Breakfast with Google's Search Team




Uploaded by GoogleVideos on 20.08.2012

Transcript:
>>Amit Singhal: Since you are having breakfast with the search time let's talk about search.
>>audience member: Woo!
>>Amit Singhal: So, when I talk to people about search, the question they ask me most
often is, "Where the search going? What's the future of search?" And everyone who asks
me that question actually knows the answer deep inside their heart. They actually have
dreamt the search engine they want in the future they just want me to say, "Yes, we
will build it." They just want to believe. Now, let me give you some more contexts for
the answer. As a little boy growing up in India, I watched endless reruns of Star Trek
on our newly acquired black and white TV. And like every child of my age, in that time,
I, too, dreamt of being Captain Kirk, flying from galaxy to galaxy on Starship Enterprise
talking to the Starship computer asking for anything I needed to know. Yes, the destiny
of search is to become that Star Trek computer, that perfect, loyal assistant who's there
by my side whenever I need it. Now, to get there we will need to solve numerous technical
challenges, numerous very, very hard scientific problems and that's what I want to talk about
today and show you some of these technical problems that we are solving inside of our
search lab. And the very first problem that you encounter when you want to build that
future, futuristic search is, indeed, to get all the knowledge of humanity online so that
you can build upon it. Now, thanks to the web and the web community we, together, have
made a huge stride in that direction. The web today is the biggest repository of human
knowledge. As of last count, we have seen over 30 trillion URLs on the web and on an
average day we crawl over 20 billion of those. And we, humanity, are a curious kind. We at
Google are now serving over 100 billion searches every month. The web is indeed the biggest
repository of human knowledge and human beings need information. Now, having all of human
knowledge online is just the first step towards building that search of the future. The very
next impediment, scientific challenge that we face, is that we need to understand what's
in that knowledge. Our knowledge graph understands this word, like you and I do. That San Francisco
is a place and Embarcadero is a neighborhood in San Francisco and so on and so forth. It
understands things not just as sequence of letters or not just as strings that we call
in computer science, but it understands this word as real world things. Now getting all
of humanity's knowledge online and then understanding it like human beings do, is just not enough
because to build this search of future that I dream of, we will need to make search truly
universal. And a truly universal search will have information available on the web, all
the information that humanity has put on the web and information that's your information
so that you can actually do things that are not possible today. These add up rather quickly
and get us closer to the search of the future that I dream of.
>>Shashi Thakur: So Amit painted this really inspiring picture of a world which is like
the Star Trek computer. Where you can ask the computer a question, it understands your
question and it understands you and gives the best answer. So let's step back for a
second and see what it would take to realize Amit's vision. So here's a demonstration of
an advance we have done in this space. Starting tomorrow, you should be able to type in Rio,
or any other query for that matter, in the search result box, we are taking it one step
further and taking the expansions of Rio and mapped those to real world objects. And not
only are these the popular expansions, these are the popular real world objects. But start
with Rio, the city, the movie or the casino, right here on the auto complete box, annotated
with the interpretation. Like, Rio, the first one, is annotated with Rio the 2011 film.
And I can do exactly what I did in the search result box, I can highlight the option I want
and there I am, back, again, on the search page this time focused on Rio the movie without
having to pick an interpretation from the ambiguous credit, I picked the interpretation
I cared about right on the auto complete box.
>>Jack Menzel: If we understand the world we can do a better job of getting you, when
you ask a question, just that answer to a question. And he showed how when the answer
is something kind of simple like a topic, we can use our own understanding to better
pivot all of the information that we find from all over the web into the most salient
summary of that information. With the knowledge graph we're able to predict. Since Cedar Point
is an amusement park we know that the next thing that you're probably going to ask about
is the rides. So instead of me having to go and click through an alphabetical list or
copy and paste a search of an alphabetical list, I am very quickly able to flip through
these. Imagine if when I clicked on this it did something more like this which is, we've
created this experience at the top where we have taken all of the results and we have
expanded out all of the entities that we know to be rides at Cedar Point and we know, since
I'm doing this task, you know, it's likely I'm gonna do a whole bunch of these. So, now,
with this almost magical organization of roller coasters at the top, I can very quickly click
through this knowledge carousel and I can say, "Hey, check it out." So we started with
Top Thrill Dragster; that's one of the tallest ones and I think there's only one other roller
coaster that's taller than it. And then, let's see, Millennium Force, I've heard about that
one so I can flip through and then I can be like, "Well, how am I gonna make a decision
here?" Well, I probably, you know, what better way to do it then to actually, look they've
got like, I can see how the rides actually look and, Whoa! This is crazy! This is a great
way to pick roller coasters. And you can really quickly and fluently like flip through and
we can, I can complete this task for Kevin in a fraction of the amount of time it would
have taken before. Um, so that's one example of the kind of, of the kind of session, the
kind of complex task that we can make much easier with the knowledge graph. Let me show
you guys, now, how this works on tablet. We can pivot the results around the entities
that you were asking for. Women astronomers; and we can flip through and pick them out.
