a very, very tough scientific problem that the research community is also thinking
about and we are also thinking about. And right now, a straightforward answer to your
question is we emphasize quality and relevance. And that often brings the truth out. I say
often because there are maybe occasional chances when the truth is somewhat grey and not black
and white. In which case, it can be debated. But it's a very hard problem because language
understanding is still an unsolved problem. We have made great strides in language understanding
at Google. However, it's still an unsolved problem. We are very excited about the algorithms
we are developing to understand language, but what you are talking about is, in some
sense, the grand challenge for language understanding. So, we are excited about the strides we have
made but that's our ultimate objective years down the road to get to that point. And that's
why after having worked in the field for 20 years, I come into work every morning like
a kid going to candy store, because I get to work on all these things.
>> MOORE: Patrick Moore. >> GARY: I'm sorry. I'm speaking. This is
Gary. I'm with SearchWise [PH]. Actually we are hosting some Real-Time information on
our Website. So, is there any way that we can submit the channels to the Google Real-Time
Search, you know, where the resource? >> SINGHAL: Sure. Please, talk to Dillon,
who's sitting here after the event. >> GUNDOTRA: Yup.
>> GARY: That's also a general question for a lot of Real-Time hosting Websites.
>> SINGHAL: Right. >> MAYER: I think it make sense at some point
to have a standard API and a standard feed that we accept and I think we will be moving
in that direction as we evolve the product. >> GARY: Okay, thanks.
>> PATRICK MOORE: Patrick Moore. One of the questions that I've noticed come up on Twitter,
yeah, I think it should be repeated is a lot of energy and effort is spent on PageRank.
What does this do in the Real-Time Search? It seems like, you know, you got a real issue
here, how are people going to deal with this as PageRank impact the Real-Time Search results?
>> SINGHAL: So, Page Rank is a very important piece of our ranking relevance technology
that we use for our entire Web search. And PageRank is indeed one of the hundreds of
factors that we use in our ranking system. And for Real-Time Search, we have all those
hundreds of factors. Some may not be as powerful in the context of Real-Time Search, some maybe
more powerful. In addition to Page Rank and those numerous other signals, we have had
to add these technologies that I talked about. Like, language modeling, right? How do you
model language that it actually is finding good relevant information? And so, PageRank
is always a very important piece of our technology including Real-Time Search. We have just had
to develop many new things to make Real-Time Search as relevant as it is.
>> MAYER: And I would add that Page Rank is really about finding authoritative pages.
One of the more fascinating things that we've seen or we're beginning to see inside some
of the real-time data is authoritativeness exist there as well and there are signals
that indicated. So, for example, re-twits and replies and the structure of how the people
in that ecosystem relate to each other, you can actually use some of our learnings from
PageRank in order to develop a, say, you know, an updates rank and/or an updater rank for
the specific people who are posting. So, this is something we're beginning to experiment
with. It was interesting to see that same parallel where PageRank looks at links. You
can actually look at the very mechanisms inside of these updates streams and searches and
in a sense authoritativeness in the same way. >> BENNET: We're going to take one more online
question from Eric Wester from wikiHow, who's asking: If there's an option to disable the
scrolling feature, he says he realized there's a pause but what about disabling it all together?
>> SINGHAL: So, we really have experimented tremendously with this user interface for
Real-Time Search. And based on early positive and good feedback like this, we did introduce
the "Pause" button. And after a lot of experimentation, I think the current interface is serving its
purpose of conveying to you the Real-Time nature of your query and providing the Real-Time
results. And all—we are always experimenting with everything that we do, including user
experience, not just for Real-Time Search, but for everything else. And we would be working
on the user experience for over the entire search system going forward, clearly Real-Time
Search is the new feature we are very excited about, and we would do a lot more work in
the user experience direction of Real-Time Search going forward.
