Placeholder Image

Subtitles section Play video

  • MALE SPEAKER: So today we're here to see Jerry Kaplan.

  • He's co-founded four startups, two of which

  • have gone public-- serial entrepreneur, widely

  • respected as a technical innovator,

  • and a bestselling author.

  • In the interest of time though, I'm

  • going to abridge his long list of many accomplishments

  • and talk about a few things that I think will especially

  • interest you before he talks about things that especially

  • interest him.

  • So one especially interesting thing that he's done-- he

  • was the co-founder, alongside his friends Kevin Doren

  • and Robert Carr, of the GO Corporation in 1987.

  • They were pioneers in the work of tablet and stylus-based

  • computing, the precursors to the Apple Newton, the Palm Pilot,

  • and later smartphones and tablets of today.

  • If you chronicled, actually, his time there,

  • they have a very interesting book called "Startup:

  • The Silicon Valley Adventure."

  • Some of you may have heard of this.

  • A fun little fact.

  • Omid Kordestani-- some of you may know him

  • as our former chief business officer, started

  • at Google in 1999-- got his start, actually, at Go Corp.

  • Jerry may talk about that.

  • I don't know.

  • It's possible.

  • Here he is today again to talk about artificial intelligence

  • and the changing world of work automation.

  • So we are here for "Humans Need Not Apply."

  • Give a warm welcome to Jerry Kaplan, everyone.

  • [APPLAUSE]

  • JERRY KAPLAN: Thanks, how's the mic?

  • Oh, that's good.

  • All right, well, a mentor of mine used to say never

  • give a talk the first time.

  • I want you to know I've put together

  • a special talk for you guys.

  • This is the first I'm giving it.

  • We'll see what happens.

  • I should leave some time.

  • I have lots of weird anecdotes about Google

  • that I will be happy to tell when I'm not on the camera,

  • as long as I'm not being recorded.

  • OK, so now for something completely different, as

  • they used to say on "Monty Python."

  • The common wisdom about artificial intelligence

  • is that we're building increasingly intelligent

  • machines that are ultimately going

  • to surpass human capabilities and steal our jobs

  • and maybe even escape human control

  • and take over the world.

  • So I'm going to present the case today

  • that that narrative is both misguided and

  • counterproductive-- that a more appropriate way to frame

  • this, which is really better supported

  • by actual historical and current events,

  • is that AI is simply a natural extension

  • of long-standing efforts to automate tasks that date back

  • at least to the start of the Industrial Revolution.

  • And then I want to talk about the consequences

  • if you think about it in that particular way.

  • But let me ask about the audience-- how many of you

  • are engineers?

  • OK, how many of you are not engineers?

  • Two.

  • How many people haven't raised their hand yet?

  • Nobody.

  • OK.

  • And that's called closure, right?

  • OK, and how many of you are doing anything

  • even vaguely related to AI?

  • Oh, not that many, OK.

  • Cool.

  • At least you won't admit it by the time I'm done with my talk,

  • I think.

  • OK, so let me start with a little bit of a history lesson.

  • I'm teaching Impact of Artificial Intelligence

  • at Stanford.

  • And much to my shock, the students

  • who studied artificial intelligence

  • don't know much about its history.

  • So here's a kind of irreverent view.

  • I'm going to start with an unorthodox history of AI.

  • Now, here's a news flash for you.

  • Science does not proceed scientifically.

  • So it's like the making of legislation and sausage.

  • Perhaps this is better done outside of the public view.

  • More than you might want to believe,

  • progress is often due to the clash of egos

  • and ideas and institutions.

  • You guys work in an institution.

  • I'm sure you see that occasionally.

  • And artificial intelligence is no exception,

  • so let me start right at the beginning.

  • Dartmouth College, 1956.

  • A group of scientists-- they got together

  • for an extended working session.

  • How many of you who John McCarthy is?

  • Oh, man, OK.

  • He's a mathematician who was then employed at Dartmouth.

  • Now, he hosted this meeting along with-- raise your hand

  • if you know these guys-- Marvin Minsky.

  • Oh, more than John, OK.

  • He was then at Harvard.

  • Claude Shannon?

  • That's good.

  • You guys should know who Shannon is.

  • He was at Bell Laboratories.

  • And Nathaniel Rochester?

  • Probably no hands.

  • One hand.

  • Are you his son?

  • Sorry?

  • AUDIENCE: [INAUDIBLE].

  • JERRY KAPLAN: OK, there you go.

  • He was at IBM.

  • Now, here's what these guys had to say,

  • or John McCarthy had to say.

  • He called his proposal "A Proposal

  • for the Dartmouth Summer Research Project

  • on Artificial Intelligence."

  • Now, this was the first known use of the term artificial

  • intelligence.

  • But what's not commonly known is why did John McCarthy choose

  • that particular name?

  • He explained this later-- much later,

  • actually-- his motivation.

  • He said, "As for myself, one of the reasons

  • for inventing the term artificial intelligence

  • was to escape the association with cybernetics.

  • Its concentration on analog feedback seemed misguided,

  • and I wished to avoid having either

  • to accept Norbert Wiener as a guru

  • or having to argue with him."

  • Now, Norbert Wiener, as you may know,

  • was a highly respected-- Norbert Wiener?

  • Anybody?

  • Oh, my god.

  • OK.

  • Cybernetics.

  • Cybernetics.

  • Good, you've heard the term at least.

  • He was a highly respected senior mathematician

  • and a philosopher at MIT.

  • Now, while he was that, McCarthy, this guy,

  • was just a junior professor at Dartmouth.

  • So he didn't want to have to go up against the powers that be.

  • So to understand the original intention of the founding

  • fathers of AI, it's worth reading

  • some of the actual text of this conference proposal.

  • I think it's on the screen.

  • "The study is to proceed on the basis of the conjecture

  • that every aspect of learning or any other feature

  • of intelligence can in principle be so precisely described

  • that a machine can be made to simulate it.

  • An attempt will be made to find out

  • how to make machines use language,

  • form abstractions and concepts, solve

  • kinds of problems now reserved for humans,

  • and improve themselves."

  • It's 1950-- what was it, 1956?

  • "We think that a significant advance

  • could be made in one or more these problems if a carefully

  • selected group of scientists work on it together

  • for a summer."

  • Now, that's a pretty dubious agenda for a summer break.

  • Now, many of the Dartmouth conference participants

  • had their own view about how to best approach

  • artificial intelligence.

