Placeholder Image

Subtitles section Play video

  • ASTRO TELLER: About five years ago,

  • right as Google X was being birthed,

  • I sat down with Larry Page.

  • And I was trying to work with him on how we were going

  • to talk about what Google X is.

  • And I was having a hard time getting something concise out

  • of him.

  • So I just started throwing things at him to respond to.

  • So I said, are we a research center?

  • He said, "no."

  • I'm glad to hear that.

  • Are we an incubator?

  • "Sort of, not really."

  • Are we just another business unit for Google?

  • Is that what we're going to be?

  • "No."

  • The original vision statement that Kennedy gave to the nation

  • in 1961, that we were going to put a man on the moon

  • and return him safely by the end of the decade,

  • was the original moonshot proposal,

  • at least in the moonshot sense.

  • So I was delighted when I-- after 10 of these things,

  • I threw out to Larry, are we taking moonshots?

  • And he said, "yes, that's what we're doing."

  • That made me really happy.

  • So from that afternoon on, what I've

  • been telling the people at Google X

  • is that we're trying to build a moonshot factory.

  • What I mean by that is that we're trying to take moonshots.

  • That word is to remind us that we're

  • trying to work on things that are very hard, that aspire

  • to make the world 10 times better in some way

  • than it currently is, not 10% better,

  • to remind us about the risks that we're taking

  • and the long-term nature of the work that we have ahead of us

  • when we try to do these things.

  • The word factory is meant to remind us

  • that even though we are doing these risky long term things,

  • that we want to pursue doing them with an eye

  • to actually having the impact that we aspire to,

  • that we're building products and services for the real world.

  • Fast forward five years, I'm tickled, I confess,

  • to see that the word moonshot has

  • made its way, fairly heavily now, into the popular lexicon.

  • I understand though-- I haven't seen the show myself--

  • that the TV show, "Silicon Valley," the Google in it

  • is called Hooli.

  • And they've now started their own Google X-like organization,

  • which they call XYZ, instead.

  • And it's taking moonshots also.

  • And I've personally been upgraded

  • from captain of moonshots to head daydreamer in the TV show,

  • apparently.

  • The fact that it's out there is important.

  • And part of the reason that I think that it's important

  • is that there's this bizarre-- it's understandable,

  • but it's this frustrating game of "Not It"

  • that we all play with ourselves.

  • So the small companies say, I can't take moonshots.

  • That's for big companies to do because it costs a lot of money

  • to take moonshots.

  • The big companies say, well, we aren't

  • going to take moonshots because that

  • means taking a lot of risk.

  • That's not really our game.

  • That's what the small companies should do because they

  • have nothing to lose.

  • The governments say, well, you know like 50 years ago,

  • we were taking moonshots.

  • But that's not really our thing anymore.

  • We have to work on popular, immediate problems.

  • We don't have any money.

  • Like, that just can't be us, sorry.

  • Academics love talking about moonshots.

  • They like writing the papers.

  • They actually produce some of the underlying science that,

  • later, can turn into a moonshot, but they're not

  • the system builders who are going to build

  • the moonshot themselves.

  • Everyone thinks it's someone else's job.

  • But we're not going to fix the biggest problems in the world

  • if everyone thinks it's someone else's job.

  • The truth is, we can all work on moonshots.

  • Working on things that aspire to be 10 times better, rather

  • than 10% better, is a mindset.

  • That's what it is.

  • It's got nothing to do specifically

  • with the risk, or the money, or the time frame.

  • It's a mindset about what we're working towards.

  • And counterintuitive as it is, if you

  • work on things that aspire to be that much better,

  • it not only isn't harder, sometimes

  • it's literally easier because, when

  • you aspire to make the world that much better,

  • you have to start over.

  • And when you've acknowledged to yourself as a team

  • that you're going to start over, you

  • know that what's going to happen next

  • can't be built on what people have done before.

  • You have to, in a meaningful sense,

  • come at it from a new perspective.

  • And that often, not always, but often unlocks

  • possibilities that make the impossible seem possible.

  • So this is our blueprint for how we

  • take moonshots, for what a moonshot should

  • be in our minds.

  • The first thing is that there has

  • to be a huge problem in the world that we want to resolve,

  • that we want to have go away or mitigate

  • in some meaningful way.

  • So for example, 1.2 million people

  • die every year in car accidents.

  • More than a trillion dollars is wasted every year

  • with people sitting in traffic.

  • That is a legitimately world scale

  • problem it would be awesome if we could make go away.

  • Number two, there has to be a radical proposal for how

  • to make that problem go away.

  • If it's something that people have tried over and over before

  • in the past, the idea that we or you or anyone else

  • by just trying harder, or staying up later at night

  • is not really a good outcome.

  • It's not very likely to work.

  • So cars that drive themselves all the way from point A

  • to point B-- I think that's like the poster

  • child for a radical sounding proposal

  • to make that kind of problem go away.

