WEBVTT

00:00:00.760 --> 00:00:04.600
When I was a kid,
I was the quintessential nerd.

00:00:05.320 --> 00:00:07.496
I think some of you were too.

00:00:07.520 --> 00:00:08.736
(Laughter)

00:00:08.760 --> 00:00:11.976
And you, sir, who laughed the loudest,
you probably still are.

00:00:12.000 --> 00:00:14.256
(Laughter)

00:00:14.280 --> 00:00:17.776
I grew up in a small town
in the dusty plains of north Texas,

00:00:17.800 --> 00:00:21.136
the son of a sheriff
who was the son of a pastor.

00:00:21.160 --> 00:00:23.080
Getting into trouble was not an option.

00:00:24.040 --> 00:00:27.296
And so I started reading
calculus books for fun.

00:00:27.320 --> 00:00:28.856
(Laughter)

00:00:28.880 --> 00:00:30.576
You did too.

00:00:30.600 --> 00:00:34.336
That led me to building a laser
and a computer and model rockets,

00:00:34.360 --> 00:00:37.360
and that led me to making
rocket fuel in my bedroom.

00:00:37.960 --> 00:00:41.616
Now, in scientific terms,

00:00:41.640 --> 00:00:44.896
we call this a very bad idea.

00:00:44.920 --> 00:00:46.136
(Laughter)

00:00:46.160 --> 00:00:48.336
Around that same time,

00:00:48.360 --> 00:00:51.576
Stanley Kubrick's "2001: A Space Odyssey"
came to the theaters,

00:00:51.600 --> 00:00:53.800
and my life was forever changed.

00:00:54.280 --> 00:00:56.336
I loved everything about that movie,

00:00:56.360 --> 00:00:58.896
especially the HAL 9000.

00:00:58.920 --> 00:01:00.976
Now, HAL was a sentient computer

00:01:01.000 --> 00:01:03.456
designed to guide the Discovery spacecraft

00:01:03.480 --> 00:01:06.016
from the Earth to Jupiter.

00:01:06.040 --> 00:01:08.096
HAL was also a flawed character,

00:01:08.120 --> 00:01:12.400
for in the end he chose
to value the mission over human life.

00:01:12.840 --> 00:01:14.936
Now, HAL was a fictional character,

00:01:14.960 --> 00:01:17.616
but nonetheless he speaks to our fears,

00:01:17.640 --> 00:01:19.736
our fears of being subjugated

00:01:19.760 --> 00:01:22.776
by some unfeeling, artificial intelligence

00:01:22.800 --> 00:01:24.760
who is indifferent to our humanity.

00:01:25.880 --> 00:01:28.456
I believe that such fears are unfounded.

00:01:28.480 --> 00:01:31.176
Indeed, we stand at a remarkable time

00:01:31.200 --> 00:01:32.736
in human history,

00:01:32.760 --> 00:01:37.736
where, driven by refusal to accept
the limits of our bodies and our minds,

00:01:37.760 --> 00:01:39.456
we are building machines

00:01:39.480 --> 00:01:43.096
of exquisite, beautiful
complexity and grace

00:01:43.120 --> 00:01:45.176
that will extend the human experience

00:01:45.200 --> 00:01:46.880
in ways beyond our imagining.

00:01:47.720 --> 00:01:50.296
After a career that led me
from the Air Force Academy

00:01:50.320 --> 00:01:52.256
to Space Command to now,

00:01:52.280 --> 00:01:53.976
I became a systems engineer,

00:01:54.000 --> 00:01:56.736
and recently I was drawn
into an engineering problem

00:01:56.760 --> 00:01:59.336
associated with NASA's mission to Mars.

00:01:59.360 --> 00:02:01.856
Now, in space flights to the Moon,

00:02:01.880 --> 00:02:05.016
we can rely upon
mission control in Houston

00:02:05.040 --> 00:02:07.016
to watch over all aspects of the flight.

00:02:07.040 --> 00:02:10.576
However, Mars is 200 times further away,

00:02:10.600 --> 00:02:13.816
and as a result it takes
on average 13 minutes

00:02:13.840 --> 00:02:16.976
for a signal to travel
from the Earth to Mars.

00:02:17.000 --> 00:02:20.400
If there's trouble,
there's not enough time.

00:02:20.840 --> 00:02:23.336
And so a reasonable engineering solution

00:02:23.360 --> 00:02:25.936
calls for us to put mission control

00:02:25.960 --> 00:02:28.976
inside the walls of the Orion spacecraft.