  • But John McCarthy's specialty was mathematical logic.

  • In particular, he believed that logical inference

  • was the key to, using his words, simulated intelligence.

  • That's what he thought AI was.

  • Now, his approach, skipping ahead quite a ways,

  • but his approach eventually became

  • known as what's called the physical symbol systems

  • hypothesis.

  • Anybody here have heard of that?

  • One.

  • Good man, OK, you can take over for the rest of the talk.

  • Now, that was the dominant paradigm

  • in the field of artificial intelligence for the first 30

  • years or so after the Dartmouth Conference.

  • Now, here's John McCarthy.

  • I'm old enough to have known John McCarthy when

  • I was a postdoc at Stanford, where he founded the Stanford

  • Artificial Intelligence Lab.

  • Now, John was definitely a brilliant scientist.

  • He invented the programming language Lisp.

  • Good.

  • And he invented the concept of time sharing.

  • Not too many people know that.

  • But he definitely had the mad professor thing going.

  • Let's see if this works.

  • Almost.

  • I'm using somebody else's computer.

  • You know, he had the wild eyes and the hair.

  • The guy on the right, as you may know,

  • is Professor Emmett Brown, who invented the-- what is it?

  • The flux capacitor time machine.

  • How many people know the flux capacitor?

  • OK, good.

  • I'm just checking to make sure this talk works.

  • But I'm confident that John McCarthy, having met him,

  • never really expected that his clever name emerging

  • field is going to turn out to be one

  • of the great accidental marketing coups of all time.

  • So it's not only inspired generations of researchers,

  • including myself, but it spawned a virtual industry

  • of science fiction and Hollywood blockbusters and media

  • attention and pontificating pundits, also including myself.

  • How do you name the field something less arousing,

  • like logical programming, or symbolic systems?

  • I doubt very many of us would have ever heard

  • of the field today.

  • The field would have just motored

  • along automating various tasks while we marvelled

  • at the cleverness not of what the creations were,

  • but of the engineers.

  • I'm getting a little bit ahead of my story.

  • In any case, McCarthy's hypothesis

  • that logic was the basis of human intelligence is, at best,

  • questionable.

  • Today, in fact, most AI researchers

  • have abandoned this approach and believe

  • it was just plain wrong.

  • The symbolic system approach has been almost entirely abandoned

  • in favor of generally what's now referred

  • to as machine learning.

  • How many people here are doing machine learning?

  • Good.

  • OK, or you certainly know about it.

  • But rejecting that old approach is throwing the baby out

  • with the bathwater.

  • Some truly important advances in computing

  • came out of symbolic systems, including

  • things like heuristic search algorithms, logical problem

  • solvers, game players, reasoning systems.

  • These were all the old approach.

  • And many of the results of all that work

  • are in wide practical use today.

  • For example, formulating driving directions-- I got

  • lost coming here.

  • I didn't know the difference between the express lane

  • and the regular lane.

  • I thought I was in the other one.

  • Take this exit.

  • No exit.

  • Laying out factories and warehouses,

  • proving that complex computer chips actually

  • meet their specifications-- this all uses early AI techniques.

  • And I'm sure that there are many more of these to come.

  • Now, did I mention machine learning?

  • It's certainly the focus of most current research,

  • and in some circles, at least where I am,

  • it's considered a serious candidate

  • for the real basis of human intelligence.

  • Now, my personal opinion is that while it's

  • a very powerful technology, and it's

  • going to have a very significant practical impact,

  • it's very unlikely to be the computational equivalent

  • of the human mind.

  • And whatever your view, you might

  • be surprised to learn a little more about what

  • are the fundamental concepts that underlie what's

  • called the connectionist or neural networking approach

  • to machine learning came from.

  • There are some other approaches, mainly in the statistical area.

  • So let's see.

  • Frank Rosenblatt, anybody heard of him?

  • Wow, OK, great.

  • I didn't until I started researching this.

  • Back in the late 1950s, John McCarthy

  • wasn't the only one interested in building

  • intelligent machines.

  • There was another highly optimistic proponent,

  • and that was Professor Frank Rosenblatt

  • at Cornell-- another competing prominent institution.

  • You've got Cornell, and you've got Dartmouth--

  • and lots of people at MIT.

  • And Rosenblatt was intrigued by some pioneering research

  • by psychologists Warren McCulloch and Walter Pitts

  • at the University of Chicago.

  • And McCulloch and Pitts had observed

  • that a network of brain neurons could

  • be modeled by, of all things, logical expressions.

  • So Rosenblatt got the bright idea

  • to implement their ideas in a computer program, which

  • he rebranded as a perceptron.

  • Anybody heard of perceptrons?

  • Oh, good.

  • Cool.

  • He built an early version of what today,

  • we would call a simple neural network.

  • This is the geekiest looking guy.

  • He looks like he's 12 years old.

  • That's him.

  • That's his actual photo cell array right there.

  • Now, he wasn't about to be outdone by McCarthy and Minsky,

  • so Rosenblatt heavily promoted his work in the popular press.

  • For instance, he was quoted in "The New York Times"

  • in 1958 saying, the machine that he is going to build

  • would be the first device to think as the human brain.

  • In principle, it would be possible to build brains

  • that could reproduce themselves on an assembly line,

  • and which would be conscious of their existence.

  • This is 1958.

  • The article went on to say that the embryo

  • of an electronic computer-- the embryo

  • of an electronic computer today that

  • will be able to walk, talk, see, write, reproduce itself, and be

  • conscious of its existence.

  • And here's what I love.

  • "It is expected to be finished in about a year

  • at a cost of about $100,000."

  • So much for the journalistic accuracy of "The New York

  • Times."

  • By the way, I'm usually debating John Markoff.

  • He's the science writer there.

  • We love to beat each other.

  • I wish he was here.

  • He'd go crazy.

  • Now, that might seem a little bit optimistic,

  • given that Rosenblatt's demonstration included

  • only 400 photo cells connected to 1,000 perceptrons, which

  • after 50 trials, was able to tell whether a card had

  • a square marked on the right side or on the left side.

  • That's what he could do.

  • Now, on a more positive note, and this

  • is also pretty remarkable, I can't help

  • but notice that many of his wilder

  • prophecies in the article have actually now become reality.

  • He went on to say-- listen to this closely-- remember, 1958.

  • "Later, perceptrons will be able to recognize people, call out

  • their names, instantly translate speech

  • from one language to speech or writing in another language."

  • He was right.