  • And then, the third one is there has

  • to be some reason to believe some breakthrough

  • technology, some aha from science or engineering,

  • which makes us believe that, even if it's not guaranteed

  • to work, we have a decent shot at learning through the process

  • and maybe, just maybe, getting there.

  • In the case of self-driving cars,

  • that was the DARPA Grand Challenge

  • work that originally happened and some advances

  • in smart software and smart sensors.

  • So each project that fits into this mold then

  • has to describe not just that it fits these things,

  • but that, in principle at least, it

  • could produce in the long run a Google scale business

  • or Google scale value to the company in order for us

  • to help it move forward.

  • Our goal is to have each of these things

  • create a ton of value for the world,

  • but then also create back to Google

  • a fair or equitable return on its investment

  • for taking these big risks.

  • And five years in, I'm very happy to say

  • that we've started to make real progress in this space

  • through the graduations that we've done.

  • Some of them play out in different ways.

  • So for example, the massive neural network project

  • that we originally built at Google X,

  • we graduated back into the main part of Google,

  • called knowledge, which is what you might think of as search.

  • And in that part of Google, it now

  • is servicing over 50 products and services helping

  • all these different parts of Google turn signals

  • into symbols more effectively, which is helping

  • Google to be successful.

  • And certainly, that's not all our credit

  • because they've done a lot since they left.

  • But we helped to get that going.

  • And that is a good example of the sort of thing

  • that we're shooting for.

  • In a very different way, the smart contact lens

  • work that we built, it wasn't going

  • to probably work out optimally for us,

  • not only to do the original work on that project,

  • but to take it all the way to the market ourselves.

  • So we developed a partnership with Alcon, the eye care

  • division of Novartis, and now we are headed towards the market

  • through this still very complex process

  • of trying to make contact lenses be

  • able to sense the glucose in your eye to help diabetics

  • manage their diabetes better.

  • But that is another example where

  • value can be released through this work, in this case,

  • through a partner.

  • Another of the critical operating principles

  • that we have at Google X is throwing ourselves out

  • into the world to get contact with the real world

  • as fast as possible.

  • It's not sometimes a natural thing to do,

  • but it is an absolute critical thing

  • to do, especially when you're taking on particularly

  • big, hard projects.

  • You can't possibly know at beginning the right thing

  • to do, but you can have a process where

  • you discover faster, rather than slower, that you're

  • on the wrong track.

  • That, you can do.

  • So we go through these processes for things

  • like the self-driving cars, for our flying wind turbines,

  • for Project Loon, for contact lens work that we do,

  • and for others.

  • We go through this process where we force ourselves

  • to seek out this contact.

  • And sometimes, this turns out to be us dragging our balloons up

  • to South Dakota to expose the balloons to Arctic winds.

  • Sometimes, it's asking a really specific tiny question,

  • like how long will this glucose sensor

  • the size of a piece of glitter actually

  • be able to sense glucose while sitting in this tear fluid.

  • The question is how and how fast can you

  • discover that what you're working on

  • is the wrong thing to be working on.

  • And the secret is it's discouraging

  • to hear these things.

  • We all avoid going out into the world,

  • throwing ourselves at the world to discover these things.

  • But no matter how discouraging it

  • is now, if you put more time into doing it,

  • you will unconsciously avoid even more doing it tomorrow,

  • or a week from now, or a month from now.

  • And that's why doing it as fast as you can

  • is actually the easiest time and the most efficient time

  • to discover that you're on the wrong path.

  • And that's why it is sort of central to how Google X works

  • on solving these problems.

  • I want to emphasize-- what this basically

  • means is we don't know.

  • And I would go so far so to say, nobody really

  • knows the right way to build any of these projects.

  • If you listen to the media stories,

  • you get this nice, tight arc where the entrepreneurs that

  • make it were destined to make it,

  • and the ones that didn't work were losers who were never

  • going to make it anyway.

  • And it completely misses the point,

  • that feeling in the pit of your stomach

  • where you know where you want to get to,

  • but you really don't know how to get there.

  • I have those feelings all the time.

  • Every single one of our project leads at Google X

  • has those feelings.

  • You're not alone.

  • That's just the truth of the world

  • that we have those feelings all the time.

  • All we can do is take our best guess about what

  • we should be building.

  • And then, don't wait.

  • Get quickly out into the world to discover

  • how wrong you are, which parts are salvageable,

  • and which parts you need to go back

  • to the drawing board about.

  • That's the only way to race forward.

  • So I'm going to tell you about Project

  • Wing as an example of this.

  • Project Wing had some kind of bumpy months--

  • some very bumpy months in late 2013, early 2014.

  • So the goal for Project Wing is self-flying vehicles

  • for delivery.

  • That huge problem with the world that we aspire to,

  • we believe that there are still a significant amount

  • of friction in how physical things are moved around

  • in the physical world.