00:02:29.000 --> 00:02:31.896
Another fascinating idea
in the mission profile

00:02:31.920 --> 00:02:34.816
places humanoid robots
on the surface of Mars

00:02:34.840 --> 00:02:36.696
before the humans themselves arrive,

00:02:36.720 --> 00:02:38.376
first to build facilities

00:02:38.400 --> 00:02:41.760
and later to serve as collaborative
members of the science team.

00:02:43.400 --> 00:02:46.136
Now, as I looked at this
from an engineering perspective,

00:02:46.160 --> 00:02:49.336
it became very clear to me
that what I needed to architect

00:02:49.360 --> 00:02:51.536
was a smart, collaborative,

00:02:51.560 --> 00:02:53.936
socially intelligent
artificial intelligence.

00:02:53.960 --> 00:02:58.256
In other words, I needed to build
something very much like a HAL

00:02:58.280 --> 00:03:00.696
but without the homicidal tendencies.

00:03:00.720 --> 00:03:02.080
(Laughter)

00:03:02.920 --> 00:03:04.736
Let's pause for a moment.

00:03:04.760 --> 00:03:08.656
Is it really possible to build
an artificial intelligence like that?

00:03:08.680 --> 00:03:10.136
Actually, it is.

00:03:10.160 --> 00:03:11.416
In many ways,

00:03:11.440 --> 00:03:13.416
this is a hard engineering problem

00:03:13.440 --> 00:03:14.896
with elements of AI,

00:03:14.920 --> 00:03:19.616
not some wet hair ball of an AI problem
that needs to be engineered.

00:03:19.640 --> 00:03:22.296
To paraphrase Alan Turing,

00:03:22.320 --> 00:03:24.696
I'm not interested
in building a sentient machine.

00:03:24.720 --> 00:03:26.296
I'm not building a HAL.

00:03:26.320 --> 00:03:28.736
All I'm after is a simple brain,

00:03:28.760 --> 00:03:31.880
something that offers
the illusion of intelligence.

00:03:33.000 --> 00:03:36.136
The art and the science of computing
have come a long way

00:03:36.160 --> 00:03:37.656
since HAL was onscreen,

00:03:37.680 --> 00:03:40.896
and I'd imagine if his inventor
Dr. Chandra were here today,

00:03:40.920 --> 00:03:43.256
he'd have a whole lot of questions for us.

00:03:43.280 --> 00:03:45.376
Is it really possible for us

00:03:45.400 --> 00:03:49.416
to take a system of millions
upon millions of devices,

00:03:49.440 --> 00:03:50.896
to read in their data streams,

00:03:50.920 --> 00:03:53.176
to predict their failures
and act in advance?

00:03:53.200 --> 00:03:54.416
Yes.

00:03:54.440 --> 00:03:57.616
Can we build systems that converse
with humans in natural language?

00:03:57.640 --> 00:03:58.856
Yes.

00:03:58.880 --> 00:04:01.856
Can we build systems
that recognize objects, identify emotions,

00:04:01.880 --> 00:04:05.256
emote themselves,
play games, and even read lips?

00:04:05.280 --> 00:04:06.496
Yes.

00:04:06.520 --> 00:04:08.656
Can we build a system that sets goals,

00:04:08.680 --> 00:04:12.296
that carries out plans against those goals
and learns along the way?

00:04:12.320 --> 00:04:13.536
Yes.

00:04:13.560 --> 00:04:16.896
Can we build systems
that have a theory of mind?

00:04:16.920 --> 00:04:18.416
This we are learning to do.

00:04:18.440 --> 00:04:21.920
Can we build systems that have
an ethical and moral foundation?

00:04:22.480 --> 00:04:24.520
This we must learn how to do.

00:04:25.360 --> 00:04:26.736
So let's accept for a moment

00:04:26.760 --> 00:04:29.656
that it's possible to build
such an artificial intelligence

00:04:29.680 --> 00:04:31.816
for this kind of mission and others.

00:04:31.840 --> 00:04:34.376
The next question
you must ask yourself is,

00:04:34.400 --> 00:04:35.856
should we fear it?

00:04:35.880 --> 00:04:37.856
Now, every new technology

00:04:37.880 --> 00:04:40.776
brings with it
some measure of trepidation.

00:04:40.800 --> 00:04:42.496
When we first saw cars,

00:04:42.520 --> 00:04:46.536
people lamented that we would see
the destruction of the family.

00:04:46.560 --> 00:04:49.256
When we first saw telephones come in,

00:04:49.280 --> 00:04:52.176
people would worry it would destroy
all civil conversation.

00:04:52.200 --> 00:04:56.136
At a point in time we saw
the written word become pervasive,

00:04:56.160 --> 00:04:58.656
people thought we would lose
our ability to memorize.