  • It only took 50 years longer than he predicted.

  • OK, now, Rosenblatt's work was well known to at least some

  • of the participants at that Dartmouth Conference.

  • In particular, he attended the Bronx High School of Science--

  • anybody here go there?

  • Not one, wrong coast-- with Marvin Minsky.

  • They were one year apart.

  • So they later wound up going to these different forums

  • and debating each other, promoting

  • their respectively favored approaches

  • to artificial intelligence.

  • Until in 1969, Minsky, who's now at MIT-- remember,

  • one guy's at Cornell, the other guy's

  • at MIT-- along with a colleague of Marvin Minsky's called

  • Seymour Papert, published a book called

  • "Perceptrons." in which he went to pains to discredit, somewhat

  • unfairly I might add, a simplified

  • version of Rosenblatt's work.

  • Now, here's the way science really works.

  • Now, Rosenblatt was unable to mount a proper defense

  • for a very simple reason.

  • Anybody guess what it was?

  • He died in a boating accident in 1971, two years later.

  • He couldn't defend himself.

  • Now, the book, however, proved highly influential,

  • effectively foreclosing funding and research

  • on perceptrons and artificial neural networks

  • in general for more than a decade.

  • So after 50 years, which is better,

  • the symbolic systems approach or the machine learning approach?

  • The plain fact is both of these approaches

  • have different strengths and weaknesses.

  • In general, symbolic reasoning is

  • more appropriate for problems that

  • require abstract reasoning.

  • And machine learning, on the other hand,

  • is better for problems that require sensory perception

  • or extracting patterns from large collections

  • of noisy data.

  • So you might ask the question, why

  • was the symbolic approach dominant

  • in the last half of the 20th century and machine learning

  • is dominant today?

  • The answer is fairly simple-- the machines.

  • They are literally a million times faster, cheaper,

  • and have a million times more memory at the same price

  • as they did back then.

  • That's a qualitative difference.

  • In the early days of AI, machines just

  • weren't powerful enough to automatically learn

  • anything of interest-- the square is on the right,

  • the square is on the left.

  • They had n only a minuscule fraction of the processing

  • speed and a vanishingly small amount of memory

  • in which to store data compared to today's computers.

  • But most importantly, there simply

  • weren't many sources of machine readable

  • data available to learn from.

  • What were you going to learn?

  • For real time learning, most communication at that time

  • was on paper.

  • And for real-time learning, the data from sensors

  • was equally primitive and only available, usually,

  • in an analog form that really resisted processing digitally.

  • So you had four trends-- improvement

  • in computing speed, memory, the transition

  • from physical to electronically stored data, and easier access

  • to large bodies of data.

  • God knows you guys know about that.

  • It's mainly due to the internet and low-cost, high-resolution

  • digital sensors.

  • I don't know how I came up with five, but I can't count.

  • These were the prime drivers-- never

  • give a talk the first time.

  • These were prime drivers in the refocusing of efforts

  • from the symbolic reasoning approach to the machine

  • learning approach.

  • OK, there's a little bit of history for you.

  • Now let me get to the main issue.

  • Can machines think?

  • So what is artificial intelligence, really?

  • After a lifetime of work in this field

  • and a great deal of reflection on this question,

  • my reluctant and disappointing answer is simple.

  • No.

  • Or at least they can't think the way people think.

  • So far, at least, there's no obvious road map from here

  • to there.

  • Machines are not people.

  • And there's simply no persuasive argument

  • that they're on the same path to becoming

  • generally intelligent, sentient beings,

  • despite what you see in the movies.

  • Now, wait a minute, you might say.

  • Jerry, can't they solve all sorts of complex reasoning

  • and perception problems?

  • Sure they can.

  • They can perform tasks that humans

  • solve using intelligence.

  • But that doesn't mean that the machines are intelligent.

  • It merely means that many tasks that we thought

  • required general intelligence are in fact subject to solution

  • by other kinds of mechanical means.

  • Now, there's an old joke in AI, which

  • is that once an AI problem is solved, it's no longer AI.

  • Anybody heard that?

  • A couple of people, good.

  • Now, personally, I don't think that's any longer a joke.

  • I'm going to look at some of the signature

  • accomplishments of artificial intelligence

  • from this different perspective.

  • Let's start with computer chess.

  • Now, for decades-- most of you guys

  • weren't around to see this, but I

  • was-- the archetypal test of the coming of age of AI

  • wasn't the Turing test.

  • It was, could a machine ever beat

  • the world's chess champion?

  • For a long time, you see, chess was

  • considered the quintessential demonstration

  • of human intelligence.

  • So surely when a computer was world chess champion,

  • AI would have arrived.

  • That's it.

  • We'd have smart machines.

  • Well, it happened in 1997 when IBM's Deep Blue

  • beat the then champion, Garry Kasparov.

  • Lots of ink was spilled in the media lamenting the arrival

  • of super-intelligent machines.

  • There was a lot of hand wringing or what this meant

  • for the future of mankind.

  • But the truth is it meant nothing other than that you

  • could do a lot of clever programming

  • and use the increases in speed of computers to play chess.

  • The techniques used have applications

  • to similar classes of problems.

  • But they hardly proved to be the harbingers of the robot

  • apocalypse.

  • So let me tell you what people said

  • after that non-event happened.

  • They said, OK, sure, computers can play chess.

  • But they'll never be able to drive a car.

  • This really was what happened.

  • That requires a broad understanding

  • of the real world-- the ability to make split-second judgments

  • in chaotic circumstances.

  • And, of course, common sense-- machines will never have that.

  • Well, as you know, this bulwark of human supremacy

  • was breached in 2004 with the DARPA Grand

  • Challenge for autonomous vehicles,

  • which are soon coming, if they're not here,

  • to a parking lot near you.

  • How many of you guys have taken a ride

  • in the Google self-driving cars?

  • What?

  • Oh, they should send one up here.

  • Have you been at least down to the Tesla dealership

  • to take a test drive?

  • I did that over the weekend.

  • The self-driving car was cool.

  • OK, now our self-driving cars do just that.

  • They drive cars.

  • They don't build houses.

  • They don't cook meals.

  • They don't make beds.

  • That's what they do.

  • So computers can play chess and drive cars.

  • But then they said-- people said,

  • but they could never play Jeopardy.

  • OK, well, that requires too much world knowledge

  • and understanding metaphors and clever wordplay.

  • Well, thanks again to be ingenious people at IBM,

  • this hurdle has also been cleared.

