WEBVTT

00:00:00.000 --> 00:00:02.077
This episode is brought to you by Pique

00:00:02.160 --> 00:00:05.637
Tea. That's p i q u e. I have had so

00:00:05.720 --> 00:00:08.117
much tea in my life. I've been to China.

00:00:08.200 --> 00:00:10.117
I've lived in China and Japan. I've done

00:00:10.200 --> 00:00:13.197
tea tours. I drink a lot of tea. And 10

00:00:13.280 --> 00:00:17.957
years plus of physical experimentation
and tracking has shown me many things,

00:00:18.040 --> 00:00:21.437
chief among them that gut health
is critical to just about everything and

00:00:21.520 --> 00:00:24.837
you'll see where tea is going to tie
into this. It affects immune function,

00:00:24.920 --> 00:00:28.157
weight management, mental performance,
emotional health, you name it. I've been

00:00:28.240 --> 00:00:32.517
drinking fermented pu-erh tea
specifically pretty much every day for

00:00:32.600 --> 00:00:36.717
years now. Pu-erh tea delivers more
polyphenols and probiotics than you can

00:00:36.800 --> 00:00:39.957
shake a stick at. It's like providing
the optimal fertilizer to your

00:00:40.040 --> 00:00:43.317
microbiome. The problem with good pu-erh

00:00:43.400 --> 00:00:44.837
is that it's hard to source. It's hard

00:00:44.920 --> 00:00:50.437
to find real pu-erh that hasn't been
exposed to pesticides and other nasties,

00:00:50.520 --> 00:00:54.317
which is super common. That's why
Pique's fermented pu-erh tea crystals

00:00:54.400 --> 00:00:58.957
have become my daily go-to. It's so
simple. They have so many benefits that

00:00:59.040 --> 00:01:01.837
I'm going to get into and I first
learned about them through my friends

00:01:01.920 --> 00:01:04.357
Dr. Peter Attia and Kevin Rose. Pique

00:01:04.440 --> 00:01:06.917
crystals are cold extracted using only

00:01:07.000 --> 00:01:11.917
wild harvested leaves from 250-year-old
tea trees. I often kickstart my mornings

00:01:12.000 --> 00:01:15.317
with their pu-erh green tea,
their pu-erh black tea, and I alternate

00:01:15.400 --> 00:01:19.397
between the two. The rich earthy flavor
of the black specifically is amazing.

00:01:19.480 --> 00:01:23.317
It's very very It's like
a a a delicious barnyard.

00:01:23.400 --> 00:01:25.727
Very peaty if you like
whiskey and stuff like that.

00:01:25.810 --> 00:01:28.197
>> [snorts] >> They triple
toxin screen all of their

00:01:28.280 --> 00:01:32.157
products for heavy metals, pesticides,
and toxic mold contaminants commonly

00:01:32.240 --> 00:01:36.877
found in tea. There's also zero prep
or brewing required as the crystals

00:01:36.960 --> 00:01:41.117
dissolve in seconds. So, you can just
drop it in your hot tea or I also make

00:01:41.200 --> 00:01:46.157
iced tea and that saves a ton of time
and hassle. So, Pique is offering 15%

00:01:46.240 --> 00:01:50.357
off their pu-erh teas for the very first
time exclusive to you, my listeners.

00:01:50.440 --> 00:01:52.277
This is a sweet offer. Simply visit

00:01:52.360 --> 00:01:53.997
piquetea.com/tim.

00:01:54.080 --> 00:01:58.157
That's p i q u e t e a dot com {forward

00:01:58.240 --> 00:02:02.437
slash} Tim. This promotion is only
available to listeners of this podcast.

00:02:02.520 --> 00:02:06.317
That's peak tea dot com {forward slash}
Tim. The discount is automatically

00:02:06.400 --> 00:02:10.237
applied when you use that URL. You also
have a 30-day satisfaction guarantee, so

00:02:10.320 --> 00:02:11.997
your purchase is risk-free. One more

00:02:12.080 --> 00:02:14.677
time, check it out. Peak tea, that's P I

00:02:14.760 --> 00:02:19.960
Q U E T E A dot com {forward slash} Tim.

00:02:21.720 --> 00:02:25.597
This episode is brought to you by
ButcherBox. ButcherBox makes it easy for

00:02:25.680 --> 00:02:29.837
you to get high-quality, humanely raised
meat that you can trust. They deliver

00:02:29.920 --> 00:02:34.317
delicious, 100% grass-fed,
grass-finished beef, free-range organic

00:02:34.400 --> 00:02:39.317
chicken, heritage breed pork, and
wild-caught seafood directly to your door.

00:02:39.400 --> 00:02:42.877
For me, in the past few weeks, I've
cooked a ton of their salmon, as well as

00:02:42.960 --> 00:02:46.677
two delicious barbecue rib racks in the
oven. Super simple. They were the most

00:02:46.760 --> 00:02:51.637
delicious pork ribs I've ever prepared,
and my freezer is full of ButcherBox.

00:02:51.720 --> 00:02:54.597
When you become a member, you're joining
a community focused on doing what's

00:02:54.680 --> 00:02:58.157
better for all. That means caring about
the lives of animals, the livelihoods of

00:02:58.240 --> 00:03:01.677
farmers, treating our planet with
respect, and enjoying better meals

00:03:01.760 --> 00:03:05.117
together. ButcherBox partners with folks,
small farmers included, who

00:03:05.200 --> 00:03:08.797
believe in going above and beyond when
it comes to caring for animals, the

00:03:08.880 --> 00:03:11.957
environment, and sustainability.
And none of their meat is ever given

00:03:12.040 --> 00:03:15.597
antibiotics or added hormones. So,
how does it work? It's pretty simple. You

00:03:15.680 --> 00:03:20.517
choose your box and your delivery
frequency. They offer five boxes, four

00:03:20.600 --> 00:03:24.157
curated box options, as well as the
popular custom box. So, you get exactly

00:03:24.240 --> 00:03:28.917
what you and or your family love. Box
options and delivery frequencies can be

00:03:29.000 --> 00:03:33.197
customized to fit your needs. You can
cancel at any time with no penalty.

00:03:33.280 --> 00:03:35.197
ButcherBox ships your order frozen for

00:03:35.280 --> 00:03:37.077
freshness and packed in an eco-friendly,

00:03:37.160 --> 00:03:40.237
100% recyclable box. It's easy. It's

00:03:40.320 --> 00:03:44.717
fast. It's convenient. I really, really
enjoy it. And best of all, looking at

00:03:44.800 --> 00:03:49.437
the average cost, it works out to be
less than $6 per meal. Skip the lines

00:03:49.520 --> 00:03:52.957
for your Thanksgiving turkey. This
holiday, ButcherBox is proud to give new

00:03:53.040 --> 00:03:54.917
members a free turkey. Just go to

00:03:55.000 --> 00:03:57.077
butcherbox.com/tim

00:03:57.160 --> 00:04:01.117
to sign up. That's butcherbox.com/tim

00:04:01.200 --> 00:04:04.557
to receive a free 10 to 14
lb turkey in your first box.

00:04:04.640 --> 00:04:08.117
>> Optimal minimal >> At this
altitude [music] I can run flat

00:04:08.200 --> 00:04:10.917
out for a half mile before
my hands start shaking.

00:04:11.000 --> 00:04:12.917
>> Can I ask you a personal question?

00:04:13.000 --> 00:04:15.077
>> No, I just ate my breakfast time.

00:04:15.160 --> 00:04:16.677
>> What if I did the opposite?

00:04:16.760 --> 00:04:22.000
>> I'm a cybernetic organism living this
year over middle [music] impulse power.

00:04:29.280 --> 00:04:30.877
>> Hello boys and girls, ladies and germs.

00:04:30.960 --> 00:04:34.277
This is Tim Ferriss. Welcome to another
episode of The Tim Ferriss Show. My

00:04:34.360 --> 00:04:36.797
guest today is Eric Schmidt. Eric

00:04:36.880 --> 00:04:40.237
Schmidt, that's s c h m i d t on Twitter

00:04:40.320 --> 00:04:44.077
@ericschmidt, is a technologist,
entrepreneur, and philanthropist. He

00:04:44.160 --> 00:04:47.237
joined Google in 2001 helping
the company grow from a Silicon Valley

00:04:47.320 --> 00:04:49.997
startup to a global technology leader.

00:04:50.080 --> 00:04:54.277
He served as chief executive officer
and chairman from 2001 to 2011 and as

00:04:54.360 --> 00:04:58.357
executive chairman and technical advisor
thereafter. Under his leadership Google

00:04:58.440 --> 00:05:01.997
dramatically scaled its infrastructure
and diversified its product offerings

00:05:02.080 --> 00:05:05.557
while maintaining a culture of innovation.
In 2017 he co-founded

00:05:05.640 --> 00:05:09.117
Schmidt Futures, a philanthropic
initiative that bets early on

00:05:09.200 --> 00:05:12.917
exceptional people making the world
better. He serves as chair of the Broad

00:05:13.000 --> 00:05:16.477
Institute and formerly served as chair
of the National Security Commission on

00:05:16.560 --> 00:05:20.037
Artificial Intelligence. He's the host
of Reimagine with Eric Schmidt, a

00:05:20.120 --> 00:05:23.837
podcast exploring how society can build
a brighter future after the COVID-19

00:05:23.920 --> 00:05:26.197
pandemic. With co-authors Henry A.

00:05:26.280 --> 00:05:28.917
Kissinger and Daniel Huttenlocher, Eric

00:05:29.000 --> 00:05:33.757
has a new book out titled The Age of AI
and Our Human Future. You can find him

00:05:33.840 --> 00:05:37.557
again on Twitter @ericschmidt
and at ericschmidt.com.

00:05:37.640 --> 00:05:39.637
Eric, welcome back to the show.

00:05:39.720 --> 00:05:43.437
>> Thank you. I really look
forward to this conversation.

00:05:43.520 --> 00:05:45.397
>> I have been looking forward to this and

00:05:45.480 --> 00:05:47.397
I want to confess first and foremost

00:05:47.480 --> 00:05:50.597
that I have tremendous insecurity around

00:05:50.680 --> 00:05:54.677
my lack of clarity on AI. So, I am

00:05:54.760 --> 00:05:56.157
really looking forward to digging into

00:05:56.240 --> 00:05:59.237
many facets, but before we get to that,

00:05:59.320 --> 00:06:02.997
I want to pick out Henry Kissinger. How

00:06:03.080 --> 00:06:06.677
did you come to collaborate with Henry

00:06:06.760 --> 00:06:10.157
Kissinger? >> About 12
years ago, I met him at a

00:06:10.240 --> 00:06:12.677
conference called Bilderberg.

00:06:12.760 --> 00:06:17.957
And my father worked for the Nixon
administration when I was very young.

00:06:18.040 --> 00:06:19.477
And my father

00:06:19.560 --> 00:06:22.277
had Henry as a hero. He said he was the

00:06:22.360 --> 00:06:25.477
most brilliant, hardest working, and he

00:06:25.560 --> 00:06:28.477
was enigmatic because he has both a very

00:06:28.560 --> 00:06:32.917
favorable reputation, but also
a very controversial reputation.

00:06:33.000 --> 00:06:35.397
So, we chatted and he said, "The only

00:06:35.480 --> 00:06:40.197
problem I have with Google is I think
that you're going to destroy the world."

00:06:40.280 --> 00:06:43.517
And I thought, "Well, that's
a challenge from Henry Kissinger."

00:06:43.600 --> 00:06:47.717
So, we invited him to Google,
where he gave a speech.

00:06:47.800 --> 00:06:52.237
And he confronted the employees directly
on the manipulation that Google was

00:06:52.320 --> 00:06:56.877
doing in his view of the public discourse.

00:06:56.960 --> 00:07:00.797
And his criticisms were so apt, and the
people enjoyed it so much that we struck

00:07:00.880 --> 00:07:03.877
up a friendship.
And although we're politically very

00:07:03.960 --> 00:07:06.597
different with very different backgrounds,

00:07:06.680 --> 00:07:11.957
I've come to learn that working with
someone this brilliant at any age is

00:07:12.040 --> 00:07:15.917
phenomenal. But when
your co-author is 98, it's a

00:07:16.000 --> 00:07:19.477
special treat. >> I looked
up the age. I'm glad you

00:07:19.560 --> 00:07:21.797
mentioned it. 98.

00:07:21.880 --> 00:07:24.437
To what do you attribute him remaining

00:07:24.520 --> 00:07:28.237
cognitively sharp into his late 90s? Is

00:07:28.320 --> 00:07:30.477
that just good hardware out of the box?

00:07:30.560 --> 00:07:32.157
Is there more to it?

00:07:32.240 --> 00:07:36.637
>> He works harder than a 40-year-old.

00:07:36.720 --> 00:07:41.197
I can tell you that he gets up in the
morning and he works all day. He has

00:07:41.280 --> 00:07:43.637
dinner with his wife and his family, and

00:07:43.720 --> 00:07:45.757
then he works at night. And he keeps up

00:07:45.840 --> 00:07:49.917
that pace 7 days a week at 98.

00:07:50.000 --> 00:07:54.797
I am convinced that the secret
to longevity is being a workaholic.

00:07:54.880 --> 00:07:59.197
>> [laughter] >> And and
the reason I say that is that

00:07:59.280 --> 00:08:03.877
Henry Kissinger at the age
of 90 knew nothing about the

00:08:03.960 --> 00:08:07.477
digital world, although he
had a lot of opinions about

00:08:07.560 --> 00:08:11.557
it. But he has mastered the digital world

00:08:11.640 --> 00:08:13.477
and artificial intelligence

00:08:13.560 --> 00:08:15.637
with the alacrity and the speed of

00:08:15.720 --> 00:08:19.237
people who are just getting into it now.

00:08:19.320 --> 00:08:22.077
That's unique to him. That's a gift. And

00:08:22.160 --> 00:08:26.597
that's why his analysis
of our world is so incredibly

00:08:26.680 --> 00:08:29.117
interesting to me. >> All right,
we're going to spend a little

00:08:29.200 --> 00:08:32.157
more time on on Henry because I must

00:08:32.240 --> 00:08:34.557
scratch this itch. So you are very good

00:08:34.640 --> 00:08:37.877
at systematizing thinking from first

00:08:37.960 --> 00:08:41.077
principles, whether it's systematizing

00:08:41.160 --> 00:08:43.557
innovation or hiring, thinking of things

00:08:43.640 --> 00:08:46.717
at scale. There's a method to not the

00:08:46.800 --> 00:08:49.757
madness, but the outcomes.

00:08:49.840 --> 00:08:53.877
Was Henry's ability to learn so quickly

00:08:53.960 --> 00:08:58.517
based on some approach that he has to
first principles or a framework [snorts]

00:08:58.600 --> 00:09:01.757
of any type that you've
seen exhibited in him

00:09:01.840 --> 00:09:04.797
that he applies to new domains?

00:09:04.880 --> 00:09:09.557
>> There've been some studies about the
age at which you are your most productive

00:09:09.640 --> 00:09:12.997
professionally. And as you
know in math and science,

00:09:13.080 --> 00:09:17.197
brilliance tends to show up young,
in their 20s. They tend to get early

00:09:17.280 --> 00:09:20.957
awards. Historians,
however, seem to get better

00:09:21.040 --> 00:09:25.557
with age. Maybe it's
the accumulation of perspective

00:09:25.640 --> 00:09:27.757
and the accumulation of reasoning and

00:09:27.840 --> 00:09:30.757
the depth of wisdom that is represented

00:09:30.840 --> 00:09:35.037
by increasing age. And so

00:09:35.120 --> 00:09:39.397
Dr. Kissinger has both
the benefit of being a brilliant

00:09:39.480 --> 00:09:41.557
historian and also having changed

00:09:41.640 --> 00:09:44.517
history and lived in the moment.

00:09:44.600 --> 00:09:46.837
And today he spends a great deal of his

00:09:46.920 --> 00:09:48.717
time with people talking to him about

00:09:48.800 --> 00:09:51.757
current affairs and judging them with

00:09:51.840 --> 00:09:54.917
his historical principles in mind.

00:09:55.000 --> 00:09:57.997
It is from that basis of insight

00:09:58.080 --> 00:10:01.397
that when he looked at AI, he said,

00:10:01.480 --> 00:10:04.637
"This is a very much bigger thing

00:10:04.720 --> 00:10:07.037
than people think it is." And I said to

00:10:07.120 --> 00:10:10.517
him, "Why?" I honestly didn't know.