>> MAYER: And I think we also have to acknowledge this very early in the evolution of the feature,
and we don't know that this is exactly the right user interface. It may ultimately change
and so--and we also want to always to honor our user's preferences. So, if they have expressed
a preference to us, we don't want to later say, "Well, hey, here's a whole new way the
user interface could work and how do we interpret that option there." A few years ago when we
first put PDFs into our search results, we had the PDFs somewhat imbalanced, never showing
up too frequently. And lot's of users mailed it and said, "Can we—can I just turn off
having PDFs in the search results?" And we also said, "Actually, you know, bear with
us. The relevance of PDFs will get much better." And sure enough, it did. And we didn't want
to have this sort of mechanism where you turn it off and then we have to re-introduce it
later once it was better. We actually think that the Real-Time Search relevance is already
very good, but that--we can anticipate changes to these search results. So, a "Pause" was
the right compromise. >> KENNEDY: Nell Kennedy. I'm wondering about
time to locate on the GeoSearch, can you talk about your coverage for non-orbital sources
of field location data worldwide through your scanning MAC addresses and looking at cell
towers right now? And have you done any--have you done any work looking at how Galileo could
possibly change how you do search internationally? >> GUNDOTRA: When you say Geo Locate, you
mean a feature like My Location? >> KENNEDY: Correct.
>> GUNDOTRA: Okay. So, our time to locate varies depending upon the source of the geo
data. So, on a cell phone, if it's GPS data, cell phones depending on make and model can
sometimes take a very long time, up to 20 minutes to get their initial GPS fixed. In
those cases, we fall back to using cell tower, and if the phone has an A-GPS chip, A-GPS
gives us assist to GPS. That assist comes from using the unique identifier of the cell
tower, and we're able because of a very large database to almost instantaneously give you
a reasonably accurate fix until we can get the true lat-long once the GPS kicks in. So,
that gives you that very, very fast experience in Google Mobile Maps.
>> KENNEDY: And do you feel like you have good coverage internationally for those types
of location sources, other than cell--other than GPS?
>> GUNDOTRA: The coverage has grown by an order of 92 this year. And, obviously, the
more and more phones that carry Google Mobile Maps increases our coverage. So, we're very
happy, most places internationally, there are a few very rural areas that we continue
to drive for a better coverage in. But the rate of growth is very encouraging. They will
have broad coverage. >> KENNEDY: Will Google's Real-Time Search
API gets supported in the Google Search API in the future? I would rather like see that
in a long tail adaptation. >> SINGHAL: So, that's a great idea. We haven't
yet looked into the details of that. And I assure you, we'll be looking at the details
of that going forward. >> BENNET: Yeah, one more question online,
which is just, when is all the stuff going to be live both in the mobile front end and
on the Web front? >> GUNDOTRA: So, I'll take the mobile front
stuff. Live today is Japanese Voice Search. Brand new in the Android marketplace is Google
Mobile Maps Version 3.3. Or if it isn't live now, it will be live in the next few hours.
And also, Google Goggles, as you saw from the Twit, is already available in the Android
marketplace. Some of the new innovations that you saw on the Google homepage, like Near
Me Now, those are in the coming weeks ahead. We can't exactly predict those dates because
of some of the holidays, but it's in the very imminent, in the near future. Things like
Product Search are probably a few more weeks beyond that. But that gives you a timeline
for everything we discussed except one thing, that fascinating demo I showed where the devise
was able to translate from English to Spanish. That was a technology concept demo. You'll
see the first products from Google that start to include that technology some time in Q1.
>> SINGHAL: So, for Real-Time Search, we are starting the rollout process today. By the
end of the day, some percentage of Google users would start seeing Real-Time Search.
And in the coming days, we will complete that rollout as our systems roll out to the various
data centers that we have. For now, if you want to access Real Time Search, please go
to google.com/trends and you can click on the "New Hot Topics" panel on the left, or
type your query under that panel in the "New Search" window we have added and you will
get Google's Real Time results now. >> You don't get at the Website [INDISTINCT].
SINGHAL: We can check that. We'll make sure. >> I have heard that...
>> SINGHAL: No, I'm sure. You know, maybe some of the binary we just did needed.
>> [INDISTINCT] before that. >> SINGHAL: Okay. We'll check it right away.
Someone's checking it. >> JONES: Bobby Jones from the Guardian, again.