  • As undoubtedly you know, IBM's Watson system

  • beat Ken Jennings, the world Jeopardy champion in 2011.

  • Now, what is Watson?

  • The reality is it's a collection of facts and figures encoded

  • into cleverly organized modules that

  • can quickly and accurately answer

  • various types of common Jeopardy questions.

  • Watson's main advantage over the human contestants,

  • believe it or not, was that it could ring in

  • before they could when it estimated a high likelihood

  • that it had a correct answer.

  • I would love to go in this in more detail for you.

  • It turns out most of the Jeopardy champions

  • know the answers.

  • They're just not that fast.

  • And so the machine had numerous advantages.

  • It's a long-- it was kind of a magic show.

  • It's a wonderful accomplishment.

  • It's a really remarkable and very sophisticated

  • knowledge-based retrieval system and an inference system

  • that was honed, at least at that time,

  • to a particular problem set.

  • Now they're trying to apply it to lots of others.

  • Now, how many of you saw that, or pictures of it?

  • OK, now here's what bothers me.

  • Is this supposed to be animated?

  • OK.

  • Now, in my opinion, IBM didn't do the field of AI

  • any favors by wrapping Watson in a theatrical suite

  • of anthropomorphic features.

  • There's really no technical reason

  • to have a system say its responses in a calm, didactic

  • tone of voice.

  • Yes, Alex, the answer is such and such.

  • Much less to put up a head-like graphic of swirling lights,

  • suggesting that the machine had a mind

  • and was thinking about the problem.

  • These were incidental adornments to a tremendous technical

  • achievement.

  • Now, without a deep understanding

  • of how these systems work, and with humans

  • as the only available exemplars with which

  • to interpret the results, the temptation

  • to view them as human-like is really irresistible.

  • But they aren't those things.

  • OK, so let me give you a couple more interesting examples--

  • more contemporary, things that you're probably

  • more familiar with.

  • What about these machine learning systems?

  • Aren't they more like human intelligence?

  • Well, not really.

  • True, I could argue this for two hours here.

  • Lots of people sticking their hand up.

  • In reality, the use of the term neural networks is

  • little more than an analogy, in the same sense as saying

  • airplane design was inspired by birds.

  • It's in the same category.

  • Consider how machines and people learn.

  • You can teach a computer to recognize cats

  • by showing it a million images.

  • You guys know Andrew Ng?

  • He was at Google when he did that work.

  • You can show it a million images,

  • or you could simply point one out to a three-year-old

  • and get the same job done.

  • That's a cat.

  • Oh, that's it.

  • Now, from then on, the three year old knows what a cat is.

  • Obviously, humans and machines do not learn the same way.

  • And let me give you another interesting example.

  • Anybody here doing machine translation?

  • One Google site.

  • OK, I'm going into the lion's den in about two weeks.

  • I'm going to talk to the machine translation people,

  • among others.

  • Now, tremendous strides have been

  • made in this field in the past few years

  • mainly by applying statistical and machine learning

  • techniques to large bodies of concordant text.

  • But how do people perform this difficult task?

  • Think about how people do it.

  • They learn two or more languages,

  • along with the respective cultures and conventions.

  • Then they read some text in one language,

  • they understand what it says, and they render the meaning

  • as closely as possible in another language.

  • Now, machine translation, as successful as it is today,

  • there's no relationship to the human translation process.

  • Its success simply means there's another way

  • to approximate the same results.

  • It's mostly just concordances of text.

  • It doesn't relate to the way people solve that problem.

  • What do we learn from this?

  • It's just a way to-- we just didn't

  • think there was another solution, but there is,

  • besides having people understand that.

  • Now, let me go on to one you are all carrying around.

  • You carry around smartphones.

  • They're reminiscent of the capabilities

  • of the computer on the Star Trek Enterprise-- "Star Trek?"

  • Everybody?

  • Good, OK.

  • I started talking about "The Jetsons"

  • in my class at Stanford, nobody knew what I was talking about.

  • What's that?

  • You know, Rosie and-- you know, OK.

  • That's called getting old.

  • So this is more like-- I lost complete track,

  • and I got the whole thing right in front of me.

  • Hey, Siri, you know?

  • You can talk to your phone, and it talks back.

  • It also becomes more capable every day

  • as you download new apps and upgrade the operating system.

  • So I'm using examples of [INAUDIBLE].

  • But do you really think of your phone

  • as getting smarter in the human sense

  • when you download an app or you enable voice recognition?

  • Certainly not in the same sense that you

  • get smarter when you learn calculus

  • or when you learn philosophy.

  • It's the electronic equivalent of a Swiss Army knife.

  • It's a bunch of different information processing tools

  • that are bound together into a single unit,

  • taking advantage of some commonalities, like detailed

  • maps, and like internet access.

  • Now, you have one integrated mind,

  • while your phone has no mind at all.

  • There's no one home.

  • So I try to make the case that machines perform

  • an increasingly diverse array of tasks

  • that people perform by applying their native intelligence.

  • Now, does that mean that machines are smart?

  • Well, now things get interesting.

  • Let's talk about how you might measure supposed machine

  • intelligence.

  • I pulled that picture off the internet.

  • So I didn't make it up.

  • That's part of the point I'm trying to make.

  • We can start by looking at how we measure human intelligence.

  • Now, a common method is with IQ tests, but even for humans,

  • this is a deeply flawed concept.

  • We love to measure and rank things with numbers.

  • But let's face it, reducing human intelligence

  • to a flat, linear scale is highly questionable.

  • Little Sally did two more arithmetic problems than Johnny

  • did in time allotted, so her IQ is seven points higher

  • than his.

  • Bull.

  • But that's not to say that some people aren't smarter

  • than others-- only that simple numerical measures provide

  • an inappropriate patina of objectivity and precision.

  • As psychologists are fond of pointing out,

  • there are many different kinds of intelligence.

  • There's social and emotional, analytic, athletic, musical, et

  • cetera.

  • But what on Earth does it mean to say that Mozart and Einstein

  • have the same IQ?

  • Now, suppose we gave the same intelligence

  • tests to a machine.

  • Wow!

  • And it only took one millisecond to accurately complete

  • all of the sums that took Sally, and Johnny, and Alan.

  • It must be super-smart.

  • It also outperforms all humans on memory tests,

  • logical reasoning tests, and god knows what else.

  • Maybe it can shoot straighter, read faster,

  • and can outrun the fastest human.

