Ray Kurzweil: “How to Create a Mind” | Talks at Google

Ray Kurzweil: “How to Create a Mind” | Talks at Google


BORIS DEBIC: Welcome, everyone,
to yet another [email protected] talk. It is my distinct privilege
today to host Dr. Ray Kurzweil, and my good friend
here, Peter Norvig, from our Artificial Intelligence
group, former director of research at Google. So just to give you a little
bit of context why I am hosting this talk. When I was a kid and wrote
my first lines of code in elementary school, I saw a
tremendous potential in that toy that I was playing with. And I said to all my friends,
you know what? One of these days, these
are going to be as smart as humans. We just have to work
a lot at it. And they would say, oh,
no, that’s impossible. How can you say something
like that? I really didn’t have a good
answer in those times. I was just a kid. But I told them, look,
I mean it’s all built of atoms, right? The CPUs in this thing,
that’s atoms, and our brains, that’s atoms. So there’s no theoretical impossibility for this to happen. Well, today, I’m very happy to
host two guys who can explain why this will happen in
much more detail. Please welcome Dr. Ray
Kurzweil to Google. [APPLAUSE] PETER NORVIG: I think it’s
redundant to introduce Ray. You all know him as an inventor,
author, a futurist. And you know, there was a book
a few years back that accused Xerox PARC of fumbling
the future. And I would say, to continue
that metaphor, Ray has intercepted the future and
returned it for a touchdown, multiple times. He’s done it with the flatbed
scanner, with OCR, with print-to-speech, text-to-speech,
speech recognition, music synthesis,
and so on and so on. I won’t list all the honors,
but he’s been recognized by Presidents Johnson,
Clinton, and Reagan, and by Bill Cullen. Those of you who are younger,
you’ll have to Google that. But let me put it this way. Have you heard of Plato,
Aristotle, Socrates? Philosophers. And Ray is a philosopher, too. But more importantly, foremost,
he’s an engineer. And when it comes to these tough
questions of creating the mind, philosophers are
useful, but I’m putting my money on the engineers. [LAUGHTER] [APPLAUSE] RAY KURZWEIL: Well, thanks
for that, Peter. I– can you hear
me back there? Yeah? I agree with that. In fact, I decided I wanted to
be– well, I called it an inventor when I was five. And I had this conceit. I know what I’m going to be,
and it kind of reflected my family philosophy that if you
have the right ideas, you can overcome any problem. And I particularly
like coming here. This is actually my third
time at [email protected] I was here in 2005. I wouldn’t exactly say
Google was a young upstart at that time. It was, I think, about
4,000 people. I did it in the lunchroom
near here. The spirit hasn’t changed. I think you’re about
10 times the size. 40,000 is like the size
of a small city. But you’re still actually a
start-up compared to the opportunity, because the world
is increasingly based on knowledge and information. In fact, 65% of American workers
are knowledge workers. So the mission of organizing
and providing intelligent access to all the world’s
knowledge is the most important task in the world,
and Google is clearly the leader in that. And there’s tremendous
potential, because knowledge is growing exponentially. So I want to say a few words
about exponential growth and my law of accelerating returns,
which was the primary message of “The Singularity is
Near.” But I think Google is actually a very good example
of that exponential growth. I happened to be on Moira
Gunn’s “Tech Nation” NPR program yesterday, and she was
reminiscing about her 2001 interview with Larry and Sergey,
who came in with dark suits and ties. And they were trying to explain
this cool computer they were going to create. And she didn’t quite understand
what it was. And Larry said, well, it’s
going to be like HAL. And then Sergey said, but
it won’t kill you, so. [LAUGHTER] RAY KURZWEIL: So I think we got
the second part of that. The first part of that we have,
in the sense that Google is pretty amazing in terms
of finding information. I’m amazed by it every hour. But I think we can go further in
that direction, and that’s what I’d like to talk about. You all have these billions of
pages of millions of books, and very good access to it,
but there’s a lot of information there that’s
reflected in the natural language ideas. And computers, now, I think can
begin to understand those. And that’s something
I’m working on. That’s something I talk
about in this book. And I’d like to share
that idea with you. First, I’ll say a few words
about the law of accelerating returns. I mentioned I decided to be an
inventor when I was five. I realized 30 years ago
that the key to being successful is timing. Those inventors whose names you
know are the ones who got the timing right. So Larry and Sergey had
this great idea about reverse-engineering the links
on the internet to provide a better search engine,
but they did it at exactly the right time. And so in 1981, I was thinking,
my project has to make sense when I finish the
project, and the world will be a different place two, three,
four years from now. That was even true in ’81. It’s even more true today. Acceleration is another
feature of the law of accelerating returns. Our first communication
technology, spoken language, took hundreds of thousands
of years to develop. Then people saw that stories
were drifting. People didn’t always retell the
story in the same way, so we needed some record of it. So we invented written
language. That took tens of thousands
of years. Then we needed more
efficient ways of producing written language. The printing press actually
took 400 years to reach a mass audience. I gave a speech at the
University of Basel recently on the occasion of it’s
550th anniversary. It was founded 20 years after
Gutenberg’s invention, right near the spot where
he invented it. And I said, well, you must have
had some of his books when you opened your doors. And they said, yes, we got
them very quickly. It was only a century later. I mean, that was the Google
of that time. It took maybe a century to find
the right information. So you didn’t really find
it in your lifetime. It took 400 years for that
really to reach an appreciable number of people. The telephone reached 25% of the
US population in 50 years. The cell phone did that
in seven years. Social networks– wikis, blogs– took
about three years. Go back three or four years ago,
most people didn’t use social networks, wikis
and blogs. Ten years ago, most people
didn’t use search engines. That sounds like ancient
history, but it wasn’t so long ago. And then we very quickly become
dependent on these brain extenders. I mean, during that one-day
SOPA strike, I felt like a part of my brain had
gone on strike. Because there was a way around
it, but I didn’t know that until the day came. So I really felt like I’m going
to lose part of my mind. Yet this was not technology
that I had even a few years earlier. What’s driving this is the
exponential growth of information technology. In 1981, I began to look at
data, being an engineer. But I started out with the
common wisdom that you cannot predict the future. And that remains true as to
which company, which standard will succeed. But if you measure the
underlying properties of information technology– the first one I looked at, in a
classical one, the power of computation per constant
dollar. So the calculations per second
per constant dollar. Or the number of bits we’re
moving around wirelessly, or the number of bits on the
internet, or the cost of transmitting a bit, or the
spatial resolution of brain-scanning, or the amount
of data we’re downloading about the brain, or the cost of
sequencing a base pair of DNA or a genome, or the amount
of genetic data we’re sequencing– I mean, these fundamental
measures follow amazingly predictable trajectories, really
belying the common wisdom that you cannot
predict the future. And what’s predictable is that
they grow exponentially. And that is not intuitive. Our intuition about the future
is that it’s linear, not exponential. If you ever wondered, why
do I have a brain? It’s really to predict the
future, so we could predict the consequences of our
actions and inactions. So I’m walking along, and, OK,
that animal’s going that way towards a rock, and I’m
going this way. We’re going to meet in about
20 seconds up at that rock. I think I’ll go a
different way. That proved to be useful
for survival. That became hardwired
in our brains. Those predictors of the future
are linear, and they work very well for the kinds of situations
we encountered when our brains evolved
1,000 years ago. It’s not appropriate for the