00:10:10.600 --> 00:10:15.557
And he said, "Because we're discussing
artificial intelligence as though it's a

00:10:15.640 --> 00:10:20.917
technology. This is like
the beginning of the Renaissance."

00:10:21.000 --> 00:10:23.717
And I said, "Tell me about
the Renaissance. What else do you say to a

00:10:23.800 --> 00:10:27.437
historian who's famous?" And we started
to talk about the Renaissance, and he

00:10:27.520 --> 00:10:29.757
said that the Renaissance is really

00:10:29.840 --> 00:10:32.637
about the age of reason. It's about

00:10:32.720 --> 00:10:34.797
individuals being able to think through

00:10:34.880 --> 00:10:37.917
their systems. It's about society

00:10:38.000 --> 00:10:40.677
allowing experts to criticize other

00:10:40.760 --> 00:10:45.397
people. Before the Renaissance,
decision-making was essentially

00:10:45.480 --> 00:10:47.317
hierarchical and from a king or a

00:10:47.400 --> 00:10:49.077
religious leader.

00:10:49.160 --> 00:10:52.437
That change allowed us to develop

00:10:52.520 --> 00:10:57.357
intellectual thought.
He is arguing that we're entering a new

00:10:57.440 --> 00:11:00.717
epoch similar to the
Renaissance, this age of

00:11:00.800 --> 00:11:02.917
artificial intelligence, because

00:11:03.000 --> 00:11:08.877
humanity has never had a competitive
intelligence similar to itself but not

00:11:08.960 --> 00:11:11.437
human. >> I'd love to hear
you elaborate just a a

00:11:11.520 --> 00:11:14.237
bit, and then we're going to dig into a

00:11:14.320 --> 00:11:15.957
whole slew of questions that I have in

00:11:16.040 --> 00:11:20.077
front of me. How you thought about the

00:11:20.160 --> 00:11:23.557
composition of the co-authors on this

00:11:23.640 --> 00:11:27.957
book, and certainly would love to learn
more about Daniel. I know a little bit

00:11:28.040 --> 00:11:32.397
about Daniel Huttenlocher, but I'd love
to hear more. And we could start with

00:11:32.480 --> 00:11:36.357
Dr. Kissinger. I realize
calling him Henry's probably

00:11:36.440 --> 00:11:38.157
going to give me bad karma, so I'll

00:11:38.240 --> 00:11:40.397
start with Dr. Kissinger.

00:11:40.480 --> 00:11:43.717
Is it his broad familiarity with history

00:11:43.800 --> 00:11:46.197
as well as his knowledge of geopolitics

00:11:46.280 --> 00:11:50.037
and statecraft that you were hoping to

00:11:50.120 --> 00:11:52.197
augment everything else in the book?

00:11:52.280 --> 00:11:56.477
Maybe you could just speak to how you
think about what each party brings to

00:11:56.560 --> 00:11:59.117
this project?
>> In our collaboration, the book

00:11:59.200 --> 00:12:00.757
suggestion was actually from Dr.

00:12:00.840 --> 00:12:04.037
Kissinger. And what he basically said

00:12:04.120 --> 00:12:07.397
is that we have an opportunity

00:12:07.480 --> 00:12:10.837
to architect the questions

00:12:10.920 --> 00:12:14.797
that need to get answered
in the next 10, 20, 30 years. I would

00:12:14.880 --> 00:12:17.557
parenthetically offer that we didn't

00:12:17.640 --> 00:12:20.477
understand when we invented social media

00:12:20.560 --> 00:12:22.837
15 years ago or so,

00:12:22.920 --> 00:12:26.117
the extraordinary and compound benefits

00:12:26.200 --> 00:12:29.077
and costs of social media to our society

00:12:29.160 --> 00:12:31.917
and particular political discourse.

00:12:32.000 --> 00:12:35.757
So, armed with that knowledge that the
tech industry had invented a tool that

00:12:35.840 --> 00:12:39.117
outstripped the governance of it,
at least initially and maybe for a long

00:12:39.200 --> 00:12:44.437
time, Dr. Kissinger said, "This
is an opportunity to ask the right

00:12:44.520 --> 00:12:45.877
questions."

00:12:45.960 --> 00:12:48.277
We recruited Dan Huttenlocher, who is

00:12:48.360 --> 00:12:53.477
the Dean of MIT computer science,
partly because he's a good friend and

00:12:53.560 --> 00:12:57.317
partly because he's such a good scientist,

00:12:57.400 --> 00:12:59.437
he will make sure that our claims are

00:12:59.520 --> 00:13:00.757
accurate.

00:13:00.840 --> 00:13:05.557
And Dan worked very hard to get the path

00:13:05.640 --> 00:13:08.757
of AI correct.
Many of the books about artificial

00:13:08.840 --> 00:13:11.117
intelligence are speculative,

00:13:11.200 --> 00:13:13.997
but I'm not aware of any book

00:13:14.080 --> 00:13:16.837
that has both the geopolitical, social,

00:13:16.920 --> 00:13:21.237
and historical context,
but also has the technology correct.

00:13:21.320 --> 00:13:25.157
>> Just to look at the opposite
of speculative from a first-hand

00:13:25.240 --> 00:13:28.197
perspective, could you tell

00:13:28.280 --> 00:13:32.117
a story or two about what you've seen or

00:13:32.200 --> 00:13:36.517
experienced firsthand with AI?

00:13:36.600 --> 00:13:39.717
>> Well, you use AI today whenever you use

00:13:39.800 --> 00:13:45.077
Google Search, Google Ads, spelling
correction, Google Translate. The story

00:13:45.160 --> 00:13:46.517
on AI

00:13:46.600 --> 00:13:48.677
that's not really well understood is

00:13:48.760 --> 00:13:52.397
that in the 1960s and 70s when I started

00:13:52.480 --> 00:13:55.917
AI was going to happen within a decade.

00:13:56.000 --> 00:13:59.477
And my friends who were AI obsessed got

00:13:59.560 --> 00:14:04.357
their PhDs in this area.
And then everything stopped. It stopped

00:14:04.440 --> 00:14:08.037
working. And there was a
period of about 20 years

00:14:08.120 --> 00:14:10.797
which is known as the AI winter

00:14:10.880 --> 00:14:13.357
where the systems didn't work.

00:14:13.440 --> 00:14:18.077
And then a series of mathematicians
in the 80s and 90s invented what is today

00:14:18.160 --> 00:14:22.197
known as deep learning.
I'll spare you the technical details.

00:14:22.280 --> 00:14:25.797
But the important thing about this deep
learning is it allows the manipulation

00:14:25.880 --> 00:14:28.757
of patterns at scale that allowed these

00:14:28.840 --> 00:14:30.797
algorithms to work.

00:14:30.880 --> 00:14:33.877
And the big breakthrough was in 2011

00:14:33.960 --> 00:14:37.117
with a process called ImageNet

00:14:37.200 --> 00:14:38.797
where there was a contest to see if

00:14:38.880 --> 00:14:42.517
computers could see better than humans.

00:14:42.600 --> 00:14:46.597
And today computers can see better
than humans. Their vision is literally

00:14:46.680 --> 00:14:47.917
better.

00:14:48.000 --> 00:14:50.597
And I didn't realize at the time how

00:14:50.680 --> 00:14:54.397
important sight was for everything.

00:14:54.480 --> 00:14:57.237
A car should be driven by a computer.

00:14:57.320 --> 00:15:00.477
The doctor should use an AI system to

00:15:00.560 --> 00:15:05.637
examine you and then give him
or her recommendations on your care.

00:15:05.720 --> 00:15:07.317
I'd much rather have the computer look

00:15:07.400 --> 00:15:11.157
at my skin rash or my retina in my eye

00:15:11.240 --> 00:15:14.757
because we now know from many many tests

00:15:14.840 --> 00:15:18.557
that humans make observational mistakes,
even the best, but computers when

00:15:18.640 --> 00:15:22.637
properly trained don't.
So it was from this insight

00:15:22.720 --> 00:15:26.917
that you could do vision at scale
that you began to be able to do

00:15:27.000 --> 00:15:28.917
prediction at scale.

00:15:29.000 --> 00:15:34.037
And all of a sudden now we're beginning
to see systems that can predict the next

00:15:34.120 --> 00:15:38.557
thing. And computers have gotten very good
at predicting what will happen next. The

00:15:38.640 --> 00:15:40.677
most recent well there sort of three

00:15:40.760 --> 00:15:43.597
events in the last three years that

00:15:43.680 --> 00:15:48.797
really are the index points. The first
is that and Dr. Kissinger wrote an

00:15:48.880 --> 00:15:54.397
article >> after AlphaGo >>
called the end of history basically

00:15:54.480 --> 00:15:58.317
and inspired by the fact that Go it was a

00:15:58.400 --> 00:16:02.357
game that humans had played
for 2,500 years. It was thought to be

00:16:02.440 --> 00:16:06.917
incomputable. Not only did
a computer solve the game

00:16:07.000 --> 00:16:11.237
but it beat the top humans both in Korea
and China. I know cuz I was physically

00:16:11.320 --> 00:16:14.277
there. Computer against
human and it was great

00:16:14.360 --> 00:16:18.357
fun. But in that process

00:16:18.440 --> 00:16:22.557
the computer invented some new moves
and strategies that had not been known to

00:16:22.640 --> 00:16:27.277
humans for 2,500 years.
Now that's a big deal.

00:16:27.360 --> 00:16:31.397
The next thing that
happened was that at MIT

00:16:31.480 --> 00:16:35.077
a set of synthetic biologists
and computer scientists did a very

00:16:35.160 --> 00:16:39.317
complicated trick involving
going through a hundred

00:16:39.400 --> 00:16:42.037
million different compounds

00:16:42.120 --> 00:16:44.037
and figuring out which compounds would

00:16:44.120 --> 00:16:48.277
create a reaction for antibiotic use.

00:16:48.360 --> 00:16:52.797
And they came up with using this
technique a new drug that could not be

00:16:52.880 --> 00:16:57.277
foreseen. It's called Halicin and it
appears to be the next broad scale

00:16:57.360 --> 00:17:01.197
antibiotic. We haven't had one
in roughly 40 years. The third thing that

00:17:01.280 --> 00:17:05.757
happened was that a group
called OpenAI built

00:17:05.840 --> 00:17:07.637
what is are now called

00:17:07.720 --> 00:17:10.037
universal models where they read

00:17:10.120 --> 00:17:13.197
everything they could find on the web

00:17:13.280 --> 00:17:16.837
into something called GPT-3.

00:17:16.920 --> 00:17:21.517
This is called a basically a transformer
and these generative models can generate

00:17:21.600 --> 00:17:23.317
things.

00:17:23.400 --> 00:17:26.557
So, all of a sudden, we have a computer

00:17:26.640 --> 00:17:30.317
that can speak what it knows.

00:17:30.400 --> 00:17:34.517
And these models are interesting because
you train them and you don't know what

00:17:34.600 --> 00:17:37.477
they know. And furthermore,
they can't tell you.

00:17:37.560 --> 00:17:41.717
You have to ask them.
So, it's like a teenager.

00:17:41.800 --> 00:17:46.797
And many people think that these
universal models are going to profoundly

00:17:46.880 --> 00:17:50.197
change our understanding
of language and thought

00:17:50.280 --> 00:17:52.917
because they only get better with scale.

00:17:53.000 --> 00:17:56.197
So, we've now got four or five
companies, couple big ones, couple

00:17:56.280 --> 00:18:00.077
startups that are building what are
called trillion parameter models.

00:18:00.160 --> 00:18:02.197
These trillion parameter models cost a

00:18:02.280 --> 00:18:07.440
hundred million dollars or so to make.

00:18:07.560 --> 00:18:09.637
That's how exciting this new area is.

00:18:09.720 --> 00:18:14.477
So, you've got strategy in the form
of Go, you've got medicine and drug

00:18:14.560 --> 00:18:18.357
discovery, really scientific discovery
in the form of AlphaFold, and now you've

00:18:18.440 --> 00:18:22.717
got language models and learning models
in We believe collectively that in the

00:18:22.800 --> 00:18:28.520
next 10 years, this is going to come
together and transform everything.

00:18:29.720 --> 00:18:33.517
>> I'm just taking a breath
to let that settle in.

00:18:33.600 --> 00:18:38.237
Uh GPT-3 really got
my attention in part and I

00:18:38.320 --> 00:18:39.797
want you to disabuse me of any

00:18:39.880 --> 00:18:43.157
misunderstandings, but I saw

00:18:43.240 --> 00:18:46.237
interviews generated between people who

00:18:46.320 --> 00:18:50.677
are no longer living. So, I saw Marcus
Aurelius interviewing >> [laughter]

00:18:50.760 --> 00:18:53.037
>> Let's just call Mark Twain. Mark Twain

00:18:53.120 --> 00:18:55.397
interviewing Abraham Lincoln and I

00:18:55.480 --> 00:18:57.997
thought to myself, once we have enough

00:18:58.080 --> 00:19:00.717
audio and video online, this will really

00:19:00.800 --> 00:19:05.037
simplify my guest recruitment for the
podcast because I won't actually need to

00:19:05.120 --> 00:19:08.597
reach out to the living human. It'll be
great. I won't have to keep banging on

00:19:08.680 --> 00:19:11.757
the door of Oprah or anyone else. I'll
just be able to generate an interview.

00:19:11.840 --> 00:19:14.037
It'll save me recording time, too.

00:19:14.120 --> 00:19:17.077
>> You say that in jest,
but we're busy building that.

00:19:17.160 --> 00:19:19.917
>> Right. >> And so,
let me just be clear how the Tim

00:19:20.000 --> 00:19:22.397
Ferriss Show would work. So, you have an

00:19:22.480 --> 00:19:26.917
enormous amount of video and audio data.

00:19:27.000 --> 00:19:28.877
So, there's enough to learn to build a

00:19:28.960 --> 00:19:31.957
Tim Ferriss question and answering thing

00:19:32.040 --> 00:19:35.117
where you would have not only your tone

00:19:35.200 --> 00:19:38.197
and your the fun that you represent,

00:19:38.280 --> 00:19:42.957
but it would also represent the sum
of your insight. So, it would say, "I was

00:19:43.040 --> 00:19:47.477
speaking with so-and-so,
and he or she said this, what do you

00:19:47.560 --> 00:19:51.037
think?" Now, in this particular
scenario, this is long after your own

00:19:51.120 --> 00:19:54.637
passing and all your guests passing, too.

00:19:54.720 --> 00:19:59.797
There's every reason to think
that we'll be able to mimic

00:19:59.880 --> 00:20:04.117
your intelligence and charm and wit

00:20:04.200 --> 00:20:06.677
and that of your family, your parents,

00:20:06.760 --> 00:20:10.317
your grandparents, historical figures.

00:20:10.400 --> 00:20:14.117
That will be fun.
The question is, will it really change

00:20:14.200 --> 00:20:18.877
society? >> Or how will
it change society? I recall

00:20:18.960 --> 00:20:22.397
being in a small meeting,
private meeting, but the content I'm

00:20:22.480 --> 00:20:24.317
going to mention is public,

00:20:24.400 --> 00:20:27.477
and the discussion centered on the

00:20:27.560 --> 00:20:32.117
legal cases that will be forthcoming
related to a lot of this technology, and

00:20:32.200 --> 00:20:36.557
there was at the time something related
to Now, I should be clear. I'm I want to

00:20:36.640 --> 00:20:39.957
take a backstep in a minute and define
artificial intelligence and what that is

00:20:40.040 --> 00:20:42.317
so that it doesn't get
conflated with other things.

00:20:42.400 --> 00:20:44.557
But in terms of developing technology,

00:20:44.640 --> 00:20:49.757
deep fakes, copyright, defamation,
there's been a lot in the

00:20:49.840 --> 00:20:53.437
news, at least in the last few years,
related to deep fakes of Taylor Swift

00:20:53.520 --> 00:20:57.677
and lawsuits, and we are
all going to have to contend

00:20:57.760 --> 00:21:00.837
with entirely new breeds of legal cases, I

00:21:00.920 --> 00:21:03.077
would imagine. So, that's another

00:21:03.160 --> 00:21:05.397
uh facet of this that is going to be

00:21:05.480 --> 00:21:10.117
incredibly fascinating,
terrifying, and complex, possibly.