We can do moons of Jupiter and all the moons of Jupiter, that's kinda fun. Given our understanding
of the world we can make, we can fundamentally change the way that you are, you're able,
you're interacting with these more complex tasks and really help you get the things you
wanna get done much faster.
>>Sagar Kamdar: Our index is bigger than ever. We're refreshing that index incredibly fast.
Now with these documents we're actually making the understanding of those documents and understanding
the entities behind them and once you understand the entities, you understand the relationship
between those entities and making this phenomenal experience. It's making the search box universal.
Our whole goal is that you have an answer; we should be able to get it for you from wherever
that answer can be derived. So today we're pretty excited, we're taking a step and actually
making this, we're trying to make this experience better for you, trying to make it that you
don't have to figure out where to go to find a bit of information by introducing this field
trial of Gmail. So you sign up for this field trial, it's a limited set of users, we're
doing this again because it's a baby step in a really complicated area. Amit just talked
about how large our web corpus is and how it continues to grow, Gmail is of that size
and you now need to make it private and secure. Like that is just such a hard problem and
we're really excited that the team was able to accomplish it and then beyond the infrastructure
it's such a hard product problem on how to make this a meaningful, cool experience. What
happens when we get Gmail integrated into the search experience? So, if I type Amazon,
I see Gmail results on the right hand side. So this is one thing we did, I wanna talk
a little bit about the design process that we've gone through over the past couple of
months cause you may find that interesting. So, I search for Amazon, I see the great web
results, one thing we learned as part of our doc feeding and a lot of our user studies
is put Gmail in a consistent spot with a consistent UI. So we've actually thought about a couple
of UI treatments and a couple of placements. So one place we decided on was on the right
hand side, you see your Gmail, if you do see your Gmail results one of the spots you can
see them is on the right hand side and the other thing I wanna highlight is that these
results show up as collapsed. Another bit of feedback we got from folks is, "Hey, we're
not searching like this is great, I'm getting awesome content, maybe when somebody's over
my shoulder or what not, like I don't want them seeing that content." It's now up to
you whether or not you wanna see that interesting information. So what I can do very quickly,
for me, is I click on 'show results'. And this is great for me cause the first thing
I got is an email from Amazon, actually late last night, about my order of this Radio Flyer.
When you really want your Gmail, tell us you want your Gmail. So one thing that happens
with me a lot is my brother and my friends are really sharing cool YouTube videos with
me and I love showing it to the next friend, trying to act cool that I found this link
without giving credit to my friend. So now what we do is if I do my brother's name's
Jinen, he usually shows a bunch of videos with like performance, I use Gmail as the
keyword for one search box you're doing this mini search against Gmail, against the web
corpus and you see your Gmail results right there. If I remove Gmail from my search box,
the messages immediately show up on the right hand side. So what if you actually ask Google,
"Hey, like when's my flight?" Like what if I could say "my flight" like, "Google please
tell me." And if you have it on the field trial it'll tell you, "Here's your flight.
You're flying out today." They'll tell you the terminal, the gate number and you can
actually see whether it's on time or delayed. Like just imagine that, you typed in the query
"my flights" you're now getting information that's relevant to you and you're also getting
it updated with the coolest information from the web. It's gonna be a pretty phenomenal
experience for you and we're pretty excited about it.
>>Scott Huffman: We've talked a lot today about the computer of the future and I wanna
think with you for one more minute about what that computer of the future would be like.