progression of information technology. And I’d say the principal
difference between myself and my critics is they look at the
current situation and they make linear extrapolations. So halfway through the Genome
Project, seven years, 1% had been completed, and mainstream
scientists who were still skeptical said, I told you this
wasn’t going to work. Seven years, 1%? It’s going to take 700
years, like we said. My reaction was, no,
we’re almost done. [LAUGHTER] RAY KURZWEIL: I mean, 1%, you’re
pretty much finished. I mean that’s– you can try that with your
product submission schedules. [LAUGHTER] RAY KURZWEIL: But over the
next– it had been doubling every year. There was reason to believe
that would continue. It was only seven doublings
from 100%. And that’s exactly
what happened. It kept doubling and was
finished seven years later. That has continued. Up to the present day,
the first genome cost a billion dollars. We’re now down to under
$10,000, and so on. And it’s true in every area
of information technology. Not everything– I mean, transportation’s not yet
an information technology. But industries are converting. It’s not just the gadgets
we carry around. Health and medicine has become
an information technology. I’ll talk about that. The world of physical things
is going to become an information technology as
three-dimensional printing gets going, and I’ll
touch on that. It’s worth just examining for
a moment the difference between linear progressions,
which is our intuition, and the reality of information
technology, which is exponential. So linear goes one,
two, three, four. Exponential, which is
information technology, goes to two, four, eight, siexteen. Is that really so different? Actually, it’s not
that different. A linear progression is a
good approximation of an exponential one for a short
period of time. I mean, look at an
exponential. Take a little piece of it. It looks like a straight line. It’s a very bad estimate over
a long period of time. At step 30, the linear
progression’s at 30. At step 30, the exponential
progression’s at a billion. And that’s not an idle
speculation about the future. This Android phone is several
billion times more powerful per constant dollar than the
computer I used as an undergraduate. It’s a million times cheaper,
it’s several thousand times more powerful, in terms of
computation, communication, memory, and so on. And it’s also 100,000
times smaller. That’s another exponential
progression. And we’ll do both of
those things again in the next 25 years. So that gives you some idea
of what will be feasible. So this is what I
wanted to cover. Any questions on any of this? [LAUGHTER] Well, this was the first
graph I had, in 1981. So I don’t know if you
can see that, but I had it through 1980. And this calculations per second
per constant dollar. It’s a logarithmic scale, which
I have to take some pains to explain to
many audiences. But every labeled point on this
y-axis is 100,000 times greater than the
level below it. So this modest little uptick
represents trillions-fold increase in the amount of
computer you can get per constant dollar over the last
century, going back to the 1890 census. Several billion-fold, just
since I was a student. People go, oh, Moore’s law. But Moore’s law is actually just
the part on the right. That had actually only been
underway for a little over a decade when I did
this estimate. This started decades before
Gordon Moore was even born. 1950s, they were shrinking
vacuum tubes, making them smaller and smaller to keep this
exponential growth going. CBS predicted the election of
Eisenhower with a vacuum-tube based computer in 1952. Remember that? [LAUGHTER] A few people here might
remember it. When I first talked to Google in
2005, I don’t think anybody remembered it. But finally, that hit a wall. Couldn’t shrink the vacuum tubes
anymore and keep the vacuum, and that was the
end of that paradigm. But it was not the end of the
ongoing exponential, it just went to the fourth paradigm. And people have been talking
about the end of Moore’s law, but the sixth paradigm will be
three-dimensional computing. We’ve taken baby steps
in that direction. If you talk to Justin Ratner,
the CTO of Intel, he’ll show you this experimental circuits
they have that are three-dimensional self-organizing molecular circuits. Those will become practical in
the teen years, before we run out of steam with flat
integrated circuits, which is what Moore’s law is all about. But the most interesting thing
about this is, just look at how smooth and predictable
a trajectory that is. People say, well, it must have
slowed down during the Great Depression, or the recent
recession– neither of which is the case. Did Google slow down during
the recent recession? I mean, these technologies
continue because we’re creating the computers and the
systems and the search engines of 2013 and 2014 with the
computers of 2012. We couldn’t do that in 2002. We had computers of 2002,
so we created the systems of 2003. That’s why the technology
builds on itself. But it goes through thick and
thin, through war and peace, through boom times
and recessions– nothing seems to affect it. And we could talk about natural
limits, but I examine that in “The Singularity is
Near,” and if you look at what we know about the physics of
computing, we do need a certain amount of matter and
energy to compute, to remember, to transmit a bit,
but it’s very, very small. And based on the limits that we
understand that have been demonstrated, we can go well
into the century and develop systems that are many trillions
of times more powerful than we have today. So I won’t dwell on these
examples of electronics, but you could buy one transistor
for $1 in 1968. I thought that was actually
pretty cool, at the time, because in the early ’60s, I
would hang out at the surplus electronic shops on Canal
Street in New York– they’re still there– and buy something this big, a
telephone relay that could switch one bit, for $50. And it was big and slow,
30-millisecond reset time. I can actually get something
much faster and smaller for $1. Today, you can get
billions for $1. And they’re better, again,
because they’re smaller, so the electrons have less
distance to travel. Cost for a transistor
cycle is coming down by half every year. That’s a measure of
price/performance. So the fact that you can buy an
Android phone that’s twice as good as the one two years ago
for half the price partly is because Google is clever,
but partly it’s because of this law of accelerating
returns. It’s a 50% deflation rate. We put some of that
price/performance improvement into better performance and some
of it into lower prices. So you get better products
for lower costs. And that’s going to continue
for a very long time. The economists actually
worry about deflation. We had massive deflation
during the Depression. That was a different source. It was not price/performance
improvement. It was the collapse of
consumer confidence. But they’re still concerned as
more and more of the economy becomes information technology,
like all of health and medicine. Peter’s working on education
becoming information technology. And if you can get
the same stuff– computes, bits of communication,
base pairs of DNA, physical things
printed out on three-dimensional printers– for half the cost of a year ago,
Economics 101 will say that you will buy more. But you’re not going to double
your consumption year after year, because after all,
how much do you need? You’ll reach a saturation
point. So maybe you’ll increase
your consumption 50%. And so the size of the economy
of these information technologies will shrink, not
as measured in bits, bytes, and base pairs, but as measured
in constant currency. And for a variety of good
reasons, that would not be a good thing. And that is not what
is happening. In fact, we more than double
our consumption each year. This is bits shipped, but I
have 50 other consumption graphs like this. Every form of information
technology has had an average growth rate of 18% per year
for the last 50 years in currency, despite the fact that
you can get twice as much of it each year for
the same price. And the reason for that is, as
we reach certain points of price/performance, whole new
applications explode. I mean search engines like we
have now, or even like we had 10 years ago, weren’t feasible
20 years ago. Search engines– there were search engines before
three or four years ago, but they didn’t take off
because they weren’t even able to upload one picture. And when the price/performance
reached a certain point, these applications exploded. And we have an insatiable
appetite for information, for knowledge– which is really information