  • And if you look, historically, at the Pony Express

  • and the introduction of the post office,

  • at mail order catalogs-- which are kind of weirdly

  • the precursor to ecommerce on the internet-- If you look

  • at boats, planes, trains, every time people have introduced

  • a way to more efficiently move physical things around

  • in the world, it has utterly and positively changed the world.

  • There's lots of friction left, let's remove some more of it.

  • But that's too hard to work on, to get your arms around

  • at the beginning.

  • It's just too much.

  • So we had to pick something to start with.

  • There was a process of brainstorming.

  • And the thing that we got excited about at the beginning

  • was to deliver defibrillators to people.

  • So if somebody is having a heart attack-- you've probably

  • seen like an ER show, they say, clear.

  • And they hit their chest.

  • And the guy wakes up again, hopefully.

  • So there is in this building, I guarantee you,

  • like five AEDSes, Automatic External Defibrillators.

  • I bet you not a single person in this room knows where it is.

  • And if somebody fell down on the floor right now,

  • I doubt you anyone would find it in the next 5 to 10 minutes,

  • sadly.

  • What if defibrillators could come

  • find you-- at least get to the front of the building,

  • maybe even get all the way to the person who needed it.

  • So this was our vision.

  • It's small.

  • And it's something where time matters.

  • It's about eight to nine minutes from when someone

  • has a heart attack to when they actually

  • get help, today on average.

  • If we could shave four minutes off of that time,

  • we would save 20,000 lives a year just in the US.

  • So the team got really excited about this.

  • That's a mission.

  • That's a great beachhead mission for this bigger goal.

  • So we started two things in parallel.

  • The engineers, all psyched up because they

  • were going to save all these lives,

  • started building a first version, or a vehicle,

  • which by the way-- this turns out

  • not to be the right vehicle.

  • Story for another day.

  • But that was another failure mode that we had.

  • And we're now doing a very different vehicle.

  • That was a painful moment for the engineers.

  • But while they were building the hardware, the software,

  • the sensors for these vehicles, we also immediately

  • had a real world team that went out to ask,

  • can we kill this project right now, or at least

  • that direction for the project, by finding out that there was

  • something wrong with the plan.

  • We went out and we talked to people,

  • actually had them, on dummies, use

  • these automatic external defibrillators.

  • We went and we talked to the emergency response people

  • in communities.

  • We literally sat and watched them take phone calls

  • and talk to the person who ran the 911 center in various towns

  • and said, well, could we plug into what you're doing.

  • And we quickly found out the whole plan

  • wasn't going to work.

  • That's good though.

  • I promise you, this was enormously depressing

  • for the team, but it was only going to be depressing for them

  • in a much bigger way if we'd let them

  • go for another year and a half til they were all done,

  • and then, discovered that it was the wrong thing to do.

  • That would have been much more expensive, much more painful.

  • It turned out that it wasn't the right thing

  • to do because number one, the defibrillators

  • are quite hard to use.

  • In principle, they're not, but most people

  • don't apply them really well.

  • So if you get someone something they

  • don't know how to use really quickly,

  • you actually haven't made the world

  • as better as you were hoping.

  • And the other one is that 911 is just not set up

  • to let us start flying to where the call came

  • from the moment they get the call.

  • In fact, you would be surprised how much they don't even

  • understand where the call comes from.

  • It's a somewhat antiquated system.

  • So it just wasn't a good plan, and this was really depressing

  • for the team.

  • The team took a big step back.

  • Initially, we were trying to figure out what

  • a better beachhead would be.

  • And we came to what I think was the right answer.

  • This is about 9, 10 months ago where we said,

  • you know what, we're not going to pick a beachhead.

  • Instead, let's just throw ourselves into the world.

  • And we went to Australia and we started doing deliveries.

  • And we discovered things in that process.

  • We discovered a ton of things in that process.

  • But among other things, we discovered

  • things people wanted delivered that we hadn't even imagined.

  • Cattle ranchers were saying to us,

  • do you know what would be awesome is vaccines.

  • We're out in the field.

  • The vaccines have to be stored in a refrigerator,

  • so we never have them with us.

  • They have a really poor shelf life,

  • so we often don't even have them in our barns.

  • But you want it like right there, that's when you need it.

  • If something could just come in the middle of the field

  • and hand me the vaccine, that would be amazing for the cows

  • that they're taking care of.

  • We would never have figured that out sitting

  • in a conference room on a whiteboard, I promise you.

  • So Project Loon.

  • The goal for Project Loon is to beam internet

  • down to the four billion people in the world who

  • do not currently have a good connection

  • to the digital world.

  • Four billion people.

  • It's sort of hard to imagine something more worth working on

  • than that.

  • And most of these people, surprisingly by the way,

  • already have the device.

  • More than half of them already have a device.

  • It's actually the absence of the connection to the digital world

  • which is stymieing their ability to participate in modern life

  • and to get access to better health outcomes, literacy,

  • the ability to vote, lots of other things.