00:21:10.200 --> 00:21:13.117
I would love for you,
if you if you don't mind, to

00:21:13.200 --> 00:21:16.757
just define for me, quite frankly,
artificial intelligence, because as I'm

00:21:16.840 --> 00:21:20.277
sure you have seen and can imagine,
there are a lot of terms that kind of

00:21:20.360 --> 00:21:23.837
get copied and pasted into startup decks
whenever they're hot. That [laughter]

00:21:23.920 --> 00:21:29.317
would be deep learning, machine learning,
AI. What is AI and what is it

00:21:29.400 --> 00:21:34.077
not? >> The simplest explanation for AI
is it's a system that gets better through

00:21:34.160 --> 00:21:37.797
learning. It is busy learning something.

00:21:37.880 --> 00:21:42.317
And that's the probably the easiest
and current definition of it.

00:21:42.400 --> 00:21:47.117
When I say AI to my non-technical friends,
they typically think of a movie

00:21:47.200 --> 00:21:51.037
that they saw.
And the movie always has a robot, and

00:21:51.120 --> 00:21:53.677
the robot goes awry,

00:21:53.760 --> 00:21:57.197
and that robot is slain by a female

00:21:57.280 --> 00:21:58.837
scientist

00:21:58.920 --> 00:22:01.517
who triumphs.

00:22:01.600 --> 00:22:04.397
And I propose a variant of that movie

00:22:04.480 --> 00:22:07.637
where the computers all conspire against

00:22:07.720 --> 00:22:10.077
humans and start killing humans.

00:22:10.160 --> 00:22:14.317
The humans notice this,
the computers run away from the humans,

00:22:14.400 --> 00:22:19.557
and a female scientist figures out how
to unplug the computers one by one by

00:22:19.640 --> 00:22:24.197
one. That's not what AI is. That's a movie

00:22:24.280 --> 00:22:27.517
fiction. And it'll be
a very long time, hopefully

00:22:27.600 --> 00:22:30.117
never, that we'll have to deal with that.

00:22:30.200 --> 00:22:32.997
What AI really is is a system of

00:22:33.080 --> 00:22:35.757
knowledge that is implemented inside the

00:22:35.840 --> 00:22:38.237
cloud, so it's around you all the time,

00:22:38.320 --> 00:22:40.997
and it's very good at looking at large

00:22:41.080 --> 00:22:43.197
data sets and predicting things. It's

00:22:43.280 --> 00:22:48.157
very good at looking and finding
patterns that humans can't see.

00:22:48.240 --> 00:22:52.077
I mentioned earlier that the
computers were very good at vision.

00:22:52.160 --> 00:22:54.037
That was the first breakthrough.

00:22:54.120 --> 00:22:55.877
But what they're now doing is looking

00:22:55.960 --> 00:22:59.037
for patterns of correlation that humans

00:22:59.120 --> 00:23:03.637
can't see. So, let's imagine that on
Tuesday you do something and on Friday

00:23:03.720 --> 00:23:07.237
someone else does something and it turns
out that you're magically linked in the

00:23:07.320 --> 00:23:11.077
universe such that Tuesday causes
Friday, but it's not a pattern that's

00:23:11.160 --> 00:23:14.717
apparent. Computers can
discover that pattern,

00:23:14.800 --> 00:23:19.197
that needle in a haystack particularly
well. When you operate this stuff at

00:23:19.280 --> 00:23:23.357
scale, you end up with systems that look

00:23:23.440 --> 00:23:26.717
human-like because they
can aggregate data and they

00:23:26.800 --> 00:23:29.837
can think what you think.
They can generate

00:23:29.920 --> 00:23:34.037
a solution. There's a technology
called GANs, uh generative adversarial

00:23:34.120 --> 00:23:38.677
networks, where the computer can
generate candidates and another network

00:23:38.760 --> 00:23:43.277
says not good enough, not good enough,
not good enough until they find it. GANs

00:23:43.360 --> 00:23:45.277
have been used for example to build

00:23:45.360 --> 00:23:47.597
pictures of actresses and actors that

00:23:47.680 --> 00:23:52.637
are fake, but look so real
you're sure you know them.

00:23:52.720 --> 00:23:57.117
So one of the concerns, the one
of the real concerns about AI,

00:23:57.200 --> 00:24:00.197
is that it's going to be very
difficult to tell the difference between

00:24:00.280 --> 00:24:02.637
information and misinformation.

00:24:02.720 --> 00:24:04.277
And I'll give you an example. Let's

00:24:04.360 --> 00:24:06.957
assume that all of your readers and

00:24:07.040 --> 00:24:11.917
listeners are obviously human and that
the standard rules of humans apply to

00:24:12.000 --> 00:24:15.197
all of us. There's something called
anchoring bias, there's something called

00:24:15.280 --> 00:24:19.637
recency bias. You know, people aren't
completely rational computers nor should

00:24:19.720 --> 00:24:22.397
they. We don't want them to be.
We love humans.

00:24:22.480 --> 00:24:26.757
But let's imagine the computer gets
enough of such people and it figures out

00:24:26.840 --> 00:24:28.717
that if I say the following outrageous

00:24:28.800 --> 00:24:32.357
thing first, you'll always believe me.

00:24:32.440 --> 00:24:34.277
And the computer discovers this and so

00:24:34.360 --> 00:24:36.957
all of a sudden everything it does boom,

00:24:37.040 --> 00:24:39.437
boom, boom. I thought a lot about the

00:24:39.520 --> 00:24:44.477
way politicians speak. If you watch
carefully, they take a set of phrases

00:24:44.560 --> 00:24:48.357
and they repeat them over and over again.
They're simple phrases. That's

00:24:48.440 --> 00:24:52.357
anchoring. They're trying to get
the audience, their voters, to start with

00:24:52.440 --> 00:24:54.597
this fact and then judge past it. Well,

00:24:54.680 --> 00:24:57.597
computers will be incredibly good

00:24:57.680 --> 00:25:02.197
at exploiting that.
That means that the world, our social

00:25:02.280 --> 00:25:04.597
world around us within the next decade,

00:25:04.680 --> 00:25:09.237
will become impossibly confusing
because there's so many actors that will

00:25:09.320 --> 00:25:11.477
want to misinform us. Businesses,

00:25:11.560 --> 00:25:14.797
politicians, our opponents, for fun, God

00:25:14.880 --> 00:25:17.637
knows. We don't know how to manage that.

00:25:17.720 --> 00:25:21.677
>> When you mentioned things coming
together in some fashion in the next 10

00:25:21.760 --> 00:25:25.877
years, could you speak to

00:25:25.960 --> 00:25:28.597
if and how generalized or general

00:25:28.680 --> 00:25:31.637
artificial intelligence fits into that?

00:25:31.720 --> 00:25:34.077
Because I know that's a big question
for a lot of people, including those who

00:25:34.160 --> 00:25:36.157
don't really understand the technology,

00:25:36.240 --> 00:25:41.237
when generalized artificial
intelligence will arrive, so to speak.

00:25:41.320 --> 00:25:44.517
>> I think it's important
to establish some of these terms.

00:25:44.600 --> 00:25:47.557
AGI stands for artificial
general intelligence.

00:25:47.640 --> 00:25:50.317
And AGI refers to computers that are

00:25:50.400 --> 00:25:51.957
human-like

00:25:52.040 --> 00:25:55.797
in their strategic and capability.

00:25:55.880 --> 00:26:00.077
Today's algorithms are extraordinarily
good at generating content,

00:26:00.160 --> 00:26:03.637
misinformation, guidance,
helping you with science, and

00:26:03.720 --> 00:26:07.117
so forth. But they're
not self-determinative.

00:26:07.200 --> 00:26:08.997
They don't actually have the notion of

00:26:09.080 --> 00:26:11.277
who am I and what do I do next? If you

00:26:11.360 --> 00:26:15.477
ask GPT-3 who it is, GPT-3 will say,

00:26:15.560 --> 00:26:20.757
"I'm a computer."
But it won't give you a philosophical

00:26:20.840 --> 00:26:24.477
basis for its existence,
its purpose, and how

00:26:24.560 --> 00:26:27.197
it's determining where it goes.

00:26:27.280 --> 00:26:29.877
That is the distinction between AI and

00:26:29.960 --> 00:26:32.317
AGI, in my view.

00:26:32.400 --> 00:26:35.757
Now, within the community, the AGI

00:26:35.840 --> 00:26:38.637
optimists think that within 10 years we

00:26:38.720 --> 00:26:41.117
will have such computers. The pessimists

00:26:41.200 --> 00:26:42.877
think that it will be far longer if

00:26:42.960 --> 00:26:46.717
ever. So, let's say 15 years from now,

00:26:46.800 --> 00:26:52.277
it may be possible to have computers
that have a sense of self-determination,

00:26:52.360 --> 00:26:56.357
that they know roughly what they're
trying to achieve against a broader

00:26:56.440 --> 00:26:58.117
objective.

00:26:58.200 --> 00:27:00.997
Many people think that this will become

00:27:01.080 --> 00:27:04.357
an enormous competition between humans.

00:27:04.440 --> 00:27:08.957
Because remember that those computers
will be faster learners than humans.

00:27:09.040 --> 00:27:11.877
They'll have access
to more data than humans.

00:27:11.960 --> 00:27:15.157
Other people believe that they'll
be fundamentally flawed.

00:27:15.240 --> 00:27:18.597
Because they won't have the nuance,
they won't have the background, they won't

00:27:18.680 --> 00:27:21.917
have the cultural background of the
countries that we all grew up in and the

00:27:22.000 --> 00:27:24.957
cultures we grow in. We don't really know.

00:27:25.040 --> 00:27:28.597
My own view is that this will occur.

00:27:28.680 --> 00:27:31.077
But that the amount of computation

00:27:31.160 --> 00:27:33.837
required to do this will be so great

00:27:33.920 --> 00:27:36.677
that there'll only be a few of them.

00:27:36.760 --> 00:27:40.197
And furthermore,
they'll become so important

00:27:40.280 --> 00:27:43.037
that they'll be like nuclear weapons.

00:27:43.120 --> 00:27:46.397
And let me tell you why.
Let's imagine you have a truly evil

00:27:46.480 --> 00:27:49.037
person who encounters this AGI. What's

00:27:49.120 --> 00:27:50.837
the first question they'll ask them?

00:27:50.920 --> 00:27:54.197
Tell me how to kill a million people.

00:27:54.280 --> 00:27:57.157
Obviously a terrible
thing to ask and answer.

00:27:57.240 --> 00:28:02.157
And the AGI, because it doesn't
have morals, could actually answer.

00:28:02.240 --> 00:28:05.917
It could actually say, well, do this,
this, this, and this. And furthermore,

00:28:06.000 --> 00:28:09.077
it could articulate something that's
not generally known because it knows

00:28:09.160 --> 00:28:11.277
everything.

00:28:11.360 --> 00:28:14.557
So I think a fair reading of these AGIs

00:28:14.640 --> 00:28:19.197
is that in the most extreme form,
they're going to be so powerful

00:28:19.280 --> 00:28:21.997
that they're going
to have to be protected.

00:28:22.080 --> 00:28:25.757
We don't want them broadly used. They're
going to have to be used in specialized

00:28:25.840 --> 00:28:29.277
scenarios. Now,
such an intelligence will be

00:28:29.360 --> 00:28:33.837
enormously valuable for drug
discovery, material science,

00:28:33.920 --> 00:28:36.197
climate change, making the world more

00:28:36.280 --> 00:28:38.997
efficient, making people more educated.

00:28:39.080 --> 00:28:43.957
And imagine with an AGI, you could say,
"Tell me how to teach a million children

00:28:44.040 --> 00:28:47.157
English better."
And he could figure that out.

00:28:47.240 --> 00:28:51.717
Cuz it could in its own thinking to the
greatest thinking, it could look at all

00:28:51.800 --> 00:28:55.197
the patterns, figure out how to do it,
and you could actually ask it to write

00:28:55.280 --> 00:28:57.397
the program.

00:28:57.480 --> 00:29:00.437
The beginning of this is today there are

00:29:00.520 --> 00:29:05.157
companies that are offering tools
which will help you write code.

00:29:05.240 --> 00:29:08.677
They don't know what code you're trying
to write, but once you start, it can

00:29:08.760 --> 00:29:11.357
fill in a bunch of it for you.

00:29:11.440 --> 00:29:14.557
That's the beginning of this phenomena.

00:29:14.640 --> 00:29:18.677
In the extreme case, there
are scientists and basically

00:29:18.760 --> 00:29:22.877
science fiction around the future
that could use something called

00:29:22.960 --> 00:29:26.797
the singularity, which
is the point where the computer

00:29:26.880 --> 00:29:29.237
evolution is so much faster than humans.

00:29:29.320 --> 00:29:33.757
Now, that's speculation.
When you talk to people in my world,

00:29:33.840 --> 00:29:37.757
most people think the singularity will
occur 20 years from whenever you ask

00:29:37.840 --> 00:29:44.760
them that question.
>> [laughter] >> So, they would say 2041.

00:29:45.160 --> 00:29:50.037
>> Where do you fall, if I may ask your
personal opinion, assuming, and this is

00:29:50.120 --> 00:29:53.637
a big assumption, that there are people
on the techno-optimist and pessimist

00:29:53.720 --> 00:29:54.997
side who are very technically

00:29:55.080 --> 00:29:56.957
sophisticated?

00:29:57.040 --> 00:30:00.757
What is your current best guess or

00:30:00.840 --> 00:30:04.597
intuitive feel, however you want
to answer it, on the arrival of something

00:30:04.680 --> 00:30:08.677
we would consider AGI?
>> The top scientists, the greatest

00:30:08.760 --> 00:30:10.797
inventors in this area,

00:30:10.880 --> 00:30:13.157
collectively believe that we need one or

00:30:13.240 --> 00:30:15.877
two more breakthroughs

00:30:15.960 --> 00:30:18.597
to get to volition,

00:30:18.680 --> 00:30:22.917
to get to consciousness in the way
that humans do it. When you wake up in the

00:30:23.000 --> 00:30:27.157
morning, you have so many choices of how
to spend your time. How do you choose?

00:30:27.240 --> 00:30:29.677
How do you handle unplanned situations

00:30:29.760 --> 00:30:31.917
and all of that? Most people believe

00:30:32.000 --> 00:30:34.477
that the computers that I'm talking

00:30:34.560 --> 00:30:38.077
about will become enormously valuable

00:30:38.160 --> 00:30:42.277
partners. My physicist friend, I said
to him, "What would you like the computer

00:30:42.360 --> 00:30:45.037
to do?" And he says, "It's really simple.

00:30:45.120 --> 00:30:49.597
There are so much physics being written
about now that I don't have time to read

00:30:49.680 --> 00:30:50.677
it all.

00:30:50.760 --> 00:30:53.957
And I want the physics assistant

00:30:54.040 --> 00:30:56.757
to go and read everything, figure out

00:30:56.840 --> 00:31:01.277
what disagrees with each other,
suggest things for me to think about

00:31:01.360 --> 00:31:03.117
while I'm sleeping."

00:31:03.200 --> 00:31:06.317
The biologists have similar questions.

00:31:06.400 --> 00:31:10.957
People who are philosophers want new
insights. Read everything. Help me think

00:31:11.040 --> 00:31:14.317
through this. I believe
that in the next decade that

00:31:14.400 --> 00:31:17.837
will be the primary achievement in AI.

00:31:17.920 --> 00:31:19.957
And that is extraordinary.

00:31:20.040 --> 00:31:22.077
Because it means, for example, that we

00:31:22.160 --> 00:31:26.997
can answer questions in physics, math,
medicine, so forth, that have been

00:31:27.080 --> 00:31:30.157
unanswerable. We've not
been able to understand the

00:31:30.240 --> 00:31:34.757
behavior of subatomic particles.
We can do it now.

00:31:34.840 --> 00:31:39.437
The enormous breakthroughs, but that's
not the same thing as real human

00:31:39.520 --> 00:31:43.197
intelligence. I think it's going
to take another breakthrough or two.

00:31:43.280 --> 00:31:47.077
>> I know this is maybe trying
to look into the crystal ball with too

00:31:47.160 --> 00:31:51.717
much intensity, but could you elaborate
on if those breakthroughs are known,

00:31:51.800 --> 00:31:55.077
what they are, what like what the problems
are that need to be solved? Are

00:31:55.160 --> 00:31:57.957
Are they known problems? >> There
are people who are working really

00:31:58.040 --> 00:32:01.637
hard on the question of goals.

00:32:01.720 --> 00:32:06.477
And the computer systems have

00:32:06.560 --> 00:32:09.437
knowledge that they cannot explain,

00:32:09.520 --> 00:32:11.957
which we can think of as intuition.