that has been shaped by meaning. That’s the mission of Google,
is to turn information into knowledge that people can
access and benefit from. So “Time Magazine” had a cover
story on my law of accelerating returns. They wanted to put a particular
computer they had covered and were fond
of on the chart. I said, well, I don’t know. It might be below the chart,
because sometimes people come out with things that are not
cost-effective, and then they don’t last in the marketplace. This has just come out. But it actually was
on the curve. It’s the last point there. This is a curve I laid
out 30 years ago. I laid it out through 2050. But we’re right where
we should be. This has been an amazingly
predictable phenomenon. Communication technology– Martin Cooper is one of the
faculty at Singularity University. He invented a product that you
sell, the mobile phone. And that’s the number of bits
of data we send around wirelessly in the world. So it’s over the last century. A century ago, this was Morse
code over AM radio. Today, it’s 4G networks. And again this is trillions-fold
increase. That’s a logarithmic scale. But look at how smooth a
progression that is. Internet data traffic. This is a graph I had just the
first few points of in the early ’80s. It was the ARPAnet. And I said, wow, this is going
to be a world wide web connecting hundreds of millions
of people to each other and to vast knowledge
resources by the late ’90s. I wrote that in the ’80s. And people thought that was
ridiculous, when the entire defense budget could only tie
together a few thousand scientists. But that’s the power of
exponential growth. That is what happened. That’s the same data on the
right, seen on a linear scale. That’s how we experience
the world. So to the casual observer, it
looked like, whoa, the World Wide Web is a new thing,
came out of nowhere. But you could see it coming. And you can see revolutions
coming if you look at these progressions. And that is what I advise
young companies to do. Because I get some business
plans and do some entering, and very often, these plans talk
about the world three, four years from now, like
nothing is going to change. And you only have to look at the
last three or four years to see that that’s
not correct. I could talk for a long time
about this phenomenon. But we are turning health and
medicine into an information technology. I mentioned the Genome
Project. But we can actually reprogram
this outdated software in our bodies. How long do you go without
updating your Android phone software? This is probably updating
itself right now. But I’m still walking around
with software in my body that evolved thousands of years ago–
like, for example, the fat insulin receptor gene, which
says, hold onto every calorie ’cause the next hunting
season may not work out so well. That was a good idea
1,000 years ago. You worked all day to
get a few calories. There were no refrigerators,
so you stored them in your fat cells. I’d like to tell my fat insulin
receptor gene, you don’t need to do that anymore. I’m confident the next hunting
season will be good at the supermarket. [LAUGHTER] RAY KURZWEIL: So that was
actually tried in animal experiments. We have a number of ways of
turning genes off, like RNA interference. And these animals ate ravenously
and remained slim and got the health benefits of
caloric restriction while doing the opposite. They lived 20% longer. They’re working with a drug
company to bring that to the human market. I’m on the board of a company
that takes lung cells out of the body of patients who
have a disease caused by a missing gene. So if you’re missing this gene,
you probably will get this terminal disease, pulmonary
hypertension. So they scrape out lung cells
from the throat, add a gene in vitro, and then inspect that
it got done correctly, replicate the cell several
million-fold– that’s another new
technology– inject it back in the body, it
goes through the bloodstream. The body recognizes them
as lung cells. You’ve now added millions of
cells with that patient’s DNA, but with the gene they’re
missing, and this has actually cured this disease in successful
human trials, and it’s doing its Phase III trial
now before it gets approved. There are hundreds of examples
of reprogramming biology. My father had a heart attack
in 1961, damaged his heart, which is the case of 50% of all
heart attack survivors, have a damaged heart. He could hardly walk. He died of that in 1970. Up until very recently, there’s
nothing you could do about it, because the
heart does not rejuvenate itself naturally. You can now reprogram stem cells
to rejuvenate the heart. Now, I’ve talked to people who
could hardly walk, and now they’re normal. We are growing organs already. Some of these simpler organs
are being used in humans. Other ones are now being
implanted in animals, where we lay down the scaffold with
three-dimensional printers and then use the three-dimensional
printer to populate it with stem cells and regrow, for
example, a kidney. So all of this is coming. It’s a complex area. But the point is that health
and medicine has become an information technology, and
therefore it’s subject to this law of accelerating returns. So these technologies, which are
already beginning to enter clinical practice, they’re going
to be 1,000 times more powerful in 10 years and
a million times more powerful in 20 years. It gives you some idea
of what’s coming. If I want to send you a music
album or a movie or a book, just a few years ago, I’d send
you a FedEx package. I can now send you a Gmail
message with those products as an attachment. I can also send you these
musical instruments, if you have the three-dimensional
printer. And this is a revolution
right before the storm. They’ve been expensive. They were hundreds of thousands
of dollars and tens, now thousands. They will, in a number of
years, go sub-$1,000. The resolution is improving at
a rate of about 100 in 3-D volume per decade. It’s still over several
microns. Needs to be sub-micron. The range of materials
is increasing. Ultimately, a substantial
fraction of manufacturing will be done this way, turning
information files into physical products. Today, you can print out 70%
of the parts you need with your three-dimensional printer
to create another three-dimensional printer. [LAUGHTER] RAY KURZWEIL: That will be 100%
in five to eight years. So that brings me
to the brain. And I want to spend
some time on that. I’ve been thinking about this
topic for 50 years, actually, thinking about thinking. I wrote a paper when
I was 14– that’s 50 years ago– that basically described the
human brain as a large number of pattern recognizers. That was my Westinghouse
Science Talent Search submission, and I got to
meet President Johnson. And I did a program that did
pattern recognition on musical melodies and then wrote original
music with the patterns it had discovered. So you could feed in Chopin,
and it would write, then, music like it was a student of
Chopin or Mozart, and you could recognize which composer
had been analyzed with the original music that
it was composing. And this book actually
articulates a very consistent thesis. Pattern recognition is
what we do well. We’re not very good at
logical thinking. Computers do a far better
job of that. One of the predictions
I made in the early ’80s was that by ’97– actually, I said ’98– a computer would take the World
Chess Championship. I also predicted that when
that happened, we would immediately dismiss chess as
being of any significance. Both of those things happened
in ’97 when Deep Blue defeated Kasparov. And people said, well, of
course that’s true. Chess is a logic game, and
computers are logic machines, so we would expect them
to do a better job than humans on chess. But what they will never do is
be able to understand the vagaries and subtleties and
ambiguities of human language. So already we’re seeing
that being overturned. And there’s actually a pretty
impressive range– it’s just a first step– but an impressive range of
language that you can say to systems like Google now, and it
will understand you pretty well, and actually begin to
develop a model of who you are, something that
Siri doesn’t do. How many of you can answer
this “Jeopardy” question? “A long tiresome speech
delivered by a frothy pie topping.” What is a
meringue harangue? [LAUGHTER] So Watson got that correct. The two humans who were the best
human “Jeopardy” players ever did not get it. And Watson got a higher
score than the best two humans put together. And there’s a lot of misunderstandings about Watson. People say, well, it’s not
really doing any true understanding of language,
because it’s just doing statistical analysis of words. Actually, what it does– I mean, it actually has many