  • So when we started because the idea was

  • to do this from stratospherically

  • based balloons-- so this is 60,000 to 90,000 feet,

  • much up above where the weather is and up above were

  • airplanes fly.

  • We were going to put thousands of these balloons, which

  • you can think of like flying cell towers.

  • And they would talk to each other.

  • And they would beam internet down

  • to cell phones, or laptops, or whatever on the ground.

  • We knew we had a lot to learn because we had not

  • spent-- truthfully, almost nobody

  • has spent any meaningful time hovering up in that area.

  • But we misestimated how much we had to learn.

  • So one of the things that was a basic part of our plan--

  • it's called the stratosphere because the winds go

  • at different speeds and different directions

  • quite close to each other.

  • It is stratified.

  • And this is a big deal for us, a good deal for us,

  • because we want the balloons to go up and down just

  • a little bit because it doesn't take much energy to cause

  • them to go up and down.

  • And then they can catch winds going in different directions

  • and at different speeds.

  • And so we can teach the balloons.

  • We have taught the balloons to sail on those winds.

  • What we did not fully anticipate,

  • once we got up there, was, if you have winds going like this

  • and winds going like that, and your balloon is

  • the size of a house, when it crosses that boundary,

  • it gets whipped around.

  • And the violence of that process,

  • every time we went through these layers

  • was something we really hadn't properly

  • designed into the balloon.

  • And there was a major redesign process

  • to make sure that it was ruggedized

  • to survive those particular kinds of things.

  • Even after that, we were flying balloons

  • and our goal was to get them to stay up for 100 days.

  • Our modeling of how this could turn into a Google scale

  • business told us that, on average, these things

  • needed to last 100 days in order for this

  • to be a viable business.

  • So at first, they would go up, they

  • would explode after five hours, and come down.

  • So we could learn pretty quickly that we

  • were on the wrong track.

  • It was kind of a mess though because, when it came down,

  • you didn't necessarily knew why it blew up.

  • It just sort of pops.

  • These things are full of helium.

  • But after a while, once we solved those problems,

  • our success became an even bigger problem

  • because they would leak really slowly.

  • We would wait a week.

  • We would wait three weeks.

  • Then it would start leaking.

  • It would take several more weeks for it to come down in a way

  • where we could get it to come down

  • where we could capture it, and then go send someone to like

  • Southern Chile to pick it up.

  • This was an incredibly inefficient way

  • to find out that we weren't doing the right thing.

  • We needed to fail faster.

  • We were in the real world.

  • It was giving us these signals, but we

  • needed to accelerate the real world in some way.

  • So we created, in this particular case,

  • something called a leak squad.

  • And their goal was to create, detect, and then fix

  • the leaks that we were having.

  • And they studied a lot of other things

  • in the world where people care about very thin things that

  • don't leak.

  • And it's important that they not leak.

  • Doritos bags, sausage casings, condoms.

  • There's a lot of work that's going on in this area.

  • All three of those are important not to leak

  • for somewhat different reasons.

  • So imagine something the size of a house rubbing soap

  • over it, and then, filling it so full of helium

  • that somewhere the leak would begin.

  • And then looking around it for little bubbles.

  • Or having a wand, about this big,

  • which is a helium detector.

  • And kind of like a metal detector,

  • it would go eee if it could smell the helium.

  • And then waving this thing around this thing

  • the size of a house while you've got

  • it overpressurized with helium, trying

  • to get it to leak so we could find the leak.

  • Then we could redesign how the balloon was structured.

  • Or taking it, as I mentioned before, to South Dakota.

  • There's like this weird polar vortex where Santa Claus

  • weather had come down.

  • It was like minus 40, minus 50 degrees below zero

  • for a couple weeks there.

  • And that was good because that's actually

  • the temperature it is most of the time up

  • in the stratosphere.

  • So most of the stuff did not work.

  • But we did find several really important things.

  • One of my favorite is that in the manufacturing process,

  • the balloons are so big that we have to stand on them in order

  • to make them.

  • And so someone brought up the question,

  • I wonder if them walking on the balloons

  • is what's causing some of the weeks.

  • And so we did a test.

  • Everything's an experiment.

  • Failing is fine as long as you set everything up

  • as an experiment with a hypothesis.

  • So the hypothesis here was are their feet making little holes?

  • So the control group was them standing on the balloons

  • the way they were, which is wearing their normal socks.

  • And then we got some really fluffy socks.

  • But we need to control the experiment.

  • So we actually had a line dance, like a can can line dance.

  • And we had all these manufacturing guys

  • on the balloon doing the line dance in their normal socks.

  • And then we got another balloon.

  • And we had them all put on their big fluffy socks

  • and do the same line dance.

  • And in fact, it turned out that the balloon

  • with the fluffy socks had fewer leaks in it.