  • Oh my god.

  • Robots can outperform us.

  • What are we all going to do?

  • So are the robots taking over?

  • Are the robots taking over?

  • Of course, by the logic I just gave you,

  • machines took over a long time ago whether they are smart

  • or not.

  • They move our freight.

  • They score our tests.

  • They explore the cosmos.

  • They plant and pick most of our crops.

  • They trade stocks.

  • They store and retrieve our documents, as Jacob knows,

  • in petabytes.

  • They manufacture just about everything,

  • including themselves.

  • And sometimes they do it with human help,

  • and sometimes without human intervention.

  • And yet, they aren't taking over our businesses.

  • They aren't marrying our children.

  • They are not watching the SyFy channel when we're not around.

  • So what's wrong with the traditional picture of AI?

  • We can build machines and write programs and perform

  • tasks that previously required human intelligence

  • and attention, but there's really nothing new about that.

  • Each new technological breakthrough

  • from the invention of the plow to to the CGI

  • rendering of Rapunzel's hair is better

  • understood as an advance in automation,

  • not as a usurpation of human primacy.

  • We can program machines to solve very complex problems,

  • and they may operate with increasing independence.

  • But as a friend of mine once observed,

  • a vehicle will really be autonomous

  • when you instruct it to take you to the office,

  • and it decides to go to the beach instead.

  • My point is simple.

  • Lots of problems we think require human intelligence

  • to solve actually don't.

  • There are lots of other ways to solve them,

  • and that's what the machines are doing.

  • Calculating used to be the province

  • of highly trained specialists.

  • Did you guys know that?

  • You used to go see somebody when you

  • wanted to do some calculation.

  • Now all it takes is the $0.99 calculator.

  • Making money in the stock market used

  • to be the province of experts.

  • Now the majority of trading is initiated by computers.

  • It's the same for driving directions, picking and packing

  • orders in warehouses, and designing more efficient wings

  • for airplanes.

  • But you don't have to worry about the robots taking over.

  • Robots don't have feelings, except in the movies.

  • Here's a news flash for you.

  • They aren't male or female.

  • As I like to say to my Stanford students,

  • what does it mean for a robot to be gay?

  • So robots don't have independent goals and desires.

  • A robot that's designed to wash and fold laundry

  • isn't going to wake up one day and say, oh my god, what a fool

  • I've been.

  • I really want to play the great concert halls of Europe.

  • So just as we can teach bears to ride bikes,

  • and we can teach chimps to use sign language,

  • we could build machines to perform

  • tasks the way people do, and even

  • to simulate human emotions.

  • We can make them say ouch when you pinch them or wag

  • their tails when you pet them.

  • But there's simply no compelling reason

  • to believe this bears any meaningful relationship

  • to human behavior or experience.

  • Machines aren't people, even if we

  • build them to talk and walk and chew gum the way that we do.

  • OK, now I've given you a new way to think

  • about artificial intelligence.

  • Let's talk about the implications

  • of this new perspective.

  • I'm going to try to run through this

  • pretty quickly because I was warned people like

  • to ask questions.

  • Now, it's certainly true that AI is

  • going to have a serious impact on labor

  • markets and employment.

  • But perhaps not in the way that people expect.

  • If you think of machines as becoming even more intelligent

  • and threatening our livelihoods, the obvious solution

  • is to prevent them from getting smarter, and to lock our doors

  • and arm ourselves with Tasers against these robots that

  • are coming to take our jobs.

  • Well, the robots are coming, but not exactly for our jobs.

  • Machines and computers don't perform jobs.

  • They automate tasks.

  • Now, except in extreme cases, you don't roll in a robot

  • and show an employee to the door.

  • Instead, the new technologies hollow out and change the jobs

  • that people perform.

  • Even experts spend most of their time doing

  • mundane, repetitive tasks.

  • They review lab tests.

  • They draft simple contracts.

  • They write straightforward press releases.

  • They fill out paperwork and forms.

  • On the blue collar side, lots of workers

  • lay bricks, paint houses, mow lawns, drive cars, load trucks,

  • pack boxes, and take blood samples.

  • They fight fires, deliver direct traffic, et cetera.

  • And many of these intellectual and physical tasks

  • require straightforward logic or simple hand-eye coordination.

  • Now, the new technologies, mainly driven

  • by artificial intelligence, are poised to automate these tasks,

  • not to replace the jobs.

  • Now, if your job involves a narrow, well-defined set

  • of duties, and many do, then indeed,

  • your employment is at risk.

  • If you have a broader set of responsibilities,

  • or if your job requires a human touch such as expressing

  • sympathy or providing companionship,

  • I don't think you have too much to worry about.

  • Now, just check out this comparison

  • of the job duties between licensed practical nurses

  • and bricklayers.

  • Whose job do you think is most at risk from automation?

  • By the way, this list is hilarious.

  • "Monitoring fluid and food intake and output."

  • I was like, OK, I didn't know they measure output.

  • "Providing emotional support."

  • What you guys are working on-- (ROBOT VOICE)

  • I am so sorry about your problem.

  • I mean, come on.

  • Most jobs, as opposed to tasks, involve

  • a mix of general capabilities and specific skills.

  • And as machines perform the more routine tasks,

  • the plain fact is that fewer people are

  • needed to get the jobs done.

  • So one person's productivity enhancing tool

  • is in fact another's pink slip, or more likely,

  • a job opening that no longer needs to be filled.

  • Now, this is called structural unemployment.

  • Automation, whether it's driven by artificial intelligence

  • or not, it changes the skills that

  • are necessary to perform work.

  • I need to move ahead, because we're running out of time.

  • So this is called structural unemployment,

  • and it's the mismatch of skills against the needs of employers.

  • People get put out of work because it's not

  • so much that there's a lack of jobs,

  • but the training that people need to perform those jobs--

  • there's a disconnect.

  • Now, historically, as automation has eliminated

  • the need for workers, the resulting increase in wealth

  • has eventually generated new kinds of jobs

  • to take up the slack.

  • And I see no reason that pattern is not going to continue,

  • but the keyword there is eventually.

  • Let's talk about farm employment.

  • This stuff is amazing, if you look into it.

  • 200 years ago, more than 90% of the US population

  • worked in agriculture.

  • Basically, almost all anyone did was grow and prepare food.

  • That's what it meant to work.

  • Now, today, less than 2% of the population

  • is required to feed everybody, as you can

  • see in the free food over here.

  • Oh my god, is everybody out of work?