different modules. What the IBM engineers did is
create a framework called UIMA, which runs these different
systems and is able to analyze their strengths and
weaknesses and combine them. So actually, the engineers
in charge of Watson don’t necessarily understand
all of those modules. The ones I think that are most
effective are ones that are statistical, but they’re
not just doing statistics on word sequences. They’re building a hierarchical
model with a whole field of probabilities
at different levels of the hierarchy. And if that does not represent
a true understanding of the material, then humans have no
true understanding, either, because that is how the
neocortex works. And another misconception is
that every fact was sort of programmed in some language
like Lisp. In fact, Watson got its
knowledge by reading Wikipedia and several other encyclopedias,
200 million pages of natural language
documents. And it is true that it actually
doesn’t do as good a job on any page as a human. So you could read a page, and if
you knew nothing about the presidency, you’d conclude,
wow, there’s a 95% chance Barack Obama’s president, having
read that one page. And Watson will read it and come
out with a conclusion, oh, there’s a 58% chance that
Barack Obama is president. So it didn’t do as good a job
of understanding that page. But it has read 200 million
pages, and maybe 100,000 of those have to do with Barack
Obama being president. And it can then combine all
those probabilities using sound probability theory– Bayes’ theorem and so on– and conclude that there’s a
99.99% chance that Barack Obama is president. It has total recall of those
200 million pages and can analyze the cross-implications
in three seconds. It’s just a first step, but that
is the kind of capability that we’re leading to. My vision of search engines in
the not-too-distant future is that they won’t wait to
be asked questions. They’ll be listening in
on our conversations– what we say, what we write, what
we read, what we hear, if you let them, and I believe
people will, because it’ll be useful to have an intelligent
assistant like this– and it will anticipate
your needs. So suddenly, it might pop up and
say, oh, just yesterday, you were talking about, if
only we could have better bioavailable means of
phosphatidylcholine. Well, here’s a study that came
out 36 minutes ago on just that topic. If it sees you struggling in a
conversation to come up with the name of that actress, right
in your field of view on your Google Glass, you’ll get
information about that actress, not even having
asked for it. It can just see you
needed that. Obviously, that could be
annoying if it’s really information you don’t want. That’ll be the key. But actually, we very much
want this information. I mean, people are constantly
Googling something at dinner. But we don’t even want to have
to put that information in. An intelligent assistant
should be listening to what we say. So some of the best evidence
for the thesis I’ve come up with on how the neocortex works
has emerged just as I was sending off the book. Actually, four times I was
about to send it to the publisher and said, no,
wait, this great research just came out. I’ve got to include this. And we actually delayed
the book as a result. The publisher wasn’t happy
with that, but these were great pieces of research
to support the thesis. The thesis is that there are
modules in the brain that are comprised of about 100 neurons,
and that each one of these recognizes a pattern and
is capable of wiring itself, literally with a wire,
biological wire, an axon and a dendrite, to other modules to
create this hierarchy that the neocortex represents. And that hierarchy doesn’t
exist when the brain is created. Even before we’re born, we
start building this, one conceptual layer at a time. And that’s actually the secret
of human thought, the ability to build these modules. One piece of research that came
out just as I was sending off the book is that the
neocortex is comprised of these modules of about
100 neurons. The wiring and structure
of those 100 neurons is not plastic. It’s stable throughout life. It is the connections between
these modules which are dynamic and plastic
and are created. And our neocortex creates our
thoughts, but our thoughts create our brain, in terms of
these connections and the patterns that each
module learns. This is different from neural
nets, and I’ve never been a fan of neural nets. I was one of the pioneers of
hierarchical hidden Markov models in the ’80s and ’90s
and used that for speech recognition, and today, that is
the dominant technique in speech recognition, speech
synthesis, character recognition. It’s one of the popular
techniques in natural language understanding. And it’s really the closest
mathematical equivalent to what I’m talking about. This 100-neuron module is more
complex than one neuron in a neural net. It’s capable of dynamically
learning a pattern, recognizing the pattern
even if parts of it are occluded or missing. It can actually tell other
pattern recognizers to expect a pattern because it’s almost
recognized a pattern and another part’s coming, and
so lower-level pattern recognizers should be
alert for that. It’s capable of creating
these connections up and down the hierarchy. And that’s much more
complex than one neuron in a neural net. So the neural net is based on
one neuron, either a model of it that we have in synthetic
neural nets or, in theory, the neural net that the
brain represents. And that’s not the right
building block, either for AI or for the brain. There was this recent research
at Google that showed an ability to do image recognition
with a neural net without any labeling
of the data. It was impressive, but it only
recognized 15% accuracy. I think a much better model is
based on not having the neuron as the building blocks. The building block are
these modules. And we have about 30 billion
neurons in the neocortex. There’s about 300 million of
these pattern recognizers. Now a word about
the neocortex. It is this part of the
brain where we do hierarchical thinking. It can think in hierarchies, and
it can solve problems in hierarchies. And it can see a solution to a
problem and then reapply it in situations that might be
a little different. And only mammals have
a neocortex. So 100 million years ago,
these mammals emerged, rodent-like creatures with a
neocortex that was the size of a postage stamp, about as thin
as a postage stamp, flat and smooth, and it covered
the brain. But it was capable of a
certain amount of this hierarchical thinking. So these mammals could solve
problems quickly, or could see another member of its species
solve a problem and learn it in a matter of hours. Animal species without a
neocortex could learn, too, but not in the course
of one lifetime. They had pre-programmed
behaviors. Those behaviors could evolve in
biological evolution, but that would take thousands
of lifetimes. So over thousands or tens of
thousands of years, they could gradually change
their behavior. And that was OK, because
the environment changed that slowly. So there would be environmental
changes that required an accommodation
in behavior over thousands of years. But then 65 million years ago,
there was a cataclysmic event that happened very quickly
called the Cretaceous extinction event. And we see archaeological
evidence of that around the globe. There’s a layer that represents
this catastrophic change in the environment that
happened very quickly. And there are theories
about that having to do with a meteor. But it’s very clear that there
was a sudden change in the environment at that time. And the animals that didn’t
have a neocortex and that couldn’t adjust quickly,
thousands of those species died out. That’s when the mammals took
over their ecological niche of small- and medium-sized
animals. So to anthropomorphize,
biological evolution said, wow, this neocortex is a pretty
good design, and it kept growing it in size through
increasingly complex mammal species. By the time it got to primates,
it’s no longer a smooth sheet. It’s got all these convolutions
and ridges to increase its surface area. It’s still a flat structure. If you take the human neocortex,
you can stretch it out into a flat structure the
size of a large table napkin. It’s about the same thickness. It’s still thin. But it has so many convolutions
and ridges, it actually comprises
80% of the brain. And that’s where we do our
hierarchical thinking. So if you take a primate, it
also has one with convolutions and ridges, but the innovation
in homo sapiens is we have this large forehead to squeeze
in more of this neocortex. And that greater quantity was