  • It is literally at that level that we've

  • had to push the real world towards us

  • so we can fail quickly, so we can discover, so we can fix it.

  • I really believe Loon has made unbelievable progress.

  • It hasn't even been four years yet,

  • and they've moved from this crazy sounding science project

  • to a viable venture.

  • Our balloons now stay up, by the way, for more than six months

  • at a time.

  • And we can sail them around the world

  • to within 500 yards of where we want them to go, routinely.

  • That they've made that progress is

  • because they've been so hungry to fail,

  • because they've been so hungry to get into the real world

  • and to try things.

  • Maybe more than any other project that we have,

  • the self-driving cars have to get into the real world

  • in order to do their learning.

  • It's just the way it's wired.

  • And of course, when we started, we

  • couldn't possibly know the 10,000 things

  • that we would need to do right in order for a car

  • to drive itself really well because pretty good most

  • of the time is not good enough when you

  • get into a self-driving car.

  • But we couldn't know what that list was.

  • We weren't going to sit on some white board

  • and write down the 10,000 weird situations that would happen.

  • So you get out into the real world.

  • And the truth is that the making of that list of those 10,000

  • things is half of all of what's hard about doing something

  • like this, like making a self-driving car is figuring

  • out what the real world wants to tell you.

  • So 2 and a 1/2 years ago-- this was in the fall of 2012,

  • we were done.

  • We were just done, pretty much.

  • We had this great commute helper.

  • We had it on highways.

  • It was executing beautifully.

  • We were so done that we gave a bunch of our Lexus vehicles,

  • like you can see in this picture,

  • to a bunch of Googlers who did not work at Google X.

  • And we said take this home with you.

  • Use it for your commute.

  • You just drive to the freeway, you push a button,

  • it'll take care of the rest.

  • Driving on the freeway is actually not that hard.

  • You pretty much stay in your lane.

  • You change lanes occasionally.

  • Don't hit the guy in front of you.

  • Occasionally, some bad driver makes

  • things a little interesting.

  • But basically, we had it covered.

  • So we gave it to these people.

  • And we made them swear that they were

  • going to pay 100% attention because when we have our safety

  • drivers in the car, they're like hovering like this

  • over the steering wheel.

  • The car drives for eight hours at a time,

  • and they never touch the steering wheel,

  • but they just have to be like this the whole time,

  • so we were telling these Googlers,

  • you have to do the same thing.

  • We're going to put cameras in the car.

  • This is for your commute, but you got to pay attention.

  • Yeah, oh yeah, totally.

  • We're going to pay so much attention you can't believe it.

  • The cars, thankfully, performed flawlessly.

  • The people did not.

  • People don't even pay attention to driving

  • when they're driving.

  • They're like makeup, and the texting, and the burrito.

  • It's horrible.

  • And that's when they're actually supposed to be driving.

  • So imagine what they do when they think the car's mostly

  • got it covered, and like once in a blue moon,

  • I'm going to need to take over.

  • It was not pretty.

  • It was sufficiently not pretty that we stopped doing it.

  • And we said, OK, this is not going to work.

  • Humans cannot be a backup system for the computer.

  • Our success was a failure when you factored in human nature.

  • And so we knew that if there was going to be a self-driving car,

  • it was going to have to be something that

  • could go all the way from point A

  • to point B by itself with no help from the person.

  • This was a major existential moment for this team.

  • They thought they were done.

  • And now all of a sudden, they weren't just not done,

  • it wasn't just like a half step back.

  • They were like, is this even possible?

  • This is 2 and a 1/2 years ago.

  • The first thing was there was like this frantic brainstorming

  • about how we could force people to always pay attention

  • to the road because then we'd be like fixed, like training

  • wheels.

  • And it quickly became clear that was not the right way

  • to solve the problem.

  • So after quite a bit of emotional process and soul

  • searching, the team did what I now believe

  • is exactly the right thing.

  • They said, OK, if humans are not a reliable backup system

  • for the computer, let's build a car that

  • doesn't have a steering wheel.

  • Then, we won't be tempted to use that as a crutch

  • when we hit these parts of the process

  • where we don't know how to handle.

  • We won't just pump that and be like, oh, we'll

  • people handle that one.

  • So we did.

  • We've been doing that for 2 and a 1/2 years.

  • And we just announced recently that these cars

  • will start being on city streets this summer.

  • Super exciting.

  • [APPLAUSE]

  • At first, they're going to have this temporary steering

  • wheel in it.

  • And the safety driver will just be sitting there.

  • But that's the right, safe thing to do so that we can make sure

  • that these cars are doing what our bigger cars have been

  • practicing for a long time now, which is to always do

  • the right thing on these city streets, which are much more

  • complicated than being on the highway,

  • going all the way from point A to point B.

  • There's no way we could have learned this sort of thing

  • without spending a ton of time out in the real world.

  • It's also a priceless opportunity for us

  • to start being in these city streets

  • in a more widespread way.