00:32:12.040 --> 00:32:14.517
So, the system gets trained and it knows

00:32:14.600 --> 00:32:19.237
things, but if you ask it why it knows it,
it can't tell you. That's intuition.

00:32:19.320 --> 00:32:22.957
So, the question here is how much
more computational power and scenario

00:32:23.040 --> 00:32:26.877
planning do you have to do to look
at all the different choices that you have

00:32:26.960 --> 00:32:31.117
every day? There are
people who believe that it's

00:32:31.200 --> 00:32:34.317
perhaps a hundred times
more computing power.

00:32:34.400 --> 00:32:37.277
And computing power goes up, it doubles

00:32:37.360 --> 00:32:41.477
every 2 to 3 years right now.
And so, that gives you a sense that

00:32:41.560 --> 00:32:46.197
within 15 years it's possible
to imagine that amount of

00:32:46.280 --> 00:32:49.797
computing power required to do this.

00:32:49.880 --> 00:32:53.357
And the reason we use computing power
is one, we don't have a better metric, but

00:32:53.440 --> 00:32:56.117
also because the computer doesn't think

00:32:56.200 --> 00:33:00.397
the way humans do. The computer looks
at scenarios and eliminates. So, it says,

00:33:00.480 --> 00:33:02.157
"Well, what if this? And what if this?

00:33:02.240 --> 00:33:06.357
And what if this?"
And the computer itself is eliminating

00:33:06.440 --> 00:33:09.997
this choice, this choice, this choice,
this choice, and finally settles on a

00:33:10.080 --> 00:33:13.157
good choice. So,
many people believe that the

00:33:13.240 --> 00:33:17.917
compounding of knowledge will
take what we say a couple of orders

00:33:18.000 --> 00:33:21.557
of more magnitude.
Which is why these computers are likely

00:33:21.640 --> 00:33:24.597
at least initially to be few, extremely

00:33:24.680 --> 00:33:28.117
expensive, and very, very large.

00:33:28.200 --> 00:33:30.357
>> Does the 15-year

00:33:30.440 --> 00:33:33.077
hypothetical time horizon

00:33:33.160 --> 00:33:35.997
factor in, and And this is me getting

00:33:36.080 --> 00:33:40.717
into very slippery territory since I'm
non-technical, but does that factor in

00:33:40.800 --> 00:33:43.637
the potential applications
of quantum computing?

00:33:43.720 --> 00:33:47.037
>> For at least 20 years, people have been
focusing on this notion of a different

00:33:47.120 --> 00:33:49.957
kind of computer called
a quantum computer.

00:33:50.040 --> 00:33:54.357
And quantum computers are hard to explain,
but basically think of them as

00:33:54.440 --> 00:33:57.237
they do all the computations
at the same time.

00:33:57.320 --> 00:34:00.517
So, instead of going ka-chunk, ka-chunk,
ka-chunk, all the ka-chunks go at the

00:34:00.600 --> 00:34:02.077
same time.

00:34:02.160 --> 00:34:04.917
And the term quantum supremacy means

00:34:05.000 --> 00:34:10.437
that everything occurs at once,
rather than taking the computer time to

00:34:10.520 --> 00:34:13.317
go through each of the
mathematical calculations.

00:34:13.400 --> 00:34:17.437
The reason that this is of such great
interest is that one of the core aspects

00:34:17.520 --> 00:34:21.877
of security in the computer age
is the difference between multiplication

00:34:21.960 --> 00:34:25.357
and division.
And with quantum computers, you could

00:34:25.440 --> 00:34:27.917
break all of the codes and all of the

00:34:28.000 --> 00:34:30.117
secrets and all of the keys and all of

00:34:30.200 --> 00:34:33.117
that that we all use every day.

00:34:33.200 --> 00:34:37.437
So, a lot of people think that national
security groups are busy working hard to

00:34:37.520 --> 00:34:42.237
do that. The consensus in the industry
is that we are 8 to 10 years away from

00:34:42.320 --> 00:34:45.597
that. And the way they
get there is that the

00:34:45.680 --> 00:34:49.157
current quantum computers
have an error rate

00:34:49.240 --> 00:34:51.317
because it's not perfect when they

00:34:51.400 --> 00:34:56.837
operate. They're essentially
mimicking a natural process digitally.

00:34:56.920 --> 00:35:00.557
And Google showed last
year that they could

00:35:00.640 --> 00:35:04.557
do this on a computer that would have
taken a million years that took roughly

00:35:04.640 --> 00:35:09.037
10 seconds because of it.
So, we we know it's possible.

00:35:09.120 --> 00:35:12.877
The question now is how do you
actually build a real one that's useful?

00:35:12.960 --> 00:35:15.317
If you had a real one, you could instead

00:35:15.400 --> 00:35:18.117
of simulating a physical process, the

00:35:18.200 --> 00:35:20.517
typical one is called annealing where

00:35:20.600 --> 00:35:23.277
two metals basically merge, and this is

00:35:23.360 --> 00:35:28.117
crucial for high-strength steel and
titanium and things like this. And is of

00:35:28.200 --> 00:35:31.837
great business interest.
But the same thing would apply to AI if

00:35:31.920 --> 00:35:35.757
it worked because the algorithm,
instead of looking at all these different

00:35:35.840 --> 00:35:38.877
scenarios, would examine them all
at the same time and pick the best one.

00:35:38.960 --> 00:35:41.637
Technically, the one
with the lowest energy state.

00:35:41.720 --> 00:35:46.677
The current quantum learning
is not proceeding very quickly.

00:35:46.760 --> 00:35:50.437
It's a very hard problem.
My assumption is that quantum learning

00:35:50.520 --> 00:35:54.357
will occur in our lifetimes,
but not very soon.

00:35:54.440 --> 00:35:59.237
>> Thank you. And uh for people listening,
you want to do a a real deep dive on

00:35:59.320 --> 00:36:01.117
quantum learning and quantum computing.

00:36:01.200 --> 00:36:04.957
I have a conversation with Steve
Jurvetson that goes very

00:36:05.040 --> 00:36:06.677
deeply into this subject matter, and it

00:36:06.760 --> 00:36:09.477
really does start to be a mind bender

00:36:09.560 --> 00:36:15.640
when you get into the details.
It is quite something.

00:36:16.840 --> 00:36:19.477
Just a quick thanks to one of our
sponsors, and we'll be right back to the

00:36:19.560 --> 00:36:24.117
show. This episode is brought to you
by ShipStation. The holiday season is fast

00:36:24.200 --> 00:36:27.997
approaching, and we We that people will
be buying more stuff online than ever

00:36:28.080 --> 00:36:31.717
before. All of these trends
to e-commerce have been accelerated due to

00:36:31.800 --> 00:36:35.797
COVID and much more. If you're an
e-commerce seller, are you ready to meet

00:36:35.880 --> 00:36:39.557
the demands of a record-breaking
online shopping season? Be ready with

00:36:39.640 --> 00:36:43.997
ShipStation. shipstation.com is the
fastest, easiest, and most affordable

00:36:44.080 --> 00:36:47.957
way to manage and ship your orders.
In just a few clicks, you're managing

00:36:48.040 --> 00:36:51.837
orders, printing out discounted shipping
labels, and getting your products out

00:36:51.920 --> 00:36:54.917
fast. Happier holidays for you and your

00:36:55.000 --> 00:36:59.157
customers. ShipStation takes the hassle
out of holiday shipping. No matter where

00:36:59.240 --> 00:37:03.637
you're selling, on Amazon, Etsy, your
website via Shopify, or other platforms,

00:37:03.720 --> 00:37:07.597
ShipStation brings all of your orders
into one simple interface. And

00:37:07.680 --> 00:37:12.277
ShipStation works with all of the major
carriers, USPS, FedEx, UPS, even

00:37:12.360 --> 00:37:15.837
international. You can compare and choose
the best shipping solution every

00:37:15.920 --> 00:37:19.557
time, and you can access the same
postage discounts that are usually

00:37:19.640 --> 00:37:22.837
reserved for large Fortune 500
companies. It's no wonder that

00:37:22.920 --> 00:37:26.597
ShipStation is the number one choice
of online sellers. And right now, my

00:37:26.680 --> 00:37:30.677
listeners, that's you guys, can try
ShipStation free for 60 days when you

00:37:30.760 --> 00:37:35.757
use offer code TIM. Just go to the
homepage, shipstation.com, click on the

00:37:35.840 --> 00:37:37.597
microphone at the top of the homepage,

00:37:37.680 --> 00:37:40.717
and type in TIM, T I M. That's it. Go to

00:37:40.800 --> 00:37:43.117
shipstation.com, then enter offer code

00:37:43.200 --> 00:37:48.960
TIM. shipstation.com. Make ship happen.

00:37:49.880 --> 00:37:54.157
I want to ask you, you mentioned
what you philosophers might ask for in

00:37:54.240 --> 00:37:58.157
terms of augmentation or help. You
mentioned the biologist, the physicist.

00:37:58.240 --> 00:38:01.157
I am going to ask you,
I'm just planting a seed,

00:38:01.240 --> 00:38:05.037
what Eric Schmidt might use AI

00:38:05.120 --> 00:38:07.997
technology for in the future. But before

00:38:08.080 --> 00:38:11.117
I get there, I want to just look at a

00:38:11.200 --> 00:38:15.077
snapshot of current day. What are

00:38:15.160 --> 00:38:19.757
some of the coolest or most impressive
things that you've seen AI figure out on

00:38:19.840 --> 00:38:21.117
its own?

00:38:21.200 --> 00:38:24.517
>> I mentioned this new drug and

00:38:24.600 --> 00:38:27.837
the new drug and drug discovery will

00:38:27.920 --> 00:38:32.597
accelerate combination
of the mRNA achievements

00:38:32.680 --> 00:38:37.997
plus the ability now to essentially
replace the way the drug lab works.

00:38:38.080 --> 00:38:41.197
My sort of stereotype is the chemist
wakes up in the morning and says, "Let's

00:38:41.280 --> 00:38:43.557
try the following seven compounds." They

00:38:43.640 --> 00:38:45.837
try the seven compounds, none of them

00:38:45.920 --> 00:38:51.037
work, and at 5:00 they go home to have
dinner and think, watch television, and

00:38:51.120 --> 00:38:53.437
the next morning they
think of another seven.

00:38:53.520 --> 00:38:57.037
Well, the computer can do,
as I mentioned, 100 million

00:38:57.120 --> 00:39:00.637
in a day. That's a huge
accelerant in what they're

00:39:00.720 --> 00:39:04.877
doing. I'm very interested
in the development

00:39:04.960 --> 00:39:07.077
of humans

00:39:07.160 --> 00:39:12.037
together with AI systems.
And the example I would offer is you

00:39:12.120 --> 00:39:16.197
have a 2-year-old and the
2-year-old gets a plush toy.

00:39:16.280 --> 00:39:18.837
It happens to have AI inside of it.

00:39:18.920 --> 00:39:21.437
And by agreement as this child ages

00:39:21.520 --> 00:39:23.677
every year, they get a better toy.

00:39:23.760 --> 00:39:27.277
And of course, the toy gets smarter.

00:39:27.360 --> 00:39:29.917
We don't know at all

00:39:30.000 --> 00:39:32.837
what happens when a child's best friend

00:39:32.920 --> 00:39:37.037
is not a human or a dog.

00:39:37.120 --> 00:39:41.597
We don't know what that does
to the child's bonding to

00:39:41.680 --> 00:39:44.917
other children, to their parents.

00:39:45.000 --> 00:39:48.317
I mean, you know frustrated parents
and the kid is busy and they give them a

00:39:48.400 --> 00:39:51.237
computer and do whatever you want.

00:39:51.320 --> 00:39:53.357
But imagine if that computer is

00:39:53.440 --> 00:39:56.317
learning, talking, thinking,

00:39:56.400 --> 00:40:01.357
educating at the same time.
It's a godsend, right?

00:40:01.440 --> 00:40:03.597
But what is it teaching? >> Yeah.

00:40:03.680 --> 00:40:05.517
>> What is it teaching?
What are its norms?

00:40:05.600 --> 00:40:10.157
What are its values? Will such a child
end up being very stilted with real

00:40:10.240 --> 00:40:13.437
humans and really
comfortable with digital?

00:40:13.520 --> 00:40:18.037
We honestly just don't know.
We know that people get attached to

00:40:18.120 --> 00:40:23.157
inanimate objects. There are many
religions where inanimate objects have

00:40:23.240 --> 00:40:26.437
what we would think of as a bit of a soul.

00:40:26.520 --> 00:40:28.517
We mentioned this in the book.

00:40:28.600 --> 00:40:32.237
But we don't know. I'll give you
another example with elderly. Uh a lot of

00:40:32.320 --> 00:40:35.077
studies indicate that the
elderly are very lonely.

00:40:35.160 --> 00:40:38.117
It's sort of sad. And

00:40:38.200 --> 00:40:42.637
imagine if their best
friend is a digital friend.

00:40:42.720 --> 00:40:47.717
What does that do? Does that extend
their lives? Does it make them crankier?

00:40:47.800 --> 00:40:51.877
Does it make their loneliness more
perverse? I don't think we know yet.

00:40:51.960 --> 00:40:56.597
>> Yeah, or is that digital
friend an emulation of a relative?

00:40:56.680 --> 00:40:58.837
>> That's right. So now we get our

00:40:58.920 --> 00:41:00.757
grandmother as an example and we

00:41:00.840 --> 00:41:03.917
recreate her husband. And what does it

00:41:04.000 --> 00:41:08.397
do when she can chat with her
husband who's now passed away?

00:41:08.480 --> 00:41:12.397
We honestly don't know.
When I go through and I look at these

00:41:12.480 --> 00:41:14.477
technologies, there really four

00:41:14.560 --> 00:41:17.477
qualities in AI that are different.

00:41:17.560 --> 00:41:22.277
The first is it's imprecise.
It's imprecise in that it can't tell you

00:41:22.360 --> 00:41:26.917
exactly what it's doing and it makes
mistakes. Don't use it for life-critical

00:41:27.000 --> 00:41:31.757
decisions. You want a human
who can make mistakes

00:41:31.840 --> 00:41:35.997
consulting a computer. You don't want
the computer flying the airplane. And

00:41:36.080 --> 00:41:37.517
maybe never, but certainly not for a

00:41:37.600 --> 00:41:40.557
long time. It's dynamic in the sense

00:41:40.640 --> 00:41:43.277
that it's changing. It can learn. It can

00:41:43.360 --> 00:41:48.237
assess the situation around it
and change its behavior. This is the AI

00:41:48.320 --> 00:41:51.677
now. It has emergent behavior.

00:41:51.760 --> 00:41:56.637
Emergent means things that come
out that we don't expect.

00:41:56.720 --> 00:42:00.317
Strange things. Now, we've seen
this a bit with social media.

00:42:00.400 --> 00:42:03.597
I don't think anyone expected the
government interference in the elections

00:42:03.680 --> 00:42:06.117
in 2016. That was an emergent behavior.

00:42:06.200 --> 00:42:10.157
It was unexpected by humanity
using these tools. And those

00:42:10.240 --> 00:42:14.237
were not AI-powered.
The final point is it's capable of

00:42:14.320 --> 00:42:18.557
learning. So, you've got
a system using the child

00:42:18.640 --> 00:42:22.437
scenario. You've got a a child that's
got a teddy bear that looks like a teddy

00:42:22.520 --> 00:42:26.197
bear, but it's imprecise, dynamic,
emergent, and capable of learning.

00:42:26.280 --> 00:42:28.277
What's the teddy bear learning?

00:42:28.360 --> 00:42:30.717
So, the teddy bear's watching TV, too.

00:42:30.800 --> 00:42:34.557
And the teddy bear says, "Look, I think
this show sucks." And the kid says, "I

00:42:34.640 --> 00:42:38.277
agree with you." >> [laughter]

00:42:38.360 --> 00:42:43.277
>> Again, we don't understand
the implication, especially on formative

00:42:43.360 --> 00:42:46.637
behaviors. There's a whole
'nother set of arguments

00:42:46.720 --> 00:42:50.397
about national security and how
governments will work and how dominance

00:42:50.480 --> 00:42:53.917
games will play and who will be winners
and losers from this technology that we

00:42:54.000 --> 00:42:56.437
don't really understand.
And we talk about that.