  • Of course not.

  • We've had plenty of time to adapt,

  • and as our standard of living has relentlessly increased,

  • which I'll get to in a minute, new opportunities

  • have always arisen for people to fill the expanding

  • expectations of our ever richer and greedier society.

  • Now, if a person from 1800 could see us today,

  • they'd think we'd all gone nuts.

  • Why not work a few hours a week, buy a sack of potatoes

  • and a jug of wine , build a shack in the woods,

  • dig a hole for an outhouse, and live a life of leisure?

  • Somehow, our rising expectations seem

  • to be magically out of pace due to our wealth.

  • OK, so what are the jobs of the future?

  • I don't see why we can't be a society of competitive gamers,

  • artisans, personal shoppers, flower arrangers, tennis pros,

  • party planners, and no doubt a lot

  • of other things that don't exist yet.

  • You might say, well, who is going to do the real work?

  • Well, our great grandchildren may

  • think of our idea of real work as so 21st century.

  • It may take, as we think of with agriculture-- it may only

  • take 2% of the population, assisted by some pretty

  • remarkable automation, to accomplish what's

  • taking 90% of our labor today.

  • So what?

  • It may be as important to them to have

  • fresh flowers in the house each day

  • as it is for us to take a shower every day,

  • which 70% of the US population does.

  • By the way, in 1900, the average was once a week.

  • I'm glad I'm not there.

  • So let me move ahead.

  • That's the good news.

  • The bad news is it it's going to take time

  • for this transition to happen.

  • And there's a new wave of AI enabled applications that's

  • likely to accelerate the normal cycle of job creation

  • and destruction.

  • So we're going to need to find new ways

  • to retrain displaced workers.

  • I was going to go into this.

  • I know Jacob's interested in this,

  • but hopefully, we're going to have to skip over this idea.

  • Our problem is our vocational training system

  • is really messed up.

  • It's mainly because the government today is the lender

  • of first resort for students.

  • So the skills that people learn are

  • disconnected from the needs of the employers

  • in the marketplace.

  • So we're not actually investing in education

  • so much as we're heading out money

  • to people to learn things that won't help them pay it back.

  • You can't get a job?

  • It's too bad.

  • Your student loan is still due.

  • How many of you guys have student loans?

  • OK, not bad.

  • So there are different ways to do this,

  • and we need to create new financial instruments

  • that tie the development of capital,

  • the deployment of capital, to the return on the investment.

  • And I've got this concept that I talk

  • about in my book, which is somewhere around here,

  • that I call a job mortgage.

  • So you get a mortgage for your education,

  • and it is payable solely out of your future earnings stream.

  • And that causes all the right incentives

  • to align so that we're teaching people the right things.

  • Otherwise, people aren't going to give them

  • the money if they don't know that there's going to be

  • a likelihood of a payback.

  • Finally, there's one other dark cloud.

  • I painted a very optimistic view of the future.

  • While it's true that automation makes a society richer,

  • there are serious questions about whose pockets

  • are filled by that wealth.

  • You may be aware while we were on high tech,

  • we tend to believe we are developing

  • dazzling technologies for a needy and grateful world,

  • and indeed, we've made great progress

  • in raising the standard of living for the poorest

  • people on Earth.

  • But for the developed world, the news is not so good.

  • Up until about 1970, on and off, we've

  • found ways to distribute at least some

  • of those economic benefits across society,

  • and this was the rise in the supposed-- the mythical middle

  • class.

  • But it doesn't take much to see that those days are over.

  • They began to diverge.

  • So as economists know, automation

  • is the substitute of capital for labor.

  • And Karl Marx was right.

  • The struggle between capital and labor

  • is a losing proposition for workers.

  • What that means is that the benefits of automation

  • naturally accrue to those who can invest in the new systems.

  • And why not?

  • People aren't really working harder than they used to work.

  • In fact, they aren't really smarter than they used to be.

  • Working hours have actually decreased

  • slowly but consistently for about the last 100 years.

  • The reason we can do more with less

  • is that the business owners invest some of their capital

  • into the process and productivity for improvements.

  • And they reap the most of the rewards.

  • So what has all this got to do with AI?

  • Now, the technologies that are on the drawing

  • boards in our labs are quickening the hearts

  • of entrepreneurs and investors everywhere,

  • as you guys are well aware.

  • And they are the ones who stand to benefit

  • while they export more and more of the risk out

  • to the rest of society.

  • Workers are less secure today.

  • Wages are stagnant.

  • Pension funds can go bust.

  • We're raising a generation of contractors

  • for the gig economy.

  • They're working variable hours, and health benefits

  • are their own problem.

  • That's not true for you guys.

  • You have regular employment jobs.

  • But if you really find out what's

  • going on in the rest of the world, this is true.

  • Now, some people have the mistaken impression

  • that the free market will naturally

  • address these problems if only we can get the government out

  • of the way.

  • And I'm here to tell you that our economy is hardly

  • an example of unfettered capitalism.

  • The fact is that there are all sorts of rules and policies

  • that drive where the capital goes, how it's deployed,

  • and who gets the returns.

  • And the basic problem is-- ah, this is great.

  • I should show the slides while I give the talk.

  • The basic problem is that our economic and regulatory

  • policies have become decoupled from our social goals.

  • And we have to fix that.

  • But the question is how?

  • Now here's the good news.

  • Most people have no idea about this.

  • The good news is that the economy isn't static.

  • It doubles about every 40 years.

  • You guys are familiar with the singularity and the Moore's

  • curve and all that.

  • That's happening with the economy too,

  • not just with computers.

  • It doubles about every 40 years, and it's done that reliably

  • since the start of the Industrial

  • Revolution in the 1700s.

  • In 1800, the average household income was $1,000.

  • And that's about the same as it is

  • today in Malawi and Mozambique.

  • And probably not coincidentally, their economies

  • look surprisingly similar to what the US was 200 years ago.

  • Yet I doubt that people in Ben Franklin's time

  • thought of themselves as dirt poor--

  • that they were barely scratching out an existence.

  • So what this means is that 40 years from now, most likely

  • there's literally going to be twice as much wealth

  • to go around.

  • So the challenge for us is to implement policies

  • that will encourage that wealth to be more broadly distributed.

  • We don't have to take from the rich and give to the poor.

  • We need to provide incentives for entrepreneurs

  • and businesses to find ways to benefit

  • ever larger swaths of society.