the enabling factor for the qualitative leap we had of being
able to make inventions like language and art and
science and Nexus phones. [LAUGHTER] RAY KURZWEIL: So how
does this work? Well, for one thing, our ability
to actually see inside the brain and confirm these
types of insights is growing exponentially. Different types of brain
scanning are growing at an exponential rate. We can now see your brain
create your thoughts. We can see your thoughts
create your brain. We can see individual links and
neural connections forming in real time. And another piece of research
that came out just as I was sending off the book is that
at the beginning of life, there is this very uniform
wiring of the neocortex, basically connections
in waiting. So you have one pattern
recognizer, and it wants to connect itself, let’s
say, to one at a higher conceptual level. It has actually connect
a wire. There’s actually a grid there,
like avenues and streets of Manhattan, and it finds the
right avenue and the right street and makes the
final connections. And we actually see that process
in real time now, inside a living brain. And then it actually finalizes
that connection. And then the connections that
are never used die away. About half of the connections
that exist in a newborn actually go away by the time
you’re two years old. So to take a simplified example
of how this works, these pattern recognizers learn
patterns, and there are different levels of the
conceptual hierarchy. And there’s a lot of redundancy,
which is one way it deals with uncertainty and
one way it can deal with variations in patterns. So I have a bunch of pattern
recognizers that have learned to recognize a cross-bar
in a capital A. And that’s all they
care about. Some exciting new technology
or a pretty girl could walk by, it doesn’t care. But when it sees a cross-bar
in a capital A, it goes, whoa, crossbar! [LAUGHTER] And it sends up a signal– I believe this is
not on or off. The whole system is a network
of probabilities. But it says there’s a
high probability we have a crossbar here. At that next higher level, it’s
getting different inputs, and it might then fire with
a high probability– ah, capital A. And at a higher
level, a pattern recognizer might think, hm, there’s a very
good probability that the word “apple” is printed here. And in another part of the
visual cortex, a pattern recognizer might go, oh, an
actual physical apple. And in another region, a pattern
recognizer might go, oh, someone just said the word
“apple.” Go up a number of levels further, where you’re not
getting input at a higher level of conceptual hierarchy,
so it’s connected to multiple senses, it may see a certain
fabric, smell a certain perfume, here a certain voice,
and say, oh, my wife has entered the room. At a much higher level, there
are pattern recognizers that go, oh, that was funny. That was ironic. She’s pretty. Those are actually no more
complicated, except for the fact that they exist at this
very high level of the conceptual hierarchy. I talk about the book this brain
surgery of a young girl. She was conscious, which you
can be in brain surgery, because there’s no pain
receptors in the brain. Whenever they stimulated a
particular point in her neocortex, she would laugh. And they thought they were
triggering a laugh reflex, but they quickly discovered, no,
they’re triggering the perception of humor. She just found everything
hilarious when they triggered that spot. You guys are so funny, standing
there, was her typical comment. But only when they were
triggering that spot. And these guys weren’t funny. [LAUGHTER] They had one spot– and we obviously have
many of them– but they had found one that
would represent the perception of humor. Where does this hierarchy
come from? Well, we’re not born
with it, obviously. That’s what we’re creating from
the moment we’re born, or even before that. I have a one-year-old grandson
now, and he’s laid down several layers. We can lay down, really, one
conceptual layer at a time. And we run through
the 300 million. One of the reasons children can
learn, say, a new language so easily is that they have
all this virgin neocortex. By the time we’re 20, it’s
really filled up. But that doesn’t mean we
can’t learn new things. We have to forget something
to learn something new. We don’t necessarily have to
completely forget it, because there’s a lot of redundancy, and
when we’re first starting to learn something, there’s lots
of redundancy and a lot of the patterns are imperfect. And over time, we can actually
perfect that model and have less redundancy and still
have a good recognition. So we can free up neocortical
recognizers for a new subject. But some people are better
at that than others. I mean, the rigidity that some
people have in learning a new idea is reflected in this
ability or inability to learn new material. Now is 300 million a
lot or a little? It was a lot compared to
other primates, who have somewhat less. And that was the enabling factor
for science and art and music and language and so on. But it’s also a big limitation,
if you recognize the limitations we have in
learning new knowledge. We ultimately will be able
to expand the neocortex. So I’m working now on synthetic
neocortexes, not in the near future to be directly
connected to the brain, but I think if you go out to
the 2030s, we will be able to do that. And we actually don’t have to
put them inside the brain. We just have to put the gateways
to it in the brain. If I do something interesting
on this– do a search, do a language translation, ask
Google Now a question– it doesn’t take place in
this rectangular box. It goes out to the cloud. And if I suddenly need 1,000
processors or 10,000 for a tenth of a second, the cloud
provides that, to the limits of the law of accelerated
returns at that point in time. Ultimately, we’ll be able to
do that with the brain and have more than 300 million
pattern recognizers, that run faster, that can be backed up. And that’s where we’re headed. We’ll have a greater quantity. The last time we added a greater
quantity, we got this qualitative leap of creating
art, science, and language. And we’ll be able to make
another qualitative leap with that expansion. Already, these devices represent
brain-expanders, but we’ll have much more powerful
means of doing that. So just a few comments. Peter will appreciate this. But we are destroying jobs at
the bottom of the skill ladder, creating new
jobs at the top. So we’re investing more
in education. We spend 10 times as much on
K-12 per capita in constant dollars, compared to
a century ago. We had 50,000 college
students in 1870. We have 12 million today. There’s a big revolution coming,
which Peter can tell you about, in higher
education. It’s fostered by this tremendous
boon in both intelligent computation
and communication. We’ve tripled the amount of
education a child gets in the developing world, doubled in the
developed world, over the last half-century. Larry Page and I actually worked
on a major energy study for the National Academy
of Engineering. And the cost of solar energy– both PPV and total installed
costs– are coming down. As a result, the total amount
of solar energy is on exponential climb. It’s doubling every two years. Right now it’s 1%. So people go, oh, 1%, that’s
a fringe player. It’s kind of a nice thing to
do, but it’s not really significant. Just the way that they dismissed
the internet or the Genome Project when
they were 1% of a usable corpus of users. It’s only seven doublings at
two years each from 100%. This was adopted by the
National Academy of Engineering. I presented it recently to the
prime minister of Israel. And he was in my class at the
Sloan School in the ’70s, and he said, Ray, do we have enough sunlight to do this with? And I said yes, we have 10,000
times more than we need. After we double seven
more times, we’ll be using one part in 10,000. So there’s a whole other
discussion about resources in general. We’re running out of resources
if we limit ourselves to 19th-century First Industrial
Revolution technologies like fossil fuels. But in terms of water,
energy, food– with vertical agriculture,
another looming revolution coming over the next decade,
we actually will have a lot of resources. So this is the progress we’ve
made in longevity over the last 1,000 years. We’ve quadrupled life
expectancy. It’s doubled in the
last 200 years. And this is from the
linear progression of health and medicine. It’s now become an information
technology. This’ll go into high gear once
we really master these techniques of biotechnology. There’s many revolutions
coming. But the most important one is
that what’s unique about the human species is that