00:42:56.520 --> 00:43:01.117
>> Well, let's actually touch
on that for a second. And

00:43:01.200 --> 00:43:03.837
this is something on the minds of a lot
of folks. Certainly, I know a lot of

00:43:03.920 --> 00:43:07.917
investors who are trying to understand

00:43:08.000 --> 00:43:10.837
AI in the capacity of acting as an

00:43:10.920 --> 00:43:13.277
investor. I'm not saying we go there,

00:43:13.360 --> 00:43:16.877
but you have spoken to Congress about,

00:43:16.960 --> 00:43:19.717
say, China's announced ambitions with

00:43:19.800 --> 00:43:21.437
AI.

00:43:21.520 --> 00:43:24.317
10 to 15 years isn't that far away. It's

00:43:24.400 --> 00:43:27.717
really close, and certain things

00:43:27.800 --> 00:43:31.477
move very slowly. Certain facets of,
say, government move quite slowly,

00:43:31.560 --> 00:43:33.397
regulation.

00:43:33.480 --> 00:43:36.237
What types of corners are most important

00:43:36.320 --> 00:43:40.477
to look around? And I'd love to hear
you speak a little bit more about the

00:43:40.560 --> 00:43:44.237
geopolitical components. >> I
was fortunate to be the chairman of an

00:43:44.320 --> 00:43:48.597
AI commission for Congress. We just
released our report, very proud of it.

00:43:48.680 --> 00:43:53.477
And we studied a lot about where the world
is in AI, not so much on these AGI

00:43:53.560 --> 00:43:57.757
questions, but the the more tactical
things that I've been discussing.

00:43:57.840 --> 00:44:02.277
We concluded that the United
States is slightly ahead of China

00:44:02.360 --> 00:44:06.557
in these areas. That China has
a national program to focus on this.

00:44:06.640 --> 00:44:10.997
They're pouring literally billions
and billions of dollar equivalents into

00:44:11.080 --> 00:44:15.477
this. They generate four times more
engineers than as we do just because of

00:44:15.560 --> 00:44:17.917
population size, and they're extremely

00:44:18.000 --> 00:44:21.317
focused on dominance of AI by 2030,

00:44:21.400 --> 00:44:24.237
which is soon. In our report, we speak a

00:44:24.320 --> 00:44:28.477
great deal about what the government
needs to do to help. It starts with more

00:44:28.560 --> 00:44:31.517
research, access to more
data, making sure our

00:44:31.600 --> 00:44:35.517
values are represented so we don't end
up with systems that have a prejudice

00:44:35.600 --> 00:44:39.477
and so forth and violate both our laws
as well as our morals. We talk about

00:44:39.560 --> 00:44:44.397
partnerships and all of this kind
of stuff. The reason this is so important

00:44:44.480 --> 00:44:48.197
is that pretty much every national
security issue will be tainted or

00:44:48.280 --> 00:44:53.277
controlled by AI in the future.
If you look think about cyber attacks,

00:44:53.360 --> 00:44:56.917
the most obvious war
scenario in the future is the

00:44:57.000 --> 00:45:01.157
following. North Korea decides
to attack the United States.

00:45:01.240 --> 00:45:05.637
It begins its attack.
China decides this is a bad idea

00:45:05.720 --> 00:45:07.917
and blocks the attack.

00:45:08.000 --> 00:45:11.357
America's defenses are awake, alert.

00:45:11.440 --> 00:45:14.637
America announces
a counterattack which stops this.

00:45:14.720 --> 00:45:16.917
The entire war

00:45:17.000 --> 00:45:22.400
that I just described
took about 10 milliseconds.

00:45:22.800 --> 00:45:26.837
Give you another example. In the military,
there's a presumption of human

00:45:26.920 --> 00:45:29.557
control, which is very important.

00:45:29.640 --> 00:45:33.877
I was part of a team that wrote some
of the AI ethics that are now used by the

00:45:33.960 --> 00:45:37.757
US military. Which I just
obviously strongly support.

00:45:37.840 --> 00:45:39.637
And they say we want the principle of

00:45:39.720 --> 00:45:43.037
human control, human authorization.

00:45:43.120 --> 00:45:45.037
So, let me give you an example.

00:45:45.120 --> 00:45:48.837
You're on a ship and a new
kind of missile, let's say a

00:45:48.920 --> 00:45:53.237
very fast hypersonic missile,
is coming in, which people are developing.

00:45:53.320 --> 00:45:56.517
Countries are. And it's
coming in so fast that the ship

00:45:56.600 --> 00:45:58.597
can't see it, the humans can't see it,

00:45:58.680 --> 00:46:00.837
but the AI has figured it out.

00:46:00.920 --> 00:46:04.957
So, the AI says to the commander,

00:46:05.040 --> 00:46:08.117
"In 15 seconds, a missile is going to be

00:46:08.200 --> 00:46:12.997
showing up and it's going to destroy you
and your entire crew. I recommend that

00:46:13.080 --> 00:46:16.957
you press this button
to launch the counterattack."

00:46:17.040 --> 00:46:18.677
And the human says,

00:46:18.760 --> 00:46:22.157
"14, 13, 12, 10,

00:46:22.240 --> 00:46:25.077
what do I do?" And at 3 seconds, that

00:46:25.160 --> 00:46:28.237
human's going to press that button.

00:46:28.320 --> 00:46:31.277
So, the compression of time, the amount

00:46:31.360 --> 00:46:35.397
of data, and the potential for error

00:46:35.480 --> 00:46:38.117
creates a whole set of problems around

00:46:38.200 --> 00:46:40.837
military doctrine.

00:46:40.920 --> 00:46:46.197
Now, everything that we live in today,
deterrence, all of the other things that

00:46:46.280 --> 00:46:49.677
we are all of us are familiar
with, great power competition,

00:46:49.760 --> 00:46:53.237
are concepts that were invented
over the last 100 years ago.

00:46:53.320 --> 00:46:57.797
And Dr. Kissinger actually invented much
of the containment strategy in the '50s

00:46:57.880 --> 00:47:02.797
and '60s before he was in government
as part of a team that sort of invented

00:47:02.880 --> 00:47:06.197
all this work. So, I asked
him, "What does this look

00:47:06.280 --> 00:47:10.717
like now? How would you address this now?"

00:47:10.800 --> 00:47:15.237
And his answer is, "Let's get
an equivalent team of people and try to

00:47:15.320 --> 00:47:20.077
figure these things out."
Containment, for example, doesn't work.

00:47:20.160 --> 00:47:23.917
Containment is about keeping another
country from getting something

00:47:24.000 --> 00:47:25.917
because these algorithms and this

00:47:26.000 --> 00:47:31.237
software is pervasive.
It leaks and it leaks through ideas and

00:47:31.320 --> 00:47:34.237
through discovery,
not just from criminal leaking.

00:47:34.320 --> 00:47:38.557
It's not like Los Alamos
where you could keep a secret.

00:47:38.640 --> 00:47:41.237
So, the way to maintain competitive

00:47:41.320 --> 00:47:45.757
dominance, that is national security,
is to invest in these areas, both in

00:47:45.840 --> 00:47:49.077
terms of data and algorithms,
and to be excellent,

00:47:49.160 --> 00:47:53.557
but also to begin some kind of dialogue

00:47:53.640 --> 00:47:57.637
about what the limitations
of automatic war are.

00:47:57.720 --> 00:47:59.877
So, another example, and I'll make one

00:47:59.960 --> 00:48:02.477
up now. So, the scenario I described of

00:48:02.560 --> 00:48:06.317
the 10 millisecond war

00:48:06.400 --> 00:48:09.477
So, let's say that China in this case

00:48:09.560 --> 00:48:13.517
develops an AGI ahead of everyone else.

00:48:13.600 --> 00:48:16.917
And this AGI is thought to be so powerful

00:48:17.000 --> 00:48:20.957
that it could defeat any of our defenses.

00:48:21.040 --> 00:48:25.397
Then, the logic on the American side
would be to do a preemptive attack to

00:48:25.480 --> 00:48:29.597
prevent that possibility.
And that's destabilizing of great power

00:48:29.680 --> 00:48:31.197
competition.

00:48:31.280 --> 00:48:35.037
So, you want to avoid an arms race.

00:48:35.120 --> 00:48:39.877
An arms race in AI could look a lot like
the arms race that we went through in

00:48:39.960 --> 00:48:44.117
the '50s and '60s.
And people have forgotten how much of

00:48:44.200 --> 00:48:46.837
our military-industrial complex, how

00:48:46.920 --> 00:48:51.917
much money and so forth, was devoted
to build more than 30,000 nuclear weapons,

00:48:52.000 --> 00:48:56.237
all of which could destroy the world many,
many times over. Many, many rounds

00:48:56.320 --> 00:49:00.557
of negotiation got those numbers down
to 3,000, 4,000 such weapons, which are

00:49:00.640 --> 00:49:05.037
still plenty to destroy everything.
It's an example of overreach.

00:49:05.120 --> 00:49:07.117
And so, I worry that because we don't

00:49:07.200 --> 00:49:11.037
have an agreement on even what the rules

00:49:11.120 --> 00:49:14.517
are, what the landscape
of limitations are,

00:49:14.600 --> 00:49:18.757
we don't have diplomats who
can have the conversations.

00:49:18.840 --> 00:49:24.037
And no single national security group,
no single country is going to self-limit

00:49:24.120 --> 00:49:26.957
and say, "Oh, we're not going to do that."
This is not Costa Rica, which

00:49:27.040 --> 00:49:28.917
doesn't have a military.

00:49:29.000 --> 00:49:32.157
So, the natural course of logic will be

00:49:32.240 --> 00:49:34.957
the development of these incredibly fast

00:49:35.040 --> 00:49:37.357
and potentially destabilizing weapons,

00:49:37.440 --> 00:49:40.197
for which we don't have a language.

00:49:40.280 --> 00:49:44.077
>> I think I'll just act
as a stand-in for listeners.

00:49:44.160 --> 00:49:47.677
That sounds very terrifying. It does sound

00:49:47.760 --> 00:49:50.517
not just terrifying, but challenging to

00:49:50.600 --> 00:49:54.277
address. And I I may come back to that

00:49:54.360 --> 00:49:58.437
and questions of policy and geopolitics.

00:49:58.520 --> 00:50:02.677
I want to ask you the personal question
that I alluded to earlier before I let

00:50:02.760 --> 00:50:04.757
that go, which is

00:50:04.840 --> 00:50:09.917
what could you achieve or what would
you want to achieve or ask of AI

00:50:10.000 --> 00:50:11.717
in the future that would allow you to do

00:50:11.800 --> 00:50:14.157
things you cannot do today?

00:50:14.240 --> 00:50:19.277
>> Almost every hard
problem in our society

00:50:19.360 --> 00:50:23.077
is based on either
the computers can't figure it

00:50:23.160 --> 00:50:25.597
out, we don't have enough data,

00:50:25.680 --> 00:50:28.397
we honestly don't know how to solve it.

00:50:28.480 --> 00:50:33.037
I'd like to get some breakthroughs.
So, I'd like for example to get better

00:50:33.120 --> 00:50:36.877
climate models.
In my philanthropy, I funded a group at

00:50:36.960 --> 00:50:38.357
Caltech

00:50:38.440 --> 00:50:40.477
which figured out a way to predict

00:50:40.560 --> 00:50:44.037
climate better using an AI system that

00:50:44.120 --> 00:50:48.077
would model clouds. I didn't know this,
but it turns out clouds are really

00:50:48.160 --> 00:50:51.277
complicated. And so,
they used an AI system to

00:50:51.360 --> 00:50:55.277
approximate how clouds
would behave, which allowed

00:50:55.360 --> 00:50:57.997
them to solve the prediction problem.

00:50:58.080 --> 00:51:03.317
It's called Clima, c l i m a. Over
and over again, I would like the AI system

00:51:03.400 --> 00:51:07.837
to educate me and entertain me

00:51:07.920 --> 00:51:12.237
and keep me curious about
the dynamism of the world.

00:51:12.320 --> 00:51:16.117
Speaking personally, when I
look at the news feeds, we're

00:51:16.200 --> 00:51:20.477
obsessed about politics
and President Trump and so forth,

00:51:20.560 --> 00:51:24.637
to the point where you get the impression
that there's nothing else going on.

00:51:24.720 --> 00:51:28.477
But one of the great things about
humanity and our world is it's

00:51:28.560 --> 00:51:32.477
incredibly dynamic. We never
hear what those people are doing.

00:51:32.560 --> 00:51:36.757
I'd like to see if we could figure out
a way to advance against these really hard

00:51:36.840 --> 00:51:40.077
problems. Let's get some solutions
around mental health. Let's get some

00:51:40.160 --> 00:51:44.997
solutions around drug abuse. Let's
get some solutions around the rise of

00:51:45.080 --> 00:51:49.917
inequality. And let's have our computers
help us with those solutions.

00:51:50.000 --> 00:51:52.437
>> And let's just talk about this [clears
throat] is something you know a

00:51:52.520 --> 00:51:55.677
lot more about than I do, but how

00:51:55.760 --> 00:51:59.317
programming may be able or

00:51:59.400 --> 00:52:01.997
incapable of addressing

00:52:02.080 --> 00:52:06.517
morals and ethics. And I'll
give an example, probably a bad

00:52:06.600 --> 00:52:08.357
example and not technically accurate,

00:52:08.440 --> 00:52:10.637
but it seems like we're already at a

00:52:10.720 --> 00:52:13.837
point with say the development of

00:52:13.920 --> 00:52:16.717
autonomous vehicles where

00:52:16.800 --> 00:52:18.957
questions that might have been presented

00:52:19.040 --> 00:52:22.877
in a freshman philosophy class, like the

00:52:22.960 --> 00:52:25.157
trolley problem, actually need to be

00:52:25.240 --> 00:52:27.597
programmed on some level in the sense

00:52:27.680 --> 00:52:31.037
that cars need to make decisions if

00:52:31.120 --> 00:52:33.357
there is an accident. They need to say

00:52:33.440 --> 00:52:35.997
choose between going on the sidewalk at

00:52:36.080 --> 00:52:39.077
high speed and hitting three

00:52:39.160 --> 00:52:43.677
elderly or swerving right and hitting
one child, let's just say. Not that we

00:52:43.760 --> 00:52:45.957
would ever want to be in that type of

00:52:46.040 --> 00:52:48.957
situation, but these things happen.

00:52:49.040 --> 00:52:51.517
How do you foresee computer scientists

00:52:51.600 --> 00:52:54.157
attempting to, if this is even going to

00:52:54.240 --> 00:52:57.037
be an objective, instill ethics and

00:52:57.120 --> 00:53:01.317
morals into AI? And does it just require

00:53:01.400 --> 00:53:06.517
a level of self-awareness that perhaps
we don't have? I'm curious how you might

00:53:06.600 --> 00:53:09.557
think about that. >> There are
many, many computer scientists

00:53:09.640 --> 00:53:13.437
working on this. Every major computer
science program I'm aware of has a

00:53:13.520 --> 00:53:16.357
computer science
and ethics research project.

00:53:16.440 --> 00:53:18.117
There are a lot of problems. One of the

00:53:18.200 --> 00:53:20.277
first problems is the computer science

00:53:20.360 --> 00:53:25.597
and AI are based on learning
data, literally learning,

00:53:25.680 --> 00:53:26.797
and therefore the information that

00:53:26.880 --> 00:53:29.197
they're learning has biases.

00:53:29.280 --> 00:53:32.677
So, whatever prejudice and bias and

00:53:32.760 --> 00:53:34.757
religious problems that society has

00:53:34.840 --> 00:53:37.437
around it, the computer will absorb.

00:53:37.520 --> 00:53:42.237
So, the research is how to identify
those biases, cuz remember in many cases

00:53:42.320 --> 00:53:45.357
these systems cannot
tell you what they know

00:53:45.440 --> 00:53:49.557
and how to mediate them.
It's pretty clear to me that there's

00:53:49.640 --> 00:53:53.557
going to end up being the system that
knows everything, and then there's going

00:53:53.640 --> 00:53:57.157
to be a supervisory system that limits it.

00:53:57.240 --> 00:54:00.677
So, two different systems. One will be
the knowledge system, and another one

00:54:00.760 --> 00:54:04.197
which will keep it within some guardrails.