  • I think if the Lexus vehicles that we had really,

  • eventually were to become the first fully self-driving cars

  • on the streets, it could provoke some anxiety in people.

  • And one of the things that we're hoping

  • is that this car, because it's smaller, it's cuter,

  • because it's only going to go 25 miles an hour,

  • because we've actually built a foam front end

  • and flexible windshield in here-- I hope we never

  • have to use it.

  • But we get to experiment with these things,

  • and on a bunch of different fronts,

  • make sure as we're learning that we're keeping the world super,

  • super safe.

  • So I'm excited to see this next step for that project.

  • Now in order for this to really be done-- that list of 10,000

  • things-- we've become-- the self-driving car team has

  • become part of its own problem.

  • We are so good at driving now on city streets.

  • We drive 10,000 miles a week in Mountain View.

  • And nothing happens because the car just does the right thing

  • all the time.

  • It is incredibly boring for the safety drivers.

  • It's not very helpful for the engineers

  • either because they aren't learning much if the cars just

  • like, do ta do ta do.

  • And they're like, OK, give it a new address.

  • Do ta do ta do.

  • We need more negative examples.

  • The teams at Google X are hungry for these negative examples.

  • So we've created a team on the self-driving car

  • team whose full time job is to find creative ways to create

  • negative examples under safe conditions for the cars.

  • Now one of the ways we can do this

  • is just by going to new cities.

  • And we are starting to do this.

  • So for example, coming to San Francisco,

  • there are interesting things like different whether.

  • It's more cloudy here.

  • Maybe you get a little bit more rain, probably not these days.

  • We're definitely not going to get snow.

  • But as we move from city to city,

  • we can rack up some new experiences.

  • San Francisco-- one of the interesting things

  • it'll give us is hills.

  • It's not that the cars are going to struggle up the hill.

  • They can go up the hills.

  • But it's a reasonable question.

  • How will the sensors on our cars have a hard time, or not,

  • when we're tilted like this up a steep San Francisco hill.

  • So that's a good thing for us to learn.

  • But that's still not learning fast enough.

  • So this team has been trying to find more and more ways

  • to challenge the cars.

  • And the ways in which we have to find

  • these rare, weird conditions fall into three

  • different categories Situation number one is people

  • drive illegally.

  • Situation number two is we just have

  • to react suddenly because people weren't paying attention.

  • And then there's just weird stuff.

  • The mail trucks are jumping out in front of us.

  • That's the canonic version of this.

  • And we have tried thousands of examples

  • to simulate that jumping out in front of us

  • just to try to stress our cars on our test track.

  • We throw beach balls at it.

  • We've gone to the Halloween store

  • and gotten big birds to swing by the windshield of the car.

  • We've had people hide in canvas bags

  • and pop out in the middle of the road

  • just to see what the car would do.

  • On take your children to work day, we parked the cars,

  • but had the sensors running just so the kids would play around

  • them, so we could get the computer

  • used to watching kids move so we could learn how they move.

  • And the reason that we could do the following

  • is because we did all that.

  • A few months ago, we were coming down a suburban side street.

  • The car turned a corner and came automatically to a stop.

  • There was a woman, an elderly woman

  • in an electric wheelchair, holding a broom,

  • in the middle of the road, trying to shoo a duck out

  • of the middle of the road.

  • And they were going around in little circles

  • in the middle of the road.

  • I'm sure this was shocking to the car

  • as well as well as to the safety drivers.

  • But eventually, she got the duck out of the road,

  • she moved off the street, and the car continued autonomously.

  • That it stopped autonomously and started autonomously again,

  • and there was no problem, is thanks to these field tests

  • that we're doing.

  • Getting out and getting into contact with the real world

  • is the most valuable thing you can do for your project.

  • All of our audacious yearly and quarterly goals

  • have this characteristic of what are

  • we going to build, how are we going to test it

  • in the real world, and how are those tests designed to reveal

  • as fast as possible the unanticipated design

  • flaws with what we're building.

  • I really believe that the progress

  • of Google X has made over the last five years

  • comes from our excitement and throwing ourselves

  • at the world, of pursuing these projects

  • with the maniacal focus of a startup,

  • and then the being hungry to be wrong,

  • being hungry to make mistakes, to get

  • these negative experiences and then harness them

  • to do it better the next time.

  • I know it sounds like a weird thing to be excited about,

  • but I think it's our special sauce.

  • I've always wanted Google X not just to take its own moonshots,

  • but to encourage other people to think this way,

  • to take on some of these problems, too.

  • So even if you're not working on a self-driving car,

  • I hope you can take something away from this approach

  • and set yourself up for creative, productive, mistake

  • filled contact with the real world.

  • Thank you.

  • [APPLAUSE]

  • I'm happy to take a few questions if you like.

  • There are microphones here.

  • AUDIENCE: Is that working?

  • ASTRO TELLER: Yes.