we have knowledge. And there’s many different ways
to measure knowledge, but no matter how you look at it,
it’s growing exponentially. So we’re doubling the amount
of knowledge, by some measures, at say every
13 months. And that’s actually
what’s hard to do. We have a much better means
already of finding knowledge with Google and other tools. That’s going to get more
and more powerful. But we need that added
intelligence in order to actually continue this
exponential growth of information technology. So Google is still very
well-positioned for fantastic growth in importance and
success over the next several decades. Thank you very much. [APPLAUSE] BORIS DEBIC: Thank you, all. We’ll do a Q&A, and please use
the audience microphone. AUDIENCE: Hi, my
name is Jason. I actually work in PR, so I
think a lot about perceptions of this kind of progress. And I’m thinking about how
people have a tremendous tendency to sort of take for
granted whatever the next progression is, or to sort of
underestimate to correct for whatever improvements
there are. What do you think about that,
the fact that you see, if you measure all these things– and I’m thinking of Steven
Pinker’s work on violence dropping over time as well. People tend to sort of correct
for that and take it for granted, and say, well, dismiss
it at each stage. Do you think that is just
sort of built in to us? RAY KURZWEIL: People have an
amazing ability to accept new changes and then assume
that the world’s always been that way. If you described self-driving
cars a decade ago, people would dismiss that as
science fiction. Now that we have it,
people shrug. Well, it’s not in everybody’s
hands, but actually, I’ve talked to people who’ve driven
in the Google cars that quickly actually gain more
confidence in the AI driver than a human driver. Maybe that’s not saying much. People very quickly then
take it for granted. I travel around the world. I don’t get that here in Silicon
Valley, but as I go to other parts of the world, there
is a common perception that the world’s
getting worse. And a big subset of that school
of thought is that technology’s responsible
for it. I’d like to show them
this graph. So this is the world in 1800. And these are countries. The x-axis is the wealth of
nations, income per person. On the y-axis is life
expectancy. And over the last 200 years,
there’s been dramatic improvement in both. A little bit of movement
in the First Industrial Revolution, but as you get to
the 20th century, there’s a wind that carries all these
nations towards the upper right-hand corner
of the graph. And there’s still a
have/have-not divide, but the countries that are worst off at
the end of the process are still far better off than the
countries that were best-off at the beginning. And I shouldn’t say “end of
the process,” because the process actually is going to go
into high gear as we get to the more mature phases of AI and
three-dimensional printing and biotechnology and so on. But people forget what the world
was like three or four years ago, before we
had social networks and wikis and blogs. And during that SOPA strike,
people were shocked that they could have to do without these
brain extenders which we didn’t have just a
few years ago. So yes, people take changes
for granted. But also, they very readily
adopt them. You describe the world 20, 30
years from now, and people say, well, I don’t know if I
want to opt in for that. It doesn’t happen that way. It happens through thousands of
product announcements and research advances. But when there’s a somewhat
better treatment for cancer, there’s no philosophical
discussion. Is it really a good thing
to extend longevity? People adopt and celebrate
it if it works. So we will continue to make
this kind of progress. I think it’s a moral imperative
that we do. There are downsides. That’s a whole other
discussion. But overall, as you can see,
life is continuing to get better in all the ways that we
can measure– health, wealth, education, so on. AUDIENCE: You mentioned one of
the great innovations of the humans is having a lot more
space up there for neocortex. What about some of our
Earth-mates, like whales? They’ve got a lot more
space up there. RAY KURZWEIL: Right. There are some other animals– actually, the whale
brain is bigger. We have one other enabling
factor, which is this opposable appendage, which
enabled us to take our ideas and our visions and say wow, I
could take that branch and I could strip it of the leaves,
and I could put a point on it, and I could create this tool. And then we had the opposable
appendage to do that. And then we had the tool
to create other tools. And these other species don’t
have that opposable appendage. I mean, we see some clumsy
ability to move things around, say, by an elephant, which
also has a big brain. But it’s actually not clear
that the neocortex, specifically, is bigger
in a whale. But it’s pretty comparable. They don’t have this opposable
appendage that enabled us. So those two things enable
us to create technology. And technology has reshaped
the world. AUDIENCE: But then what about
sort of deep thought, as opposed to just being able
to shape the world? Right? So taking is on a slightly
different vector. RAY KURZWEIL: It depends what
you mean by deep thought. I mean, the fact that we can
develop these greater number of levels of abstraction– the neocortex, in most other
mammals, is really devoted to the challenges of being
a raccoon or whatever. And we’ve been able to actually
then create these abstract levels. So we still have the old brain,
and so the neocortex is a great sublimator. And it can take the sex and
aggression of the old brain and convert it into
poetry and music. And that then becomes
an end in itself. And we’ve really been the only
species to master these additional levels, which you
would consider deep thought. But it’s in an extension of
the neocortical hierarchy. AUDIENCE: It seems pretty clear
that the size of the pie for 3-D printing is growing
significantly, such that, like, I’ve already
made a couple investments in that market. And I’m wondering if based
on your research, you’ve identified any other markets
where you see the size of the pie growing so much, where if
you make a broad play across the industry, that it’s nearly
guaranteed to grow. [LAUGHTER] RAY KURZWEIL: I think search
is very well-positioned. [LAUGHTER] RAY KURZWEIL: Even though it may
seem to be saturated, its role in our lives is not. ‘Cause search is going to become
much more intelligent. Our knowledge bases continue to
expand, and we can really use this as an intelligent
assistant to help guide us, to actually help us solve problems
and be more of an assistant as we make search
more intelligent. And it’s not just the way we
traditionally think of search. It’s this whole world
of knowledge. And Google is very much
committed to knowledge in all of its different forms and in
finding intelligent ways to find that information
and use it. So that’s very well-positioned. Virtual reality is going
to become a big deal. Google has an interest
in that. The project Glass, Google Glass,
will be a first step. But ultimately, I
mean, this is– actually, I like the big screen,
but it’s actually still pretty little. It’s still like looking at the
world through a keyhole. I’ve got this big screen– AUDIENCE: Check out Ingress,
if you haven’t yet. RAY KURZWEIL: Of real reality. And we will be online all the
time, with augmented reality, and just used to looking at
people and having pop-ups tell us who they are. And just telling us their name
will be very useful. That’ll be a killer app. [LAUGHTER] AUDIENCE: Hi. So I had a question. Once we have these pattern
recognizers that we can access remotely, obviously, a best
of breeds will emerge and everyone will want to copy the
best, most accurate, most efficient one. At that point, if I did that,
would I still be me? RAY KURZWEIL: I talk about
that in the book. There are three great
philosophical questions– consciousness, free will,
and identity. And you’re asking about
the identity issue. And I think, in my view,
identity comes from a continuity of pattern. People say, well, no, Ray. You’re this physical stuff. You’re flesh and blood. That’s actually not true. I’m completely different
physical stuff than I was six months ago. And I go through that
in the book. All these different cells
die and are recreated. OK. The neurons persist, but the
parts of the neuron, like the tubules and the actin filaments
and all of these, turn over– some in five hours,
some in five days. And we’re completely different
stuff a few months later. So we’re like a river. Charles River goes
by my office. Is that still the same river
it was yesterday? It’s completely different water,