00:54:04.280 --> 00:54:08.597
So, the question that was asked is not
a permissible question, don't send it to

00:54:08.680 --> 00:54:12.477
the system. And you're
going to have all sorts of

00:54:12.560 --> 00:54:15.317
problems. My favorite
self-driving car problem

00:54:15.400 --> 00:54:19.797
is here in New York. Eventually,

00:54:19.880 --> 00:54:24.077
all the cars are self-driving
because it's such a crowded grid, and

00:54:24.160 --> 00:54:28.317
traffic moves perfectly.
The engineers at Google and everywhere

00:54:28.400 --> 00:54:33.077
else have figured out exactly how to
optimize under a set of assumptions, the

00:54:33.160 --> 00:54:37.197
aggregate and average delivery time
of a human from place to place. It's a

00:54:37.280 --> 00:54:39.477
perfect computer science solution.

00:54:39.560 --> 00:54:42.197
So, now we've got a woman
who's going into labor

00:54:42.280 --> 00:54:44.317
who needs to get there faster.

00:54:44.400 --> 00:54:48.037
Is there a button in the car that says,

00:54:48.120 --> 00:54:52.837
"I'm pregnant, I'm about to give birth,
and I need to get there faster." Okay,

00:54:52.920 --> 00:54:56.037
so she presses the button,
and somehow it works.

00:54:56.120 --> 00:54:58.397
How do we make sure that a gentleman

00:54:58.480 --> 00:55:00.917
who's lying

00:55:01.000 --> 00:55:03.637
cannot press the same button?

00:55:03.720 --> 00:55:08.077
In the trolley problem, all of these are
of the form, how do you choose one life

00:55:08.160 --> 00:55:11.917
for the other?
The correct answer is, don't kill

00:55:12.000 --> 00:55:15.157
anything. And so, we're
going to have to find ways

00:55:15.240 --> 00:55:18.917
to avoid being in the situation where
you have to choose between three old

00:55:19.000 --> 00:55:22.997
people and one child.
And these are the great debates of

00:55:23.080 --> 00:55:26.637
philosophers, and how do we value life,
and so forth. But, I think we should

00:55:26.720 --> 00:55:30.317
start with the premise that the system
should be designed such that it

00:55:30.400 --> 00:55:32.717
optimizes overall happiness and overall

00:55:32.800 --> 00:55:35.837
wealth, in the societal sense of wealth.

00:55:35.920 --> 00:55:39.797
Uh as an example,
we tolerate double parking.

00:55:39.880 --> 00:55:42.477
We tolerate people speeding. We don't

00:55:42.560 --> 00:55:46.877
track every car and give them
an automatic ticket every time they go

00:55:46.960 --> 00:55:50.277
above the 30 mile an hour
speed limit by 1 mile.

00:55:50.360 --> 00:55:52.677
But technically that would be easy.

00:55:52.760 --> 00:55:56.477
If you want to eliminate
the vast majority of crime

00:55:56.560 --> 00:55:59.357
in our society, put cameras in every

00:55:59.440 --> 00:56:02.637
public space with face recognition.

00:56:02.720 --> 00:56:05.157
It's by the way illegal to do so.

00:56:05.240 --> 00:56:06.877
But if you really care about crime to

00:56:06.960 --> 00:56:09.117
the exclusion of personal freedom and

00:56:09.200 --> 00:56:14.357
the freedom of of not being surveilled,
then we could do it. So the technology

00:56:14.440 --> 00:56:17.517
gives you these choices and those are
not choices that computer scientists

00:56:17.600 --> 00:56:20.997
should make. Those are
choices for society. I use the

00:56:21.080 --> 00:56:24.157
surveillance cameras
as an example to say that

00:56:24.240 --> 00:56:26.877
in Britain they're very widely accepted.

00:56:26.960 --> 00:56:30.317
When you're in Britain on a street,
you're on camera. In the United States,

00:56:30.400 --> 00:56:33.917
it's partial. Sometimes you are,
sometimes you're not. But in Germany

00:56:34.000 --> 00:56:38.957
with a long history of the Stasi,
they are very violently opposed to such

00:56:39.040 --> 00:56:42.317
surveillance. These are three different
democracies that have made the choices

00:56:42.400 --> 00:56:45.957
in different ways. So I don't think
that computer scientists and the tech

00:56:46.040 --> 00:56:49.037
industry should make these decisions,
but I do think what we should tell

00:56:49.120 --> 00:56:52.717
people is our tools give
you this range of choices.

00:56:52.800 --> 00:56:56.517
If you look at leadership in China,
China is a couple generations ahead in

00:56:56.600 --> 00:57:00.037
surveillance. They can
actually spot you in your gate

00:57:00.120 --> 00:57:04.157
now, the way you walk, not just your face
recognition. This is not something

00:57:04.240 --> 00:57:08.877
we want to lead in. >> I'm just
letting that all sink in for a second.

00:57:08.960 --> 00:57:12.197
>> [laughter] >> This is one of those
conversations for me. It's great.

00:57:12.280 --> 00:57:14.877
There's so much to chew on and I'm going

00:57:14.960 --> 00:57:17.637
to take a 90 degree turn

00:57:17.720 --> 00:57:19.557
back to

00:57:19.640 --> 00:57:22.677
the physicist and the biologist

00:57:22.760 --> 00:57:26.997
and others who will be able to

00:57:27.080 --> 00:57:29.677
consult AI in the way I think it was

00:57:29.760 --> 00:57:34.077
presented to help assimilate
new information, new

00:57:34.160 --> 00:57:37.637
developments, get up to speed.
They can't possibly

00:57:37.720 --> 00:57:42.317
even with physical constraints, keep up
or digest the amount they would like to

00:57:42.400 --> 00:57:45.517
digest. Now, many people, Dr. Kissinger

00:57:45.600 --> 00:57:49.437
certainly over decades
and many, many others,

00:57:49.520 --> 00:57:54.717
spend [snorts] a lot of their time
ingesting information. And that could be

00:57:54.800 --> 00:57:56.157
in the form of reading, it could be in

00:57:56.240 --> 00:57:59.837
the form of conversations like this.

00:57:59.920 --> 00:58:02.797
I suppose there are two questions. One is,

00:58:02.880 --> 00:58:05.157
how do you think

00:58:05.240 --> 00:58:07.477
thought workers or humans will spend

00:58:07.560 --> 00:58:09.877
their time once that ingestion is

00:58:09.960 --> 00:58:12.437
dramatically reduced or removed? And

00:58:12.520 --> 00:58:17.837
along with that, if we get
to the point where that is possible,

00:58:17.920 --> 00:58:22.597
and there's a question of how it
would be conveyed to the human,

00:58:22.680 --> 00:58:25.237
how far are we from there simply being a

00:58:25.320 --> 00:58:28.877
direct brain-computer interface where

00:58:28.960 --> 00:58:33.477
that information is somehow seamlessly

00:58:33.560 --> 00:58:38.957
integrated into our consciousness
without being spoken or imparted to us

00:58:39.040 --> 00:58:42.837
externally? >> To answer
the the last part first,

00:58:42.920 --> 00:58:45.597
we are in our industry, we're working on

00:58:45.680 --> 00:58:50.437
direct brain connections of one kind
or another. They're all in startups.

00:58:50.520 --> 00:58:52.277
They're all speculative at this point.

00:58:52.360 --> 00:58:55.597
They've shown some gains.

00:58:55.680 --> 00:58:59.557
If you think about 200 years from now,

00:58:59.640 --> 00:59:03.437
to pick a random number,
it's probable that we'll know the

00:59:03.520 --> 00:59:05.757
complete details of the human brain, how

00:59:05.840 --> 00:59:08.477
it's wired, how it works.

00:59:08.560 --> 00:59:10.517
So, if you think far enough in advance,

00:59:10.600 --> 00:59:13.877
well past our lifetimes, it's probable

00:59:13.960 --> 00:59:17.717
that we'll know a great deal about how
to manipulate the brain externally to

00:59:17.800 --> 00:59:21.397
the brain. We'll understand how it
works, we'll understand the wiring,

00:59:21.480 --> 00:59:25.277
maybe there'll be some attachment and so
forth. Most people think that that will

00:59:25.360 --> 00:59:27.917
occur. No one has any idea when. When

00:59:28.000 --> 00:59:32.917
you go back to the the gains, you want
to get the basis of what the next 20

00:59:33.000 --> 00:59:34.917
years look like. And I think it's fair

00:59:35.000 --> 00:59:37.837
to say that we're going to be awash in

00:59:37.920 --> 00:59:40.637
information and misinformation

00:59:40.720 --> 00:59:42.997
at a scale that is overwhelming and it's

00:59:43.080 --> 00:59:48.157
already overwhelming.
So, the most likely scenario

00:59:48.240 --> 00:59:52.117
is that each of us will have
the equivalent of an AI communications

00:59:52.200 --> 00:59:56.397
assistant that will watch,
like while you and I

00:59:56.480 --> 01:00:00.717
are speaking, my AI assistant
is watching what's going on,

01:00:00.800 --> 01:00:03.877
and it knows my preferences, it knows
what I care about, and it has a good

01:00:03.960 --> 01:00:06.117
sense of judgment.

01:00:06.200 --> 01:00:08.237
And when we're finished speaking, it'll

01:00:08.320 --> 01:00:13.317
say, "By the way, over here in Arkansas,
the following thing happened that you

01:00:13.400 --> 01:00:18.397
might want to check out."
More importantly, this AI assistant will

01:00:18.480 --> 01:00:23.317
battle with the misinformation assistants,
and it'll say, "Prove to me that you're

01:00:23.400 --> 01:00:27.837
a human before I expose you to my human."

01:00:27.920 --> 01:00:32.077
And so, you can imagine a scenario
where the solution to the misinformation

01:00:32.160 --> 01:00:34.357
problem and the solution to this

01:00:34.440 --> 01:00:36.877
information space problem is that each

01:00:36.960 --> 01:00:41.757
of us has our own AI assistant
that think of it as the equivalent of a

01:00:41.840 --> 01:00:44.957
phone, although in practice it's
a supercomputer that's accessed through

01:00:45.040 --> 01:00:49.037
your phone or equivalent,
that sort of keeps you sane.

01:00:49.120 --> 01:00:52.077
If you go back to my earlier
comments about children,

01:00:52.160 --> 01:00:57.597
we have no idea what the rule should be
for that assistant. So, let me give you

01:00:57.680 --> 01:01:00.997
an example. We've learned,
unfortunately, in the last few years

01:01:01.080 --> 01:01:03.797
that there are still horrific racial

01:01:03.880 --> 01:01:08.077
prejudice, horrifically hateful people.

01:01:08.160 --> 01:01:11.477
Well, they get an assistant, too.

01:01:11.560 --> 01:01:15.917
Is their assistant going
to pattern their racism

01:01:16.000 --> 01:01:19.917
or their misogyny or their
violence against children or

01:01:20.000 --> 01:01:23.957
all these sort of hateful behaviors, cuz
these are part of humanity too. Or will

01:01:24.040 --> 01:01:27.357
there be some regulation that says
that your assistant has to be politically

01:01:27.440 --> 01:01:31.237
correct? It has to say,
are you a he, a she, or

01:01:31.320 --> 01:01:35.357
you know, in other words,
how will society resolve that?

01:01:35.440 --> 01:01:39.677
And my prediction will be that each
government and each culture will adopt

01:01:39.760 --> 01:01:41.757
different rules about them.

01:01:41.840 --> 01:01:45.117
But they're not going to be unregulated.

01:01:45.200 --> 01:01:48.677
What we've learned, and I've
learned really the hard way,

01:01:48.760 --> 01:01:51.557
is that the technology that I helped

01:01:51.640 --> 01:01:55.757
work on and that is being invented now
is no longer optional. 15 years ago I

01:01:55.840 --> 01:02:00.077
used to give this speech saying, "Look,
you hate the internet, turn your phone

01:02:00.160 --> 01:02:04.117
off. Have dinner with your family. Do

01:02:04.200 --> 01:02:07.877
whatever you're going to do, but get off."

01:02:07.960 --> 01:02:10.437
And that's not a practical

01:02:10.520 --> 01:02:15.037
answer today. The internet
is no longer optional. It's

01:02:15.120 --> 01:02:18.637
essential, partly because of the pandemic,
but because of e-commerce and

01:02:18.720 --> 01:02:21.357
business and knowledge and so forth.

01:02:21.440 --> 01:02:23.237
So, it's going to get regulated, and the

01:02:23.320 --> 01:02:25.957
question is how and under what terms.

01:02:26.040 --> 01:02:30.797
>> I've I believe read you use the analogy

01:02:30.880 --> 01:02:35.357
of the telephone. Maybe I'm making
that up, but in the sense that

01:02:35.440 --> 01:02:37.437
just because there are bad actors who

01:02:37.520 --> 01:02:40.317
can use the telephone for

01:02:40.400 --> 01:02:42.677
wrongdoing, for crime, [snorts] etc., it

01:02:42.760 --> 01:02:45.357
doesn't mean we eliminate the telephone.

01:02:45.440 --> 01:02:47.437
It means that we have means of

01:02:47.520 --> 01:02:50.957
regulation and enforcement and so on. I

01:02:51.040 --> 01:02:54.717
guess the telephone
by comparison seems to be such a

01:02:54.800 --> 01:02:58.037
clean, although increasingly, maybe not,

01:02:58.120 --> 01:03:00.197
discrete system.

01:03:00.280 --> 01:03:02.757
What are the most important next steps

01:03:02.840 --> 01:03:06.037
from your perspective with respect to

01:03:06.120 --> 01:03:10.957
even thinking about regulation, the best
questions to ask, or just concrete

01:03:11.040 --> 01:03:13.837
steps? >> I always like
to use self-driving cars,

01:03:13.920 --> 01:03:17.077
cuz everyone can relate to this. So,

01:03:17.160 --> 01:03:22.237
in California, there's a concept
called a rolling stop where you

01:03:22.320 --> 01:03:25.117
>> [laughter] >> The California roll, yes.

01:03:25.200 --> 01:03:28.957
>> And when you come to the stop sign,
you sort of forget to fully stop. You roll

01:03:29.040 --> 01:03:32.277
through. So, the policeman comes over
and says, "You did the rolling stop."

01:03:32.360 --> 01:03:35.957
And I say to the policeman,
"Sir, I did not."

01:03:36.040 --> 01:03:40.357
And then it said, "Who did?"
And then the car says, "I did, sir."

01:03:40.440 --> 01:03:45.157
And the policeman says, "Why?"
And the car says, "I don't know."

01:03:45.240 --> 01:03:48.037
>> [laughter] >> Now, >> Very
frustrating for the police officer.

01:03:48.120 --> 01:03:51.037
>> Yes, so So, who gets the ticket?

01:03:51.120 --> 01:03:56.117
So, your choices are the human,
the car itself, the manufacturer of the

01:03:56.200 --> 01:03:59.477
car, or the data that the car was trained

01:03:59.560 --> 01:04:03.437
with. So, that's the debate to have.

01:04:03.520 --> 01:04:08.077
Give you another example. I feel very
strongly in favor of free speech for

01:04:08.160 --> 01:04:11.477
humans, but I don't feel very strongly for

01:04:11.560 --> 01:04:15.477
amplification by computers.
And so, what we're seeing now is we're

01:04:15.560 --> 01:04:18.277
seeing humans who have wacky, false,

01:04:18.360 --> 01:04:23.157
conspiratorial ideas, what have you,
get picked up and amplified in these

01:04:23.240 --> 01:04:26.797
systems. That drives
everyone crazy because we

01:04:26.880 --> 01:04:28.957
can't tell the difference between a

01:04:29.040 --> 01:04:32.277
genuine social movement and concern

01:04:32.360 --> 01:04:34.757
versus a single crazy person who has an

01:04:34.840 --> 01:04:39.357
amplified idea that seems plausible,
but is basically false. We have no way of

01:04:39.440 --> 01:04:42.557
falsifying such things.
That's something that's got to get

01:04:42.640 --> 01:04:44.597
sorted out. We can't live in an

01:04:44.680 --> 01:04:47.317
information space

01:04:47.400 --> 01:04:50.597
that is so full of misinformation

01:04:50.680 --> 01:04:54.597
and manipulation that we
can't get through the day.

01:04:54.680 --> 01:04:58.197
When we go back in the book to talk
about the Renaissance, we say this is a

01:04:58.280 --> 01:05:02.597
new epoch because in addition to this
misinformation that we've been speaking

01:05:02.680 --> 01:05:06.997
about, we also have never had a situation

01:05:07.080 --> 01:05:11.277
in our human experience where there
was an intelligence that was similar to

01:05:11.360 --> 01:05:14.437
ours, but not the same,
that was non-human.