  • AUDIENCE: There we go.

  • So looking at a product at a typically development

  • IT organization and balancing innovation

  • from product management.

  • So from a company who's typically

  • focused on building the next product,

  • pounding the sales team lines, building, building, building,

  • do you have any suggestions or recommendations of how

  • to actually put engineering culture first

  • on top of that innovation.

  • Is it just a matter of saying, OK, let's put you guys aside.

  • Let's insulate you.

  • Is there an insulation process in that,

  • inside of the heavily driven, product management,

  • sales driven company?

  • ASTRO TELLER: I'm not sure this is the answer you want.

  • It's a weird thing to say at a developers conference.

  • I don't think an engineering heavy approach is

  • any better than a sales heavy driven approach.

  • I think any company that is overweighted

  • on any of these things is missing the point.

  • When we look to fail in the real world,

  • we depend equally on finance, on product managers,

  • on designers and UX people, on engineers.

  • Anyone who can find a reason why we're on the wrong track

  • is going to be the super hero of the month for us.

  • And so I guess what I would say is

  • rather than trying to isolate engineering

  • from these pressures, what I would try to do

  • is evangelize to them that they're missing

  • half of their opportunities.

  • And that if they think about it as discovering

  • as fast as possible why what we're currently doing

  • isn't the right thing to do, that's the most efficient way

  • to make progress.

  • And cutting off some of the reasons why it

  • might be a bad reason.

  • If you know that it breaks the law of physics

  • and you're not allowed to say so, your company is broken

  • and it's wasting its money.

  • And surely like the sales people are like, yeah, that's true.

  • Then you can say, then, invite me to the meetings.

  • Yeah

  • AUDIENCE: I just had a question about the self-driving car.

  • I think there was a quote from Elon Musk or something.

  • He was saying sort of what you said, right?

  • Freeway speeds are easy.

  • 0 to 35 range is easy.

  • But it's that 35, to say, 55 mile an hour

  • zone where you can't simply stop, right?

  • And a lot of crazy things can happen.

  • So I guess I'm just curious to hear

  • what the testing you do in that range is

  • and what your experiences are.

  • ASTRO TELLER: Look none of it's easy, to be fair.

  • The highways are very constrained situations.

  • There are some oddities that happen,

  • but a microscopic fraction of the oddities

  • that happen on city streets.

  • So that's partly about speed.

  • Part of the issue, which is some of what

  • Elon is getting at there is that the energy of the car, m v

  • squared.

  • When you double the velocity of the car, you get four times

  • the energy, four times the braking distance,

  • it does complicate things when you're going faster.

  • Our experience has been with that going slower

  • will help from a safety perspective.

  • That's part of why we're focused on 25 miles an hour

  • and under for the moment.

  • But what I would highlight is the really hard bits

  • of this are the duck.

  • When someone walks out into the street holding a stop sign,

  • like a crossing guard, that's a legal stop sign.

  • The car has to figure that out.

  • Figuring out the 10,000 things like that.

  • When a bike is coming the wrong way

  • down a bike lane, which should dominate in the computer's way

  • of approaching that problem?

  • The fact that they should be going that way, and bikes

  • tend to be going that way, or the physical reality

  • that its momentum is this way.

  • Those 10,000 things-- my experience--

  • are the really hard things.

  • Yeah.

  • AUDIENCE: How would you characterize

  • Google X's experience when bringing Google Glass

  • into the real world?

  • ASTRO TELLER: So I've said this before.

  • So first of all, Google Glass, I think,

  • is making really good progress.

  • It's graduated from Google X to Tony Fidel's part of Google.

  • And we will be hearing more in the future about that.

  • But I think we did something right and something wrong.

  • The thing that we did right, that I would absolutely

  • do again, is getting it out into the world, and in some ways,

  • even in the way we got it out into the world.

  • The Explorer Program was exactly the right thing to do,

  • and we owe a ton to Explorers.

  • Thank you, if you're one of them.

  • But we were trying to learn also about the social issues

  • around Glass.

  • And there were ways in which by trying to figure out

  • how people would think about Glass,

  • we ended up sending signals that it wasn't a prototype,

  • that it was a finished product.

  • Things like putting it on a runway.

  • I mean, even that was not crazy as an experiment

  • to see how people would think about it and respond to it.

  • But it did leave people with the feeling

  • that it was a finished product when

  • we weren't productizing it yet.

  • We were just trying to learn.

  • And I think we left people with some confusing messages there.

  • And that I wish we had done differently.

  • AUDIENCE: Thank you.

  • AUDIENCE: How do you think about financial resourcing

  • each of these products?

  • Almost by definition, a moonshot,

  • these could have infinite, or almost infinite,

  • financial resources.

  • So how do you decide which products

  • are going to get a dollar or which one's

  • going to get a billion dollars?

  • ASTRO TELLER: I'm not sure I agree

  • that you need near infinite resources for these things.