but the pattern has a continuity, so we call
it the same river. We’re the same thing. Now we can augment that pattern
by, say, introducing non-biological parts to it. And I think it’s very clear if
that’s done in a continuous manner, it’s very analogous to
what’s happening naturally, which is that we’re constantly
changing the stuff and gradually changing the pattern,
but there’s a continuity of pattern,
and that’s the nature of our identity. So I to talk about that
in that chapter. AUDIENCE: Hi. Could you comment on the
progress in the field of nanotechnology since you
wrote “Singularity?” RAY KURZWEIL: What
was the last? AUDIENCE: Could you just comment
on the progress in the field of nanotechnology since
you wrote “The Singularity is Near?” RAY KURZWEIL: Well,
there’s been– nanotechnology is a further-off
revolution than biotechnology. But there have been advances in
our ability to create small structures which being
applied, actually, to electronic devices. And electronics is clearly
nanotechnology. The feature sizes are
approaching 20 nanometers, which is like 100
carbon atoms. We’re starting to build
three-dimensional structures. So there’s definitely been a
lot of technology there. MEMS, there are MEMS devices
now that are under 100 nanometers, ’cause it’s using
the same technology as semiconductors. There are experiments with
devices in the human body. There are dozens of
experiments of blood-cell-sized devices that
are nanoengineered doing therapeutic interventions
in animals. I think that’s a further
evolution than the biotech. Biotech is really here. It’s kind of on the experimental
cutting edge. Like if you want to fix your
heart if you’ve had a heart attack, you actually can’t. It’s not FDA-approved. It will be soon, but right
now you have to go to Israel or Thailand. So it’s kind of on the edge, but
it’s very close at hand. Nanotechnology is still, I
think, late 2020s for those types of applications. AUDIENCE: I hope this doesn’t
come across as a flaky question, but– RAY KURZWEIL: No question
is flaky. AUDIENCE: In your research, have
you found the same law of accelerating returns in
happiness, fulfillment, satisfaction? RAY KURZWEIL: Well, this is
actually a similar question to the first one, in that
our expectations are constantly changing. If you talk to a caveman or
woman thousands of years ago, they would say, gee, if I
could just have a bigger boulder to keep the animals
out of my cave and prevent this fire from going out,
I would be happy. Well, don’t you want
a better website? [LAUGHTER] RAY KURZWEIL: So we don’t even
know what we want until somebody invents these ideas. And our expectations
of what should be are constantly changing. People who are poor today still
generally have access to refrigerators and to
communications and clothing. You go back several hundred
years ago, even a middle-class person only had one shirt before
there was automation in the textile industry. So our expectations of what it
takes to be happy change. I think people are happier,
because a much higher percentage of the population
gets part of their satisfaction and definition
in life from their work. Not everybody, apparently. I was interested by this French
strike where they were very upset at extending the
retirement age from 60 to 62. And I thought, gee, these people
really must not like their work. But then I realized that I had
retired when I was five, because I’m really doing
what I love to do. And I think that should be
the objective of work. And many more people have the
opportunity to do that. Work done in the information
sector, people really have a passion for it, whereas 100
years ago, they were just glad if they could earn a living. But it’s a moving frontier. And I think that’s a good thing,
and that’s part of what propels humanity forward,
is we’re constantly questing for more. And more doesn’t necessarily
mean greater quantity of physical things. It could be just more music and
more opportunity to have relationships, which social
networks gives us the opportunity to do, and so on. AUDIENCE: So with the increase
in knowledge work, it requires a lot of knowledge transfer
between humans. Do you envision any efficient
methods of knowledge transfer within humans beyond? RAY KURZWEIL: Could you
speak a little louder? I’m missing some words. AUDIENCE: Do you envision any
efficient methods of knowledge transfer between humans? Not like reading books or
anything, just beaming. RAY KURZWEIL: Yeah, well, when
we can have massively distributed communication points
in a neocortex, it could provide a higher-bandwidth
way of communicating. But we have to appreciate that
there’s actually a very kind of challenging translation job
for one neocortex to another. I talk about this in the book. If you could actually get this
information at any bandwidth, and even process it quickly, of
someone else’s neocortex, you’d have no idea what it
means, because that pattern recognizer, say, fires with
a higher probability. But you can only interpret that
based on the ones that are connected to it. And each of those, you go only
understand by the ones connected to it, all the
way down the hierarchy. You’d have to actually have a
complete dump of most of their neocortex to understand it. And so just– it’s not like we would readily
understand someone else’s neocortex, even if you could
transfer that information without translating it. We have a translation mechanism,
which is language. So we could take thoughts from
one neocortex, even though it’s very different from someone
else’s, because we’ve each built this hierarchy, and
actually communicate a thought that the other person
can understand. That’s what language
enables us to do. We could perhaps do some
automatic translation, just like we translate languages
now, from one neocortex to another and provide
higher-bandwidth connection. I mean, it’s something we could
speculate once we’re able to do that in the 2040s. AUDIENCE: Excuse me, if you’ve
already covered this. I was way in the back, and it
was a little hard to hear you, but do we have software
engineering stuff to model these clusters of neurons and
create these models already? RAY KURZWEIL: Well, the closest
that we’ve had is these hierarchical hidden Markov
models, which as I mentioned, have become a
common technique in AI. They’re missing certain things,
in that generally the hierarchy is fixed. So I mean, I began pioneering
this in the ’80s, and we did it for speech recognition, and
then we added simple natural language understanding and we
had some fixed levels of spectral features, phonemes,
words, and then simple syntactic structures. But it was relatively fixed. It could prune some elements,
some of these recognizers, if they weren’t used. But it didn’t actually
self-organize, in terms of creating the connections, which
is really the essence of what the neocortex does. If you want to get into a
better level of natural language understanding, you need
to be able to do that, because one of the features of
language is that it doesn’t just have two or three fixed
levels of hierarchy. Language reflects the hierarchy
of the neocortex. It can have many different
levels. And you really need to model
quite a few levels in order to make semantic sense
of language. And we need to be able to dynamically build that hierarchy. But it’s interesting, actually,
that I think there’s a mathematical similarity
between this hierarchical hidden Markov model technique
and what happens in the brain. And it’s not because we were
trying to emulate the brain in the ’80s and ’90s, because we
didn’t really understand– we didn’t have enough
information to confirm that that’s how the brain works. It’s just that technique
worked, and biological evolution evolved neocortexes
that way for the same reason. AUDIENCE: Speaking of assuming
that the world will not change a lot, I’d like you to comment
on the non-technical aspect of this change. We all assume that 20 years from
now we’ll be living in a stable democracy, with
free market and a capitalist economy. Those changes that you predict,
how much of that are they going to change,
politically and economically? RAY KURZWEIL: Well, I do think
the distributed communication technologies we have
is democratizing. I wrote that in the 1980s, and
then it was discussed in my first book, which I
wrote in the ’80s. I said the Soviet Union would
be swept away by the then-emerging social network,
which was communication over Teletype machines and fax
machines, by this clandestine network of hackers. And so people heavily
criticized that. At that time, the Soviet Union
was a mighty nuclear superpower. It’s not going to get
swept away by a few Teletype machines. But that’s exactly what happened