01:05:14.520 --> 01:05:19.157
And so, imagine a situation where these
intelligences exist and they can be

01:05:19.240 --> 01:05:24.037
consulted. Well, who gets to consult them?
What happens to their answers?

01:05:24.120 --> 01:05:27.717
So, for example, if it's an oracle,
do you have a rule that every answer from

01:05:27.800 --> 01:05:30.357
the oracle is published to the world?

01:05:30.440 --> 01:05:33.837
Because it's presumably beneficial.
But, what if it's a private question or a

01:05:33.920 --> 01:05:37.197
secret? What if it's being used
in national security? Again, these are

01:05:37.280 --> 01:05:39.797
questions that we have not resolved. So,

01:05:39.880 --> 01:05:41.957
what we say is, these are really hard

01:05:42.040 --> 01:05:44.757
questions. What happens when the AI

01:05:44.840 --> 01:05:47.437
perceives aspects of reality that humans

01:05:47.520 --> 01:05:48.997
do not?

01:05:49.080 --> 01:05:52.437
That it sees a connection. There are

01:05:52.520 --> 01:05:56.517
physicists who talk about
new universes and new worlds.

01:05:56.600 --> 01:05:58.957
And you can imagine a situation where

01:05:59.040 --> 01:06:03.797
there's some thing that we
as humans cannot conceive of

01:06:03.880 --> 01:06:08.157
because of our limited intelligence, but
the computer detects it and then we say,

01:06:08.240 --> 01:06:11.917
"Oh my god." In math, there's a

01:06:12.000 --> 01:06:14.557
long story about an object that lives in

01:06:14.640 --> 01:06:18.997
a two-dimensional world and the
third dimension appears, but if

01:06:19.080 --> 01:06:22.077
you're in a two-dimensional world, you
never see the third dimension, so you're

01:06:22.160 --> 01:06:26.237
always confused. So,
what if this AI can find a

01:06:26.320 --> 01:06:30.157
dimensionality of the world that we
as humans, not a single one of us can

01:06:30.240 --> 01:06:34.197
understand? And we're now confused.

01:06:34.280 --> 01:06:38.677
We don't know how to handle that.
What happens, another example is

01:06:38.760 --> 01:06:42.317
Dr. Kissinger believes that

01:06:42.400 --> 01:06:47.037
I'll paraphrase.
When something is not understandable by

01:06:47.120 --> 01:06:48.597
humans,

01:06:48.680 --> 01:06:52.197
they will perceive it as an act of God.

01:06:52.280 --> 01:06:55.557
Or they will resist it.

01:06:55.640 --> 01:06:59.637
And you can imagine a situation
where you end up with nativists

01:06:59.720 --> 01:07:01.717
who decide that this world that I'm

01:07:01.800 --> 01:07:05.437
describing is so impossibly unpleasant

01:07:05.520 --> 01:07:10.117
that they turn it all off and they go
to the equivalent of uh the woods of

01:07:10.200 --> 01:07:13.637
Pennsylvania and they say leave me alone.

01:07:13.720 --> 01:07:16.637
And you have other people who learn to

01:07:16.720 --> 01:07:21.357
drive it and manipulate
it for good and evil.

01:07:21.440 --> 01:07:26.397
>> You'll have a choice and then
one final scenario to think about

01:07:26.480 --> 01:07:28.237
goes like this.

01:07:28.320 --> 01:07:33.597
Let's say you're older, kids are grown
and it's time for you to take a break

01:07:33.680 --> 01:07:37.037
from the world. So, you put on headphones

01:07:37.120 --> 01:07:41.397
and VR glasses and in those
VR glasses and headphones

01:07:41.480 --> 01:07:44.597
you have a life of you
as a much younger person.

01:07:44.680 --> 01:07:49.477
A much more beautiful, handsome,
wealthier, stronger, whatever

01:07:49.560 --> 01:07:53.197
with the friends that you remember
of the time recreated even though they may

01:07:53.280 --> 01:07:57.117
have passed away.
That might be a more fun life for you

01:07:57.200 --> 01:08:01.877
every day than the life
that you have in this scenario.

01:08:01.960 --> 01:08:06.277
What happens when we lose people?
I call that crossing to the other side.

01:08:06.360 --> 01:08:09.797
When they wake up in the morning,
they just want to be over there.

01:08:09.880 --> 01:08:11.717
That will happen, too.

01:08:11.800 --> 01:08:14.517
>> It strikes me that

01:08:14.600 --> 01:08:17.437
it is science fiction can be

01:08:17.520 --> 01:08:22.117
prescient in some instances and it
doesn't make it future fiction, it just

01:08:22.200 --> 01:08:24.957
makes it current day fiction, but there's
[clears throat] I don't want to

01:08:25.040 --> 01:08:29.757
say predictive power, but I think
of really good science fiction

01:08:29.840 --> 01:08:33.157
of course subjective, but Snow
Crash, the description of the

01:08:33.240 --> 01:08:36.477
metaverse or you have some
of William Gibson's work.

01:08:36.560 --> 01:08:40.717
I'm just curious, do you or have you
read science fiction much yourself?

01:08:40.800 --> 01:08:42.797
>> Well, Neuromancer is particularly good.

01:08:42.880 --> 01:08:46.477
If you look at Seven
Eves which is a good book.

01:08:46.560 --> 01:08:51.477
It's an extraordinary composition of of
the importance of humanity surviving the

01:08:51.560 --> 01:08:56.397
destruction of the earth.
And in Seven Eves toward the end

01:08:56.480 --> 01:08:58.357
in their science fiction, they have

01:08:58.440 --> 01:09:01.037
assistants

01:09:01.120 --> 01:09:03.477
that have a funny name that serve this

01:09:03.560 --> 01:09:08.397
function that I'm describing.
So, as usual, science fiction

01:09:08.480 --> 01:09:10.277
anticipates many of the things in

01:09:10.360 --> 01:09:14.037
technology. But, if you think about it,

01:09:14.120 --> 01:09:18.637
we believe today, from the Reformation,

01:09:18.720 --> 01:09:20.957
that we have the sole power of

01:09:21.040 --> 01:09:26.117
understanding reality. But,
at some point, that's not going to

01:09:26.200 --> 01:09:29.837
be true. When that happens, what will our

01:09:29.920 --> 01:09:33.677
self-conception look like?

01:09:33.760 --> 01:09:38.197
How will we conceive of ourselves?
How will we organize ourselves?

01:09:38.280 --> 01:09:41.917
>> How will we value ourselves?
Let me ask you

01:09:42.000 --> 01:09:44.197
some follow-up questions on this word

01:09:44.280 --> 01:09:47.797
reality and perception because it makes

01:09:47.880 --> 01:09:51.837
me think, this conversation makes me
think of, and maybe getting the the full

01:09:51.920 --> 01:09:55.037
name wrong, so I'll I'll correct it in the

01:09:55.120 --> 01:09:57.477
show notes if I do, but I believe

01:09:57.560 --> 01:10:00.757
his name is Donald Hoffman, who is a

01:10:00.840 --> 01:10:05.517
cognitive scientist, also computer
scientist at UC Irvine, and

01:10:05.600 --> 01:10:09.357
he has a book that's called, I believe,
The Case Against Reality, and also a TED

01:10:09.440 --> 01:10:10.557
Talk,

01:10:10.640 --> 01:10:14.837
and the general exploration focuses on

01:10:14.920 --> 01:10:17.757
how easy it is to prove and demonstrate

01:10:17.840 --> 01:10:22.597
that we are optimized for very few
things, right? Reproduction principal

01:10:22.680 --> 01:10:23.677
among them,

01:10:23.760 --> 01:10:26.957
and that what we perceive is not some

01:10:27.040 --> 01:10:31.597
objective reality, and that if you were
a mantis shrimp, for instance, you would

01:10:31.680 --> 01:10:35.077
see things very differently. They have
these incredible optic systems, and so

01:10:35.160 --> 01:10:36.477
on and so forth.

01:10:36.560 --> 01:10:39.637
What does the word reality mean to you,

01:10:39.720 --> 01:10:43.077
and how do you think, personally, that

01:10:43.160 --> 01:10:47.597
might change once we are
deeper into the era of

01:10:47.680 --> 01:10:52.477
well-developed AI? >> 50 years
ago, reality was television and

01:10:52.560 --> 01:10:54.517
normal life.

01:10:54.600 --> 01:10:58.197
Today, reality is this online world that

01:10:58.280 --> 01:11:01.077
is constantly demanding attention,

01:11:01.160 --> 01:11:05.797
constantly full of stressors of one kind
or another, constantly demanding your

01:11:05.880 --> 01:11:08.597
engagement because
that's the way it's built.

01:11:08.680 --> 01:11:12.557
The engagement is around monetization,
uh spreading information and

01:11:12.640 --> 01:11:14.797
misinformation, and so forth.

01:11:14.880 --> 01:11:17.997
Everyone I know is being driven crazy by

01:11:18.080 --> 01:11:19.957
the explosion of that.

01:11:20.040 --> 01:11:22.797
I think that in the next decade,

01:11:22.880 --> 01:11:25.837
the explosion will continue and the

01:11:25.920 --> 01:11:28.997
solutions will need to to be developed

01:11:29.080 --> 01:11:31.597
because more and more the digital world

01:11:31.680 --> 01:11:36.877
becomes the real world.
An example that I used, which is uh for

01:11:36.960 --> 01:11:39.757
example, you can see it
in the movie Ready Player One,

01:11:39.840 --> 01:11:41.637
where people cross over and they spend

01:11:41.720 --> 01:11:46.957
most of their time in a virtual world,
is true for many, many teenagers.

01:11:47.040 --> 01:11:49.837
And there's every reason to believe
in the next 10 or 20 years those worlds

01:11:49.920 --> 01:11:53.397
will become extraordinarily sophisticated.

01:11:53.480 --> 01:11:56.437
So, I'll give you a simple example.

01:11:56.520 --> 01:11:59.037
At Google, because there
was so much going on, we

01:11:59.120 --> 01:12:02.357
developed a pattern where people would
be in meetings, but they would also be

01:12:02.440 --> 01:12:04.917
doing their email at the same time.

01:12:05.000 --> 01:12:07.877
And we became very good at doing two

01:12:07.960 --> 01:12:12.677
things at once, and this was a cultural
norm. But I had to put in a rule that

01:12:12.760 --> 01:12:16.237
when you had outsiders in the room,
you couldn't do this

01:12:16.320 --> 01:12:20.237
because it was seen as so incredibly rude,

01:12:20.320 --> 01:12:21.677
which it is,

01:12:21.760 --> 01:12:24.117
but it was a cultural norm that we

01:12:24.200 --> 01:12:27.357
adopted and invented and worked well.

01:12:27.440 --> 01:12:31.357
So, you can see that the addiction cycle

01:12:31.440 --> 01:12:34.717
in our world is largely now locked
into addiction cycles, whether it's

01:12:34.800 --> 01:12:38.117
drug-related or whether it's
technology-related or it's

01:12:38.200 --> 01:12:39.997
attention-related. I mean, again, humans

01:12:40.080 --> 01:12:43.397
are like this. I am very, very concerned

01:12:43.480 --> 01:12:46.157
that we will lose perspective on what

01:12:46.240 --> 01:12:51.437
reality really is.
Humans are not built for the kind of

01:12:51.520 --> 01:12:54.557
stressors that occur every day.
We just aren't

01:12:54.640 --> 01:12:58.277
organized. We're organized around
the campfire and the lion and that sort of

01:12:58.360 --> 01:13:00.677
thing, which didn't occur at warp speed.

01:13:00.760 --> 01:13:03.877
And so, I'm assuming that this pressure

01:13:03.960 --> 01:13:05.797
on human development,

01:13:05.880 --> 01:13:08.277
this sort of craziness, this constant

01:13:08.360 --> 01:13:11.557
anxiety, will lead to more neuroses,

01:13:11.640 --> 01:13:14.917
more paranoia, not less.

01:13:15.000 --> 01:13:19.677
And that people will then have to figure
out a way, just in the same way that

01:13:19.760 --> 01:13:23.717
parents are trying to keep their kids
offline some of the time, cuz kids are

01:13:23.800 --> 01:13:27.677
online from the moment they wake up
to the moment they go to sleep, kids want

01:13:27.760 --> 01:13:31.037
them go play ball for an hour,
I'm taking your phone away. We're going to

01:13:31.120 --> 01:13:32.877
have to do the same thing for adults.

01:13:32.960 --> 01:13:36.357
>> What rules or

01:13:36.440 --> 01:13:39.317
constraints do you have for yourself

01:13:39.400 --> 01:13:43.277
around social media or other types of

01:13:43.360 --> 01:13:47.197
digital stimuli?
>> One of my friends said that he stopped

01:13:47.280 --> 01:13:51.877
watching television because
he reads all his news online.

01:13:51.960 --> 01:13:55.237
And I thought, okay, well, that's
a reasonable time trade-off. Of course,

01:13:55.320 --> 01:13:58.117
you're you're changing
one stressor for another,

01:13:58.200 --> 01:14:00.037
but that says, I only have time for one,

01:14:00.120 --> 01:14:03.197
I've chosen the online world.

01:14:03.280 --> 01:14:07.077
And so, I think the first question
is there's limited time. And then the

01:14:07.160 --> 01:14:11.757
second thing is you've got to have some
rules. So, for me, I'm online all the

01:14:11.840 --> 01:14:15.037
time, but I try to turn
it off during dinner.

01:14:15.120 --> 01:14:17.757
I can usually get it done for about an

01:14:17.840 --> 01:14:21.757
hour. So, the question is, when you go

01:14:21.840 --> 01:14:25.637
without it for a while,
give you another example, in 2012,

01:14:25.720 --> 01:14:29.877
a group of us went to North Korea and we
went from Beijing. And this was at a

01:14:29.960 --> 01:14:34.517
time it was legal to do so
as an American, and we left our

01:14:34.600 --> 01:14:38.237
phones in the Beijing
airport in a trusted person.

01:14:38.320 --> 01:14:41.077
And we sat in the lounge as the plane

01:14:41.160 --> 01:14:45.597
was about to take us to Pyongyang
without our phones, and it's It's

01:14:45.680 --> 01:14:50.757
feeling. And by the time we got there and
got settled in the hotel, it's a group of

01:14:50.840 --> 01:14:54.157
maybe 10, we started
talking to each other, which

01:14:54.240 --> 01:14:56.317
we would never have done.

01:14:56.400 --> 01:14:59.517
And within 3 days, we were best friends.

01:14:59.600 --> 01:15:04.397
The moment we got back to Beijing
and got our phones, we were on them.

01:15:04.480 --> 01:15:09.557
We lost all context of what we were
doing, and we were back in the soup.

01:15:09.640 --> 01:15:13.237
So, there's going to be some in the same

01:15:13.320 --> 01:15:17.437
sense that people go to spa vacations,
there's going to have to be some

01:15:17.520 --> 01:15:21.237
equivalent of a spa
vacation from your devices.

01:15:21.320 --> 01:15:23.517
>> Yeah, I'm wondering what what type of

01:15:23.600 --> 01:15:28.917
agency humans will be able to harness to

01:15:29.000 --> 01:15:32.277
position themselves competitively in the

01:15:32.360 --> 01:15:36.117
world of AI. It brings up so many
questions. I mean, such as the AI

01:15:36.200 --> 01:15:39.917
assistant, will there be
tiered assistants, and will

01:15:40.000 --> 01:15:44.077
the well-off be able to afford better
assistants? I mean, certainly, the

01:15:44.160 --> 01:15:48.477
better-off have different types
of access now that help them to separate

01:15:48.560 --> 01:15:52.837
signal from noise better than
the majority. So, that will probably

01:15:52.920 --> 01:15:55.957
continue to be the case.
Do you think there And there's no right

01:15:56.040 --> 01:15:59.877
answer here, by the way. Do you think
there is anything uniquely human that

01:15:59.960 --> 01:16:01.477
ultimately

01:16:01.560 --> 01:16:03.237
we will be able to

01:16:03.320 --> 01:16:07.357
find value in in a world of

01:16:07.440 --> 01:16:09.357
of AGI, for instance?