I wanted to ask Vic about the visual search. You said there are a billion images already
in the computer vision data. I'm just wondering who--whether Google owns, like, the canonical
encyclopedic entry of what that image is or whether you determine what that is by, you
now, a Web search algorithm. You know, is the Empire State building determined by everyone
on the Web saying it's the Empire State building or you deciding that's what it is?
>> GUNDOTRA: So I could give you a very compelling answer but I would be remiss as the person
sitting directly to your left is the engineering lead behind it. And so, in this case, if you
could hand over the microphone to Neven--no, Hartmut. Hartmut will give you the answer.
>> NEVEN: Actually, one of the most interesting parts of our system is a technology called
"Unsupervised Learning." So, essentially, the algorithms will go out and build a model
for visual recognition in a completely unsupervised memory based on photos we find. And then,
a model, let's say in your example, for the Empire State building, will emerge as a reflection
of what's on the Web. >> JONES: Okay. My follow-up question then
was: Would it be possible to Google-bomb visual search?
>> NEVEN: In principle, yes, but, I mean, let's say, we have techniques to prevent things
like this. >> JONES: Okay.
>> BENNET: So, I think we have time for just a couple more questions. One over her in the
back, yeah. >> MACENA: So, currently--I'm Chris Macena.
I'm sorry. You're doing a lot with text-based status updates. I'm curious if you're looking
to expand that to other types of activities that people are doing on the Web that are
being recorded through other types of social networking services and systems, again, just
beyond status updates. >> SINGHAL: Indeed. We are very excited about
what the future holds. Today, we are starting our new Real-Time Search with text-based status
updates and the rest of the Real-Time Web. And we are learning about how to do relevance
in this world and we have done a very good job of getting you relevant real-time results.
As time progresses, we have image search technology, we have video search technology, and we will
be accepting all those forms of real-time information. Some of the greatest information
I've seen real-time is held in Twitpic, for example, right? The cable broke on the Bay
Bridge or something like that. There's excellent information there and, indeed, we will be
integrating that going forward. >> MAYER: As also, MySpace, one of our partners,
is already looking at how they can take some of the non-textual updates and make them available
to us. >> RIPER: Hi. I'm Van Riper. I am actually
one of the co-leaders of the Silicon Valley Google Technologies User Group. But in my
day job, I work for Crillon (ph), which is a product-based local search in real-time
inventory look-up. So I was kind of curious--I should be looking for another job. Could you
say a little more about the availability of the integration of product availability in
your results that you mentioned early on before all these real-time stuff? It's just very
exciting. Did you have a question? Early on--yeah, early on demo and something about it being
able to get real-time product availability, product inventory.
>> GUNDOTRA: Yes. >> RIPER: And when you're--when you then answered
the follow-up question about availability of stuff, you didn't even mention that so
I was just curious. >> GUNDOTRA: Oh, I'm sorry. That will be integrated
into our product search sometime in Q1. The partners I demonstrated there were Sears and
Best Buy. And, obviously, we're working with many other retail partners to get inventory
data so that we can combine that with the user's location and deliver that experience.
>> RIPER: Got it. Thanks. >> GUNDOTRA: Yeah, sure.
>> SINGHAL: Okay. Let me just take a moment. And Dillon tells me that google.com/trends
should be working now. Maybe we hit a minor glitch so we will check that again. But please
test it. >> SINGLE: Ryan Single from Wire.com, again.
Could you give a sense of what relationship there is between the Real-Time Search Index
and the Google Web Index, and whether one is feeding into the other or not?
>> SINGHAL: Another great architectural question. Google Search Index over the years has evolved
to be updated every few minutes, even like, you know, within a few seconds. And Real-Time
Search Index is just a continuation of that technology that we have been building for
many years now to do things like Google New Search, or even calling the Web very fast.
The new thing that we have added into this is update receiving and indexing and merging
with the index. And that has been a great new technology because, at that end, we have
built some of the technologies I showed you for modeling how information is flowing in
the system. >> BENNET: So I think that's all the time
we have for now. We'll stick around if you all have questions here. And thanks so much
for coming. >> SINGHAL: Thank you.
[END]