It would be right here in the room with us or maybe in my pocket. I'd be able to talk
to it. I would ask it whatever I want and it would give me, understand me and give me
back exactly what I wanted. And I'd be able to ask in an actual way, the same way that
I may ask one of you. So we'd love to build that computer and to build it we think there
are a few very hard problems that need to be solved. The first hard problem, uh, which
we spent a lot of time on, is the basic problem of relevance. So when I ask my question, when
I do my search, out of all the information out there we've gotta bring back the information
that is really pinpoint relevant to the question or the task that I'm trying to do. Second
hard problem that we need to solve to have this computer of the future is the problem
of speech recognition. So, sound comes out of my mouth and we gotta turn that sound into
words and phrases that we can do something with. We've been working on it for a long
time starting back from group 411 and through the years and, as you'll see today, with some
of what I show you, we feel like our speech recognition is getting faster and getting
more accurate and enabling you to speak naturally to this computer of the future. Third problem
that we need to solve is the problem of natural language understanding. Once I've turned the
sounds that I'm making into words and phrases, we've got to understand what do those words
and phrases mean? And, of course, this is also an incredibly hard problem and what makes
it hard is that words and phrases mean different things depending on the context. And then
the last bit, which we've talked a lot about today already is underpinning all this with
knowledge of what the entities in the world are. So that we know that Tom Cruise isn't
just the word Tom and the word Cruise kind of separately but that, in fact, we know what
Tom Cruise is, he's a person, he's an actor, he's connected to other things and that enables
us to give pinpoint answers. So, today I'm excited to show you and announce our latest
version of the Google search app for the Ipad and the Iphone now with voice based question
answering. [Beep]
>>Scott Huffman: What will the weather be like this weekend?
[Beep]
>>woman's voice on app: The forecast for San Francisco this weekend is 68 degrees with
fog.
>>Scott Huffman: I asked for the weather in a natural way and we understood not only what
those words were, we understood, for example, that even though I didn't say it, that I'm
here on the Embarcadero in San Francisco so that's where the weather has come from. We
understand that this weekend actually probably means Saturday and Sunday and so here we're
showing the weather from Saturday. And it looks like it's gonna be foggy here so I might
wanna get out of town and try something else. I've been meaning to go on a hike. One of
my friend's told me about this place called the Pinnacles National Monument, let's check
that out. What does Pinnacles National Monument look like?
[Beep]
>>woman's voice on app: Check out these pictures for Pinnacles National Monument.
>>Scott Huffman: Hm. It's a beautiful place. Okay. So that's a possibility. I wonder how
far it is. [Beep]
>>Scott Huffman: How far is Pinnacles National Monument from here?
[Beep]
>>woman's voice in app: To drive from your location to Pinnacles National Monument is
122.3 miles.
>>Scott Huffman: Hm, okay. I don't know if I wanna hike that bad.
[Laughter] >>Scott Huffman: That's a long way. Here's
another idea. [Beep]
>>Scott Huffman: Who are the Giants playing? [Beep]
>>woman's voice in app: The Giants beat the Cardinals 4 to 2. They are playing the Cardinals
today at 5:15 pm.
>>Scott Huffman: Okay, so maybe I can take in a Giants game. For those of you who follow
the Giants you know they're actually doing pretty well but their ace, Tim Lincecum's,
been struggling and so this was a question I had the other day.
[Beep] >>Scott Huffman: What is Tim Lincecum's salary?
[Beep]
>>woman's voice in app: Tim Lincecum's salary is US 18 million.
>>Scott Huffman: Okay, so he needs to throw a few strikes then.
[Laughter] >>Scott Huffman: Alright, so Giants is a possibility.
Here's another thing I could do. [Beep]
>>Scott Huffman: What movies are playing at the Metreon?
[Beep]
>>woman's voice in app: Movies playing at AMC Metreon 16.
>>Scott Huffman: Okay, so I got Dark Knight, I've got, oh, there's the new Spiderman movie.
Let's check that out. The movies at the Metreon, they cost like 12 bucks. I've got 2 kids,
my wife and if I take them to a movie they're also gonna wanna eat so that's gonna be like
50 bucks. Let's just see what this adds up to.
[Beep] >>Scott Huffman: What is 12 times 4 plus 50?
[Beep]
>>woman's voice in app: the answer is 98.
>>Scott Huffman: Hm, 98 bucks, okay. This is getting expensive, Alright, so maybe I
can save some money by eating cheap, let's try this.
[Beep] >>Scott Huffman: Find a good pizza place near
the Metreon. [Beep]
>>woman's voice in app: There are several listings for a good pizza place near Metreon.
>>Scott Huffman: Okay, so there's some possibilities. I wanna show you one more thing because, in
fact, I'd actually already decided to head out of town this weekend and to show you this
one I'm gonna actually bring up the demo version of the app; it'll work like this.
[Beep] >>Scott Huffman: When is my flight?
[Beep]
>>woman's voice in app: Your flight is on time. It departs at 11:25am today.
>>Scott Huffman: So there it is, bringing it all together. So you can tell we're very
excited about this product. We think that it brings this computer of the future just
a little bit closer.
>>Amit Singhal: Wow, you can just see that, with what Scott just showed you, we are so
much more closer to that dream computer of mine from my childhood. I think this team,
this ace search team, is going to make my dream come true. I have no questions about
that.