00:04:58.680 --> 00:05:00.736
These things are all true to a degree,

00:05:00.760 --> 00:05:03.176
but it's also the case
that these technologies

00:05:03.200 --> 00:05:06.576
brought to us things
that extended the human experience

00:05:06.600 --> 00:05:08.480
in some profound ways.

00:05:09.840 --> 00:05:12.120
So let's take this a little further.

00:05:13.120 --> 00:05:17.856
I do not fear the creation
of an AI like this,

00:05:17.880 --> 00:05:21.696
because it will eventually
embody some of our values.

00:05:21.720 --> 00:05:25.216
Consider this: building a cognitive system
is fundamentally different

00:05:25.240 --> 00:05:28.536
than building a traditional
software-intensive system of the past.

00:05:28.560 --> 00:05:31.016
We don't program them. We teach them.

00:05:31.040 --> 00:05:33.696
In order to teach a system
how to recognize flowers,

00:05:33.720 --> 00:05:36.736
I show it thousands of flowers
of the kinds I like.

00:05:36.760 --> 00:05:39.016
In order to teach a system
how to play a game --

00:05:39.040 --> 00:05:41.000
Well, I would. You would too.

00:05:42.600 --> 00:05:44.640
I like flowers. Come on.

00:05:45.440 --> 00:05:48.296
To teach a system
how to play a game like go,

00:05:48.320 --> 00:05:50.376
I'd have it play thousands of games of go,

00:05:50.400 --> 00:05:52.056
but in the process I also teach it

00:05:52.080 --> 00:05:54.496
how to discern
a good game from a bad game.

00:05:54.520 --> 00:05:58.216
If I want to create an artificially
intelligent legal assistant,

00:05:58.240 --> 00:06:00.016
I will teach it some corpus of law

00:06:00.040 --> 00:06:02.896
but at the same time I am fusing with it

00:06:02.920 --> 00:06:05.800
the sense of mercy and justice
that is part of that law.

00:06:06.560 --> 00:06:09.536
In scientific terms,
this is what we call ground truth,

00:06:09.560 --> 00:06:11.576
and here's the important point:

00:06:11.600 --> 00:06:13.056
in producing these machines,

00:06:13.080 --> 00:06:16.496
we are therefore teaching them
a sense of our values.

00:06:16.520 --> 00:06:19.656
To that end, I trust
in artificial intelligence

00:06:19.680 --> 00:06:23.320
the same, if not more,
as a human who is well-trained.

00:06:24.080 --> 00:06:25.296
But, you may ask,

00:06:25.320 --> 00:06:27.936
what about rogue agents,

00:06:27.960 --> 00:06:31.296
some well-funded
nongovernment organization?

00:06:31.320 --> 00:06:35.136
I do not fear an artificial intelligence
in the hand of a lone wolf.

00:06:35.160 --> 00:06:39.696
Clearly, we cannot protect ourselves
against all random acts of violence,

00:06:39.720 --> 00:06:41.856
but the reality is such a system

00:06:41.880 --> 00:06:44.976
requires substantial training
and subtle training

00:06:45.000 --> 00:06:47.296
far beyond the resources of an individual.

00:06:47.320 --> 00:06:48.536
And furthermore,

00:06:48.560 --> 00:06:51.816
it's far more than just injecting
an Internet virus to the world,

00:06:51.840 --> 00:06:54.936
where you push a button,
all of a sudden it's in a million places

00:06:54.960 --> 00:06:57.416
and laptops start blowing up
all over the place.

00:06:57.440 --> 00:07:00.256
Now, these kinds of substances
are much larger

00:07:00.280 --> 00:07:01.995
and we'll certainly see them coming.

00:07:02.520 --> 00:07:05.576
Do I fear that such
an artificial intelligence

00:07:05.600 --> 00:07:07.560
might threaten all of humanity?

00:07:08.280 --> 00:07:12.656
If you look at movies
such as "The Matrix," "Metropolis,"

00:07:12.680 --> 00:07:15.856
"The Terminator,"
shows such as "Westworld,"

00:07:15.880 --> 00:07:18.016
they all speak of this kind of fear.

00:07:18.040 --> 00:07:22.336
Indeed, in the book "Superintelligence"
by the philosopher Nick Bostrom,

00:07:22.360 --> 00:07:23.896
he picks up on this theme

00:07:23.920 --> 00:07:27.936
and observes that a superintelligence
might not only be dangerous,

00:07:27.960 --> 00:07:31.816
it could represent an existential threat
to all of humanity.