in the 1991 coup against Gorbachev. The authorities grabbed the
central TV and radio station, which had always worked in the
past, ’cause it kept everybody in the dark. But now this clandestine
network, this sort of first social network, kept everybody
in the know. And it just swept away the
totalitarian government. And with the rise of the web,
there was a great wave of democratization in
the late ’90s. We see the effect of social
networks today. It is democratizing for people
to share knowledge at that grassroots level, see how other
people live and think. It really is able to harness
the wisdom of crowds rather than the wisdom of
a lynch mob. And we’ve also democratized
the tools of creativity. So a kid with a notebook
computer could start Facebook. And a couple of kids in a
late-night dorm room challenge started Google. And we see now, younger kids
doing quite dramatic things, teenagers with tools that
everybody has, a kid in Africa with a smartphone has access
to more knowledge than the president of the United States
did 15 years ago. So these are having an impact
on our economy, on society. Here’s a very dramatic
demonstration of the political power of this organized group
of people who are able to communicate. The SOPA legislation was headed
for bilateral passage. Both Democrats and Republicans
were for it. It was going to be passed, one
of the few examples where there was agreement on a
piece of legislation. Well, users saw that as a threat
to the freedom on the web and organized this
demonstration. Within hours, it was dead. So I mean, just think of the
tremendous political power that was demonstrated there. Google participated in that, but
suddenly Wikipedia becomes a great political power. It just snaps its
fingers, and. So I think that these are
very positive phenomena. And it’s affecting society. It’s affecting communication. People criticize online
education now because it’s missing a social component that
you have with a campus. But we can actually do a
better job with social networks and social
communication online, because we overcome the geographic
barrier. AUDIENCE: I’m struggling to find
the exact words– sorry. But I wanted to ask you whether
you see power– not as in electronic power, but
power as in control over individuals– as something that’s
exponentially accelerating, in terms of the state or security
apparatus versus freedom. It seems like both are
accelerating quite quickly, and there’s this tension between
the power that’s being centralized versus of
the individual. RAY KURZWEIL: Well you
can I imagine– these tools can be used
to spy and wreck privacy, invade privacy. The recent scandal going on in
Washington raises issues of the privacy of emails
and so on. On the other hand, I think
it’s also been very democratizing, as I mentioned. I think it’s led to
greater freedom. I think that trend has been more
pronounced, the ability of individuals to organize
around a set of ideas that they quickly support in
terms of freedom. And we’ve seen the democratizing
effect of decentralized electronic
communication. Privacy is a very
important issue. There’s certainly an important
issue here. I think Google does a good job
of it, but it’s something that has to be a high priority. If any service like Facebook
or something did not keep faith with its users, in terms
of these social issues, there would be a reaction. And it raises complicated
issues. Like privacy, it used to be
enough to just close the curtains in your bedroom, and
now we have 1,000 virtual windows on our lives. Nonetheless, I think we’re
doing pretty well. I almost never encounter someone
who says, oh, my life was ruined by the loss of
privacy because of all these new technologies. Now, maybe those people
don’t talk to me. But I think we’re doing OK. But it is making these
once-routine issues much more complicated. AUDIENCE: So when you were
talking about the digitization of or the information age of
manufacturing with printers, 3-D printers, I had a question
about resources. Like if you print with, like,
hydrocarbons, for example, then you might need an oil rig
and a ship and a truck to get the resources from the Earth
into the printer, and that takes a lot of time
and a lot of fuel. Whereas if you build with
plants, then you need to farm somewhere, and again, you need
transport to where the printers are. So how do you see
things changing? RAY KURZWEIL: There’s not that
many resources you need to create these physical things. By far the most hydrocarbons are
used in burning them for fossil fuels. Yes, some of those products
are used now in chips, for example. But it’s a very small
part of the output. And if we can actually create
the right products at the destination in a distributed
manner, and then also recycle those these materials, that’s
a pretty efficient use of these materials. Peter Diamandis has a book
called “Abundance” that deals with, in detail, this issue
of energy, these kinds of resources for three-dimensional
printing– water, food, building
materials. And as we adopt new
technologies, we actually find that there’s a tremendous
abundance of resources, like 10,000 times more sunlight than
we need to meet all our energy needs. Larry Page was fond of going
a mile or two that way, and there’s a lot of heat in the
Earth, geothermal energy, which is also thousands of
times more than we need. And there are a lot of
other scenarios. So as we find new 21st-century
technologies, we can tap these resources. There’s new water technologies,
like Kamen’s Slingshot machine, which are
decentralized and can create clean water very inexpensively,
vertical agriculture to grow food in
AI-controlled buildings, recycling all the nutrients so
in fact it would not be the wasteful and
ecologically-damaging food-production techniques we
use now, but we can create food very inexpensively. Including in vitro-cloned
meat– I mean, why grow meat from
animals when we only need a small part of the animal? We know how to, in fact, grow
the muscle tissue, which is what we want. It’s been demonstrated. This can be done in
AI-controlled buildings at very low cost, ultimately. AUDIENCE: But do you think that,
say, a computer will be able to be printed with
resources that were sourced locally? RAY KURZWEIL: There’s actually
some experimental three-dimensional printing
systems that can print electronics. Being able to actually print
electronics in a distributed matter, there are pros
and cons to it. An argument can made– computation and communication
is very universal, so let’s have plants that really do
that efficiently and then customize it for people
with software. That’s the model we’re
using now. I mean, it’s remarkable how
powerful a computational communication device you can
get for very little money. And that’s continuing
to improve. BORIS DEBIC: I hope you all
made some new neocortical connections today which will be
useful in your work and in your lives. And please join me in thanking
Dr. Kurzweil. [APPLAUSE]

12 thoughts on “Ray Kurzweil: “How to Create a Mind” | Talks at Google

  1. Historically interesting that he only speaks about HMM's as being state of the art in modelling the hierarchical structure of the neocortex. 2012 was right on the turning point when deep learning (ANN's) really took off due to increase in compute (GPU's).

  2. I've just watched Malcolm Gladwell's google talk right before this one (https://youtu.be/5L0GGfQblrc , highly recommendable)
    Apart from talking about different topics, they could hardly be more different.
    Kurzweil's argument is: "Trust me, I've been right before: here, here and here."
    Gladwell's argument is: "This is what scientists found. This is what I think. Please don't blindly agree with me."
    I'm a scientist. I know which presentation style I prefer.

  3. Just ask him: "Ray-gun, will 5G have a negative impact on human health or not? Psychosomatically and/or physiologically? Please reply under the assumption that you would take personal responsibility for your answer…."
    "I…I…B…But… You… Well… I … I mean YOU…. must undestand that….. uh…."
    "Just a ´yes´ or ´no´ will suffice pizza face…."
    One thing these little devils do know is "price performance." Inside and out.

  4. 31:11 Already true. And scary as hell. For those who have the data a blessing. For those produce the data and never got them to see or only a fraction of it, it's terrible.

  5. Sounds like the image of the beast
    So all info – full pride and wickedness – days noah
    Are we retarded? This is literally the pathway to the Image of the beast
    Its already here =- a Narcassist holocaust….
    Im gangstalked…I know.

Leave a Reply

Your email address will not be published. Required fields are marked *