  • That's true only if you're too shy to say

  • for long periods of time that you're

  • working on the wrong thing.

  • It's amazing how in every project when you're

  • doing something dumb, when it's going in the wrong direction,

  • I guarantee you 20% or more of the team knows it within weeks.

  • You guys are engineers.

  • You know what I'm talking about, right?

  • You know, and no one will listen to you.

  • What if you lived in a culture where people are like, really.

  • Well, let's stop then.

  • It's unbelievable how much money you can save when you do that.

  • [APPLAUSE]

  • Thanks.

  • But just to answer the rest of your question,

  • it is also the case that as long as we start on a moonshot

  • and we believe that there's a good return on investment

  • over some very long period of time, where a reward to risk

  • ratio is good for Google, then as we go through the process,

  • if you're building this moonshot for us,

  • I'm going to have a dialogue with you about how much

  • resources have I given you and how much have

  • you do derisked the project.

  • And as long as the risks are coming

  • down faster than I'm giving you money,

  • our ROI is actually getting better.

  • And as long as that's true, I'm just

  • going to keep handing you the resources.

  • And if we go the other way, where

  • I'm giving you more resources and it's not

  • getting any less risky, then probably the resources

  • shouldn't keep coming.

  • I think I have time for one or two more.

  • AUDIENCE: Specifically about the self-driving cars,

  • could you talk about the 11 accidents

  • that the cars have been involved in over the past year

  • or so, both how they happened and what you've learned

  • from them going forward.

  • ASTRO TELLER: So there are certain things

  • that I'm not even allowed to get into.

  • There's legal issues around it.

  • I don't mean at Google.

  • I mean the DMV and things like that.

  • But what we've said publicly, and I encourage everyone here,

  • if you're interested, to go look at the blog

  • that we post that we put out about this is 7 of those 11,

  • we were parked.

  • I mean not parked on the side, we were stationary.

  • So the car wasn't driving.

  • The human wasn't driving either.

  • We were just rear ended.

  • I don't know what we could do about that.

  • That's just going to keep happening.

  • We should expect that to happen at roughly the rate it's

  • been happening to us because we can't not

  • have a back end to the car.

  • The other four, the person was driving.

  • Our safety drivers are pretty miraculous.

  • And they have a much better record than me or you guys do.

  • But I'm sure we could improve that in some ways.

  • We have yet to have the car be driving itself

  • when any of these have happened.

  • So I don't have more to report to you on that.

  • But I guess that's good news.

  • But we plan to continue to do these blog posts like we've

  • been doing them to help people understand what we're learning

  • and how we can improve.

  • All right.

  • Last question.

  • AUDIENCE: I had a question about you personally.

  • ASTRO TELLER: I'll get you too.

  • AUDIENCE: You do a lot of projects.

  • You manage a lot of projects.

  • You speak a lot.

  • And like I've always seen you at your 100%.

  • So what are some things that we can

  • do that you do to switch contexts between projects

  • while remaining efficient.

  • ASTRO TELLER: I'm not a maker.

  • I mean, I used to be a maker.

  • I have a PhD in computer science.

  • But to be fair, I'm good at context switching.

  • Makers, the people who are actually building the stuff,

  • do not context switch at the speed that I context switch.

  • So there's obviously a much longer thing.

  • Maybe I'll do a op-ed on it or something.

  • I think it's an interesting question.

  • But my short answer is hold yourself to the energy level

  • that you aspire to, but don't necessarily hold yourself

  • to the context switching level that I

  • do because that's not always the way to be the most productive.

  • AUDIENCE: I'm an engineer at Salesforce.

  • So I was wondering, I always had this attitude

  • where I go to the different departments.

  • I go to the strange people in the different departments

  • to get some code, to just tell me

  • how they did some things so I'm not writing this from scratch.

  • And I'm talking to the people from the boss,

  • like what they think.

  • I have these strange things.

  • And it's very beneficial.

  • I found this progressing faster.

  • I'm just curious from the people around me,

  • I didn't see a lot of such people.

  • And do you think from your experience, it comes naturally

  • or like you can teach or motivate someone

  • to do this because a lot of people are just shy.

  • ASTRO TELLER: I understand the question.

  • I believe everyone here is a moonshot thinker,

  • in your hearts.

  • I promise you, I would not be giving this talk this way

  • if I did not believe that.

  • I believe that most of us are not

  • in a context where we can be as open minded, as honest,

  • as dispassionate when appropriate,

  • as authentic as we want to be, as our natural selves would be.

  • And I think you need to ask your context for that opportunity.

  • And if it's really not going to give it to you, be humble,

  • try a few times.

  • And if it doesn't work, go find a new context.

  • You all deserve to be able to let the best

  • part of your selves out.

  • Thank you very much.

  • [APPLAUSE]

ASTRO TELLER: About five years ago,

Subtitles and vocabulary

Click the word to look it up Click the word to find further inforamtion about it