00:07:31.840 --> 00:07:34.056
Dr. Bostrom's basic argument

00:07:34.080 --> 00:07:36.816
is that such systems will eventually

00:07:36.840 --> 00:07:40.096
have such an insatiable
thirst for information

00:07:40.120 --> 00:07:43.016
that they will perhaps learn how to learn

00:07:43.040 --> 00:07:45.656
and eventually discover
that they may have goals

00:07:45.680 --> 00:07:47.976
that are contrary to human needs.

00:07:48.000 --> 00:07:49.856
Dr. Bostrom has a number of followers.

00:07:49.880 --> 00:07:54.200
He is supported by people
such as Elon Musk and Stephen Hawking.

00:07:54.880 --> 00:07:57.280
With all due respect

00:07:58.160 --> 00:08:00.176
to these brilliant minds,

00:08:00.200 --> 00:08:02.456
I believe that they
are fundamentally wrong.

00:08:02.480 --> 00:08:05.656
Now, there are a lot of pieces
of Dr. Bostrom's argument to unpack,

00:08:05.680 --> 00:08:07.816
and I don't have time to unpack them all,

00:08:07.840 --> 00:08:10.536
but very briefly, consider this:

00:08:10.560 --> 00:08:14.296
super knowing is very different
than super doing.

00:08:14.320 --> 00:08:16.216
HAL was a threat to the Discovery crew

00:08:16.240 --> 00:08:20.656
only insofar as HAL commanded
all aspects of the Discovery.

00:08:20.680 --> 00:08:23.176
So it would have to be
with a superintelligence.

00:08:23.200 --> 00:08:25.696
It would have to have dominion
over all of our world.

00:08:25.720 --> 00:08:28.536
This is the stuff of Skynet
from the movie "The Terminator"

00:08:28.560 --> 00:08:30.416
in which we had a superintelligence

00:08:30.440 --> 00:08:31.816
that commanded human will,

00:08:31.840 --> 00:08:35.696
that directed every device
that was in every corner of the world.

00:08:35.720 --> 00:08:37.176
Practically speaking,

00:08:37.200 --> 00:08:39.296
it ain't gonna happen.

00:08:39.320 --> 00:08:42.376
We are not building AIs
that control the weather,

00:08:42.400 --> 00:08:43.736
that direct the tides,

00:08:43.760 --> 00:08:47.136
that command us
capricious, chaotic humans.

00:08:47.160 --> 00:08:51.056
And furthermore, if such
an artificial intelligence existed,

00:08:51.080 --> 00:08:54.016
it would have to compete
with human economies,

00:08:54.040 --> 00:08:56.560
and thereby compete for resources with us.

00:08:57.200 --> 00:08:58.416
And in the end --

00:08:58.440 --> 00:08:59.680
don't tell Siri this --

00:09:00.440 --> 00:09:01.816
we can always unplug them.

00:09:01.840 --> 00:09:03.960
(Laughter)

00:09:05.360 --> 00:09:07.816
We are on an incredible journey

00:09:07.840 --> 00:09:10.336
of coevolution with our machines.

00:09:10.360 --> 00:09:12.856
The humans we are today

00:09:12.880 --> 00:09:15.416
are not the humans we will be then.

00:09:15.440 --> 00:09:18.576
To worry now about the rise
of a superintelligence

00:09:18.600 --> 00:09:21.656
is in many ways a dangerous distraction

00:09:21.680 --> 00:09:24.016
because the rise of computing itself

00:09:24.040 --> 00:09:27.056
brings to us a number
of human and societal issues

00:09:27.080 --> 00:09:28.720
to which we must now attend.

00:09:29.360 --> 00:09:32.176
How shall I best organize society

00:09:32.200 --> 00:09:34.536
when the need for human labor diminishes?

00:09:34.560 --> 00:09:38.376
How can I bring understanding
and education throughout the globe

00:09:38.400 --> 00:09:40.176
and still respect our differences?

00:09:40.200 --> 00:09:44.456
How might I extend and enhance human life
through cognitive healthcare?

00:09:44.480 --> 00:09:47.336
How might I use computing

00:09:47.360 --> 00:09:49.120
to help take us to the stars?

00:09:49.760 --> 00:09:51.800
And that's the exciting thing.

00:09:52.400 --> 00:09:54.736
The opportunities to use computing

00:09:54.760 --> 00:09:56.296
to advance the human experience

00:09:56.320 --> 00:09:57.736
are within our reach,

00:09:57.760 --> 00:09:59.616
here and now,

00:09:59.640 --> 00:10:01.320
and we are just beginning.

00:10:02.280 --> 00:10:03.496
Thank you very much.

00:10:03.520 --> 00:10:05.720
(Applause)