  • So in my book, again, I just give you

  • an example of the kinds of policies

  • that smart folks like you could come up with.

  • And the idea here is to make corporate taxes progressive.

  • I'm not saying this is the answer or even an answer.

  • It's just the kind of thinking we need to do.

  • You can make corporate taxes progressive

  • based on how widely distributed the equity in a company is.

  • So companies that have larger stockholder bases

  • have a lower tax rate.

  • Microsoft, to use them as an example,

  • they should pay a far lower tax rate

  • than Bechtel, which is privately held.

  • Now, progressive policies like this,

  • to promote our social goals-- by the way,

  • I flesh that out in the book in quite a bit of detail,

  • how it would work, and I encourage to you to buy a copy,

  • if not read one.

  • Progressive policies like that can promote our social goals

  • without stifling the economy.

  • We just have to get on with it and stop believing the myth

  • that unfettered capitalism is the answer to the world's

  • problems.

  • So let me wrap things up and recap.

  • I don't want you to think I'm anti-AI.

  • Nothing's further from the truth.

  • I think the potential impact on world is similar,

  • and I'm not exaggerating this-- the potential impact is

  • about the same as the invention of the wheel.

  • We need to think of it not of some sort

  • of magical discontinuity in the development of intelligent life

  • on earth, but as a powerful collection of automation tools

  • with the potential to transform our livelihoods

  • and to vastly increase our wealth.

  • The challenge we face is that our existing institutions,

  • without some enlightened rethinking,

  • run a serious risk of making a mess of this opportunity.

  • I'm supremely confident that our future

  • is very bright-- that it it's more

  • "Star Trek" than "Terminator."

  • But the transition is going to be protracted and brutal

  • unless we pay attention to the issues

  • that I tried to raise with you here today.

  • We have to find new and better ways

  • to ensure that our economy doesn't motor on,

  • going faster and faster, while throwing ever more people

  • overboard.

  • Our technology and our economy should serve us, not

  • the other way around.

  • So thank you.

  • I'm sorry to run so long.

  • Next time I give the talk, it would be half this long.

  • [APPLAUSE]

  • Do we have time for questions?

  • MALE SPEAKER: Yeah, so we do have a little bit of time

  • for questions.

  • There is a mic placed right over there

  • for those who want to ask questions, please line up.

  • You guys need to do that.

  • Go for it.

  • AUDIENCE: So for the task of taking over the world,

  • are there other means except for having human intelligence?

  • JERRY KAPLAN: Yeah, it's a very dangerous issue,

  • as a matter of fact, because a lot of the AI technologies,

  • we talk about the productivity and all of that.

  • But they have very serious applications in, for example,

  • military use.

  • And this is a very difficult problem.

  • A lot of very smart people are actually--

  • I wouldn't say secretly, but not publicly working on.

  • A friend of mine is on his way to Geneva right now.

  • There's meetings regularly with the UN.

  • There's a lot going on in the US military,

  • because they recognize that just going ahead and implementing

  • the kinds of technologies to the battlefield that are currently

  • being applied to driving cars and other things

  • might backfire, because it would be a lot easier

  • for non-state actors, to put it politely,

  • and dictators to-- today it takes enormous investment

  • to make and buy bombers and all that kind of stuff.

  • It's going to be really cheap, just like everything else,

  • like cellphones.

  • And there are some really creepy things

  • that can be done to take over the world

  • and wipe out humanity at a very low cost,

  • and that's going to be a big problem.

  • AUDIENCE: So you gave a lot of examples of problems

  • that we thought that were innately human,

  • but we were later able to describe it in a different way.

  • So why should we doubt that unbounded learning

  • is a problem we can't describe in a different way?

  • By unbounded learning, I mean the example

  • you gave-- oh, that's a cat.

  • Or for humans, you know, if you ask them, pass me

  • the purple cup, they'll learn that's purple

  • if they didn't know that word.

  • Why can't that unbounded learning be described in a way

  • that we can train machines to learn in the same way

  • that babies learn?

  • JERRY KAPLAN: Well, I'm going to turn your question

  • around a little bit to put it in the context of what

  • I said here.

  • I'm not saying that any particular task is completely

  • impervious to machine learning.

  • In fact, it might very well be.

  • However, machine learning is really

  • good at picking out patterns in large volumes of data

  • as it's practiced today.

  • It could be a future form of software technology, which

  • could do something more into what you said,

  • but that doesn't mean that they have human-like characteristics

  • or human learning.

  • And it doesn't mean that all of our jobs

  • will go away because a lot of our jobs

  • require face-to-face interaction or the expression of empathy.

  • And I don't buy the idea that machines can effectively

  • express empathy in their dealings with other people.

  • So the tasks can go away, and maybe they

  • will be able to learn as well.

  • That would be great.

  • That's a cat.

  • That's a chair.

  • That's that-- boom, boom, boom, the machine's got it.

  • But that's just another step in a long line of things

  • where people looked, and they went, wow!

  • It used to take people to do that.

  • Now a machine can do it.

  • It's just the next step.

  • So a lot of that stuff is going to go away.

  • And if we all wind up-- nobody wants

  • to watch a robot play competitive tennis.

  • It's just not interesting.

  • So I mean, there are lots of these jobs that inherently

  • require human beings.

  • So I'm trying to separate the task of automation

  • from what will people do.

  • And I hope that began--

  • AUDIENCE: Yeah, I guess it was even just the empathy example.

  • You say that robots can't be empathetic, but maybe they can.

  • Maybe we just think that's an innately human thing.

  • The tennis example actually is very convincing to me,

  • like, I would never watch a robot play tennis.

  • But if a robot was just as empathetic

  • and had a human form, and you couldn't

  • tell that it was a robot when you walked into the doctor.

  • We think it's a human problem, but maybe it's not.

  • JERRY KAPLAN: Boy, this is a really complicated topic.

  • What you said is right.

  • We can fool people.

  • And we do this all the time.

  • We can build a way-- this has gone on since the 16th century,

  • where they built automatons.

  • We went, oh, my god, it's just like a person.

  • And they were amazing.

  • Have you ever seen these?

  • These mechanical devices are absolutely incredible-- 16th

  • and 17th century.

  • It was fun entertainment for the courts of kings.

  • But if you know it's a machine, the fact that the screen comes

  • up and says, thank you, I really appreciate

  • that you placed your order.

  • I mean, come on.

  • It just doesn't compute emotionally.

  • We're not going to buy that story.

  • So a lot of it has to do with, like, toys that look like dogs

  • or look like children or play.

  • It's all very complicated, because play--

  • if you're doing it knowingly, that's perfectly reasonable.

  • If you're doing play because you're being fooled, or being

  • persuaded to buy something because the machine has gone,

  • oh, come on.

  • I got a whole bunch of hungry mouths to feed.

  • Oh, please, buy this car for me.

  • We're not going to like that.

  • That's my point.

  • AUDIENCE: Thank you.

  • JERRY KAPLAN: Thanks.

  • Yes, sir.

  • No waiting on check stand 2.

  • AUDIENCE: So I agreed with most everything you said.

  • I had a problem with one of your examples,

  • and this may seem like a nitpick.

  • But I'm going to flip this around.

  • You said teaching a machine a new task was like teaching

  • a primate to use sign language.

  • So do you think primates don't use intelligence the way

  • we do and don't understand what's being said?

  • JERRY KAPLAN: That's a good point.

  • I think that the point I was trying to make

  • is different than the one that you-- I'm not saying I

  • didn't say what you said, but that's not really what I meant.

  • What I meant was you can take something

  • that has no natural affinity for that particular task,

  • and you can get it to do that task

  • to some level of competence.

  • That's what I was trying to say.

  • Your point about chimp stuff is pretty interesting,

  • because obviously they have brains.

  • And I think most reasonable people think

  • they have rudimentary minds.

  • But the point is they don't naturally use sign language,

  • and as I say, you teach a bear to ride a bike.

  • That's not like, oh, my god, what are we going to do?

  • Next thing you know we'll be teaching bears to drive cars.

  • It's that we can make machines that also appear human-like

  • and do these human-like activities,

  • but it's not a natural part of the process.

  • Machines have certain characteristics,

  • and I can give another talk on that subject.

  • What are those characters ?

  • And they're different than people,

  • and we need to understand that and stop

  • thinking about ourselves as we're making more and more

  • intelligent machines.

  • I'm just giving you the framing to help

  • us to understand the economic results that are appropriate.

  • Thank you.

  • MALE SPEAKER: If we go quickly, we

  • have time for two more questions.

  • JERRY KAPLAN: Two more, OK.

  • AUDIENCE: You were talking a bit about the social impacts,

  • and in the end, the people who own the machines

  • get the benefit from the machines.

  • And I agreed there when you talked about ways

  • to change policy, but historically,

  • social-focused policies have come

  • from things like labor movements-- people

  • controlling the means of production or whatever.

  • These kinds of things where people go on strike.

  • Who is going to strike if all the people doing tasks

  • have just been replaced?

  • I own a machine.

  • I don't have any workers, right?

  • I got a factory that builds everything I need,

  • and I try to sell it to people.

  • And eventually it might implode, I don't know,

  • if everybody is doing it.

  • But the remaining jobs are really

  • just kind of these neat, supervisory things or the 1%

  • of the population that can get a big audience on whatever

  • medium.

  • How do you run a whole economy based on that?

  • JERRY KAPLAN: Well, there are two basic points you're making.

  • Let me try to respond to them each briefly.

  • Because you're right in a lot of senses.

  • When you automate people out of jobs, for those jobs,

  • we don't need people.

  • And the question is when are the new jobs

  • going to arrive, if ever?

  • Most people are thinking about this statically,

  • like, well, we're just going to automate away 90% of the jobs

  • when we can build machines that dig ditches

  • and do all that kind of stuff.

  • But I think historically what happens

  • is new jobs are created that require humans

  • for one reason or another.

  • And I tried to kind of make that point.

  • We really can be a leisure society.

  • What we would think of as a leisure society,

  • that's what you'll get paid for in the future.

  • In 80 years, the average American, if these trends hold,

  • the average American household would

  • be making $200,000 a year.

  • Now, most of you guys make $200,000 a year, I understand.

  • It pays well here, and I was making a joke.

  • But my point about that is that there

  • are going to be people-- when you're

  • making that kind of money, you want those fresh flowers

  • every day, and you may be willing to pay somebody

  • to do that and pay them a living wage to do it.

  • So I think the nature of work is going to shift.

  • Those people from 1800 who look at us today and think

  • we're crazy because we are we're doing

  • stuff we don't need to do for people who don't need it done.

  • And that is the way they would look at it.

  • And I can't say for sure, but I think that pattern

  • is likely to continue.

  • It's just very hard to visualize what that's

  • going to be like in 80 years.

  • AUDIENCE: My question is when do you

  • think AI will get to the point where

  • it can predict the behavior of other AI actors?

  • Because I think that's the heart of human intelligence

  • in the social context, and we haven't really

  • seen much in that task space.

  • JERRY KAPLAN: Wow, again, a very complicated-- I

  • could go on for a long time on this.

  • This has come up already in things like the flash

  • crash of 2010, which I cover in my book, which you're all

  • encouraged to take a look at.

  • It's a real problem, because people are stealthy

  • developing these systems, and it creates

  • what's called systemic risk.

  • Because these machines are like gods in terms of the damage

  • they can inflict in milliseconds.

  • And so it shut down the US power grid.

  • You can bet that China today, or I shouldn't pick out

  • China-- powerful players today have the ability

  • to completely decimate our economy for a fair period

  • of time almost instantly-- almost like a press

  • of a button.

  • And so these are difficult issues,

  • because you get two of those.

  • You ever seen the old movie "Colossus-- The Forbin

  • Project"?

  • Anybody?

  • So one guy, two guys will know what I'm talking about.

  • It's about just that.

  • They created an intelligent system.

  • The Russians-- this was like 1960 when they made the film,

  • it was great-- also did that.

  • And the two of them figured there had to be another one,

  • and they got together, and they took over the world.

  • It's actually a pretty cool film.

  • It's not as stupid as it sounds.

  • So it's a real issue.

  • These side effects, the unintended

  • consequences-- the kinds of technology

  • we're developing-- that's another hour-long talk.

  • It doesn't mean we shouldn't do it.

  • It means we need to be aware of it

  • to figure out how to control it in reasonable ways.

  • So I apologize [INAUDIBLE].

  • If anybody wants to stay and hear stories about early Google

  • like I would-- well here he is.

  • I won't do it on camera because I don't want anybody

  • to record it.

  • But thank you.

  • Thank you so much for listening to my crazy rants.

  • [APPLAUSE]

MALE SPEAKER: So today we're here to see Jerry Kaplan.

Subtitles and vocabulary

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