01:16:09.440 --> 01:16:12.717
>> A lot of people are speculating

01:16:12.800 --> 01:16:16.357
that the eventual future is a much

01:16:16.440 --> 01:16:21.197
richer world where
people are largely idle,

01:16:21.280 --> 01:16:25.317
and that the units of production are
so efficient, this is food production,

01:16:25.400 --> 01:16:29.117
buildings, and so forth,
that everyone can live like a millionaire,

01:16:29.200 --> 01:16:31.997
which means they don't have to work.
And a lot of people would prefer not to

01:16:32.080 --> 01:16:35.717
work. Many of the jobs they find not
that interesting, and they have hobbies

01:16:35.800 --> 01:16:38.717
that are more interesting
to them, and so forth.

01:16:38.800 --> 01:16:42.997
It's possible that will could true.
It's also possible that that world is

01:16:43.080 --> 01:16:47.397
dystopian because humans
need meaning. So, I think

01:16:47.480 --> 01:16:49.357
the answer to your question has a lot to

01:16:49.440 --> 01:16:52.397
do with whether the systems produce more

01:16:52.480 --> 01:16:54.837
meaning for humans

01:16:54.920 --> 01:16:59.597
or less meaning.
If the computer replaces me, that's less

01:16:59.680 --> 01:17:03.237
meaning. If the computer
augments me, it's more

01:17:03.320 --> 01:17:07.477
meaning. This is true
at every level of society.

01:17:07.560 --> 01:17:10.797
So, there will be a small
number of countries.

01:17:10.880 --> 01:17:14.077
The US will be one, China will be another,
there will be others which will

01:17:14.160 --> 01:17:16.557
be the leaders in these technologies.

01:17:16.640 --> 01:17:20.677
And because the way network platforms
work, those leaders will get far ahead

01:17:20.760 --> 01:17:22.877
of the other countries.

01:17:22.960 --> 01:17:25.037
So, there is going to be a division

01:17:25.120 --> 01:17:28.077
between the AI-enabled

01:17:28.160 --> 01:17:30.037
powerhouse countries

01:17:30.120 --> 01:17:32.277
and the following countries who are

01:17:32.360 --> 01:17:35.677
using it but not inventing it.

01:17:35.760 --> 01:17:38.437
That divide will lead to enormous

01:17:38.520 --> 01:17:43.077
societal and economic changes
which we don't fully understand. Put

01:17:43.160 --> 01:17:46.317
another way, if you don't have a leading
university in your country that's doing

01:17:46.400 --> 01:17:49.757
this kind of stuff,
you're going to be left out.

01:17:49.840 --> 01:17:52.917
And so, at the human level,

01:17:53.000 --> 01:17:55.157
some people will be comfortable in this

01:17:55.240 --> 01:17:58.197
new and very, very curiosity-driven,

01:17:58.280 --> 01:18:03.277
very interesting age, but an awful
lot of people will feel displaced.

01:18:03.360 --> 01:18:06.197
Will they be less
motivated or more motivated?

01:18:06.280 --> 01:18:10.677
Another example. We should
be able, using these tools, to

01:18:10.760 --> 01:18:14.837
build learning systems that teach in
the most efficient way possible each and

01:18:14.920 --> 01:18:18.197
every person.
We know people learn differently,

01:18:18.280 --> 01:18:20.717
but the old rear ends and seats 30

01:18:20.800 --> 01:18:23.397
people in front of a chalkboard is an

01:18:23.480 --> 01:18:27.837
outmoded concept. So, there will be
a completely different way for kids to

01:18:27.920 --> 01:18:31.797
learn and grow up, which has got
to be good because it'll maximize their

01:18:31.880 --> 01:18:35.317
potential. So, you've got a person who's
been maximized their potential, will the

01:18:35.400 --> 01:18:39.917
economic system give them opportunity
to be maximally potentialized? Give you

01:18:40.000 --> 01:18:43.397
another example. I use science examples,
but let's imagine that you really want

01:18:43.480 --> 01:18:46.397
to be a painter or a musician.

01:18:46.480 --> 01:18:48.117
Well, in the future you'll say to the AI

01:18:48.200 --> 01:18:50.757
system, I'm imagining a song about a

01:18:50.840 --> 01:18:55.517
woman and a riverboat in New Orleans
and it's a sad song, but it has willows in

01:18:55.600 --> 01:19:00.917
the background. Or something like that and
the computer will generate a song for you.

01:19:01.000 --> 01:19:03.317
Which won't be as good as what you can do.

01:19:03.400 --> 01:19:07.317
But you'll listen to it and you'll say,
oh, that then stimulates your idea.

01:19:07.400 --> 01:19:10.517
So, I want to invent
a form of cubism that's

01:19:10.600 --> 01:19:14.437
different from Picasso. So, the computer
will then go through a series of

01:19:14.520 --> 01:19:18.597
scenarios of cubism that doesn't look
like Picasso and I'll say, I can do

01:19:18.680 --> 01:19:21.237
better and then I'll be stimulated. So,

01:19:21.320 --> 01:19:23.797
the people who can engage in that, I

01:19:23.880 --> 01:19:26.277
used musicians and painters and

01:19:26.360 --> 01:19:31.197
scientists and thinkers and writers are
going to be the economic winners. But

01:19:31.280 --> 01:19:35.397
what about everyone else? >> Yeah,
it's a really great question. I

01:19:35.480 --> 01:19:38.917
also immediately start thinking,
this just shows you how boring I am, but

01:19:39.000 --> 01:19:42.437
about questions of copyright
and intellectual property. You know, will

01:19:42.520 --> 01:19:44.797
you have some type of blockchain

01:19:44.880 --> 01:19:47.117
identified

01:19:47.200 --> 01:19:53.267
and verified assistant who is kind
of one-to-one correlated with your

01:19:53.350 --> 01:19:55.597
>> [laughter] >> your identity
and social security

01:19:55.680 --> 01:19:58.397
number, therefore you can copyright and

01:19:58.480 --> 01:20:02.877
own anything that AI helps you
to generate or generates itself for that

01:20:02.960 --> 01:20:05.957
matter. It's going to raise
all sorts of questions.

01:20:06.040 --> 01:20:08.717
>> Yeah, I think we should build that
company immediately, you and I. Let's

01:20:08.800 --> 01:20:11.637
found it right now. >> [laughter]

01:20:11.720 --> 01:20:16.437
>> The use of blockchain to do
authentication is going to be central in

01:20:16.520 --> 01:20:20.237
this world. Cuz otherwise,
where do you know the source?

01:20:20.320 --> 01:20:22.837
Who invented this?
How did it actually happen?

01:20:22.920 --> 01:20:24.697
>> Yeah, we'll should talk about it.

01:20:24.780 --> 01:20:26.917
>> [laughter] >> We're
going to need source

01:20:27.000 --> 01:20:30.317
authentication very, very much in this.

01:20:30.400 --> 01:20:34.557
>> I was looking back at my notes
from our last conversation. We

01:20:34.640 --> 01:20:37.157
covered a lot of ground. I recommend
everyone also listen to the last

01:20:37.240 --> 01:20:41.437
conversation because we went so deeply
into your your background, your history,

01:20:41.520 --> 01:20:44.997
the trajectory, different mentors.
It was a great conversation. I really

01:20:45.080 --> 01:20:48.637
enjoyed it. And we spoke
a lot about Bill Joy. And

01:20:48.720 --> 01:20:53.517
when in the notes at least, if I'm
remembering correctly, once he became a

01:20:53.600 --> 01:20:56.717
venture capitalist, he would read
research papers, figure out who the best

01:20:56.800 --> 01:20:59.557
authors or participants were, and then
call them and ask, "What's the most

01:20:59.640 --> 01:21:01.637
interesting thing in your field?"

01:21:01.720 --> 01:21:05.117
And I would love to hear

01:21:05.200 --> 01:21:08.477
any examples of startups, bigger

01:21:08.560 --> 01:21:11.717
companies, academics, particular teams

01:21:11.800 --> 01:21:14.117
at universities, anyone or any groups

01:21:14.200 --> 01:21:16.117
that stand out to you as doing very very

01:21:16.200 --> 01:21:20.157
interesting things in the sphere of AI.

01:21:20.240 --> 01:21:22.317
>> There are two well-known examples in

01:21:22.400 --> 01:21:25.157
AGI. One is called DeepMind, which is in

01:21:25.240 --> 01:21:29.437
the UK. It's owned by Google.
I've been heavily involved with them.

01:21:29.520 --> 01:21:33.917
And another one is OpenAI,
which is the inventor of GPT-3,

01:21:34.000 --> 01:21:36.877
uh which has a big
partnership with Microsoft.

01:21:36.960 --> 01:21:42.197
Those two are probably the largest
focuses in these new areas.

01:21:42.280 --> 01:21:45.917
So-called reinforcement learning,
so-called generative learning.

01:21:46.000 --> 01:21:47.997
And there are a couple of other labs.

01:21:48.080 --> 01:21:49.957
There's a series of university projects.

01:21:50.040 --> 01:21:51.637
I've been funding

01:21:51.720 --> 01:21:54.677
AI applied science in the leading

01:21:54.760 --> 01:21:58.517
universities. And the AI science

01:21:58.600 --> 01:22:00.357
goes something like this.

01:22:00.440 --> 01:22:03.317
The physicists know how something works,

01:22:03.400 --> 01:22:05.717
but our computers can't calculate it.

01:22:05.800 --> 01:22:08.037
It's too complicated a calculation.

01:22:08.120 --> 01:22:10.397
But they can make an approximation.

01:22:10.480 --> 01:22:14.477
So often in science, a AI system is used

01:22:14.560 --> 01:22:19.997
to approximate something good enough
so that we can understand how the system

01:22:20.080 --> 01:22:22.917
works. The most obvious
area where this will

01:22:23.000 --> 01:22:27.277
play out will be in biology.
Another friend of mine said that math is

01:22:27.360 --> 01:22:29.877
to physics what AI will be to biology.

01:22:29.960 --> 01:22:33.317
You needed math to understand how
physics worked. You'll need AI to

01:22:33.400 --> 01:22:35.557
understand how biology works cuz it's so

01:22:35.640 --> 01:22:38.437
incredibly complicated. We still don't

01:22:38.520 --> 01:22:42.717
understand how to model a cell.
We still don't understand how the brain is

01:22:42.800 --> 01:22:46.677
organized. There are so many basic
things that you would think as humans we

01:22:46.760 --> 01:22:51.397
would want to know that have not been
calculable for us and that's where all

01:22:51.480 --> 01:22:52.837
the great discoveries will be.

01:22:52.920 --> 01:22:54.917
>> You know, I have
to share something just

01:22:55.000 --> 01:22:58.517
because it's on this topic in a sense. I

01:22:58.600 --> 01:23:02.757
have a set of papers from the late

01:23:02.840 --> 01:23:05.517
Richard Feynman on which he drew the

01:23:05.600 --> 01:23:09.677
Krebs cycle and I would just found that

01:23:09.760 --> 01:23:14.917
crossover so to speak so fascinating
and we don't have to go deeply into a

01:23:15.000 --> 01:23:18.557
Feynman but certainly one of the people
I've paid a lot of attention to.

01:23:18.640 --> 01:23:22.757
Biology's incredibly complex and I
was thinking of the examples you've given

01:23:22.840 --> 01:23:27.197
and thinking of say protein
folding or trying to identify

01:23:27.280 --> 01:23:31.957
receptors and the shapes of receptors or
even structure the and modify the shapes

01:23:32.040 --> 01:23:36.717
of receptors. The process
as you've already said is so

01:23:36.800 --> 01:23:40.077
incredibly labor intensive and we've
tried all sorts of things to try to pick

01:23:40.160 --> 01:23:42.077
up the slack using idle computers in a

01:23:42.160 --> 01:23:43.717
distributed fashion

01:23:43.800 --> 01:23:48.237
but once you have AI as a player on the

01:23:48.320 --> 01:23:52.957
field, I mean that could change things
so fundamentally as you mentioned with

01:23:53.040 --> 01:23:56.677
that one example. I think Halicin
was the example that you mentioned. It's a

01:23:56.760 --> 01:23:59.437
big big deal. I wonder how AI may

01:23:59.520 --> 01:24:04.197
augment natural prospecting
since nature often

01:24:04.280 --> 01:24:07.877
times the molecules that we identify
in nature are just so beyond the wildest

01:24:07.960 --> 01:24:09.717
imaginations of anyone who would start

01:24:09.800 --> 01:24:15.517
from scratch with a say ground-up
synthetic approach. It raises so many

01:24:15.600 --> 01:24:19.237
interesting questions.
>> It's worth noting that there was a

01:24:19.320 --> 01:24:23.557
competition between two groups. One
a group called Rosetta at University of

01:24:23.640 --> 01:24:24.917
Washington,

01:24:25.000 --> 01:24:28.437
and another one in DeepMind to develop

01:24:28.520 --> 01:24:31.197
proteins, protein folding algorithms.

01:24:31.280 --> 01:24:35.717
And this year, both have reported their
results in the open source, and they've

01:24:35.800 --> 01:24:37.997
published how proteins, the most common

01:24:38.080 --> 01:24:41.677
proteins that are part of biology, fold.

01:24:41.760 --> 01:24:45.157
The reason this is important is the way
that they fold determines the way they

01:24:45.240 --> 01:24:47.357
interact with other cells.

01:24:47.440 --> 01:24:49.917
And it is the basis, again, of drug

01:24:50.000 --> 01:24:52.277
discovery, medicine, how our bodies

01:24:52.360 --> 01:24:54.757
work, genetic expression, and so forth.

01:24:54.840 --> 01:24:57.437
These proteins are super important.

01:24:57.520 --> 01:25:01.957
These are the kinds of discoveries that,
if they were done by humans, would merit

01:25:02.040 --> 01:25:03.597
Nobel Prizes.

01:25:03.680 --> 01:25:06.397
And this is happening right now.

01:25:06.480 --> 01:25:09.317
The way they do them is the same as I

01:25:09.400 --> 01:25:12.277
described, which is they generate

01:25:12.360 --> 01:25:17.277
many, many different candidates,
and they evaluate them using AI, and

01:25:17.360 --> 01:25:21.797
then they choose the one that has the best
fit, most optimal outcome. And then

01:25:21.880 --> 01:25:24.317
they release it, and then that technology
[snorts] will be used for the

01:25:24.400 --> 01:25:28.277
next ones, and so forth. >> What
do you hope the impact of this book

01:25:28.360 --> 01:25:31.117
will be? It's so all-encompassing in

01:25:31.200 --> 01:25:35.397
some respects, this topic. What is the

01:25:35.480 --> 01:25:39.997
the hope with this book in terms
of impact, or what people will do, or how

01:25:40.080 --> 01:25:44.797
they will change their assumptions
or beliefs after being

01:25:44.880 --> 01:25:48.997
exposed to it? >> 20 years
ago, when we started the

01:25:49.080 --> 01:25:53.037
internet as you currently know it,
uh the social media world, many of the

01:25:53.120 --> 01:25:54.717
other tools,

01:25:54.800 --> 01:25:57.637
no one debated what the impact on

01:25:57.720 --> 01:26:01.437
society would be. We were way busy

01:26:01.520 --> 01:26:04.477
building these systems to great success

01:26:04.560 --> 01:26:07.237
without understanding the impact.

01:26:07.320 --> 01:26:10.877
Artificial intelligence
is much more powerful

01:26:10.960 --> 01:26:13.397
than any of the technologies up till now.

01:26:13.480 --> 01:26:18.357
Because, as I mentioned,
it's imprecise, it's emergent, it's

01:26:18.440 --> 01:26:21.437
learning. and insightful.

01:26:21.520 --> 01:26:25.557
We need to understand how we're going
to deal with these things ahead of their

01:26:25.640 --> 01:26:29.157
development. Over and over again,
technologists build these technologies

01:26:29.240 --> 01:26:32.757
without understanding how
they'll be used and misused.

01:26:32.840 --> 01:26:37.557
The goal of the book is to lay
out the fundamental questions

01:26:37.640 --> 01:26:41.717
that society has to decide around
these emergent technologies,

01:26:41.800 --> 01:26:45.917
which will happen, and they
will happen over the next 5 to

01:26:46.000 --> 01:26:48.997
10 years. If our book

01:26:49.080 --> 01:26:51.597
turns out to be the index case

01:26:51.680 --> 01:26:53.837
where after reading this book and after

01:26:53.920 --> 01:26:59.157
its publication, people say,
"Holy crap. I've got to get ahead of

01:26:59.240 --> 01:27:03.197
this. I've got to figure out a philosophy

01:27:03.280 --> 01:27:07.717
around this. I have to figure out a way
where humans can coexist with these new

01:27:07.800 --> 01:27:10.117
systems that doesn't drive humans crazy

01:27:10.200 --> 01:27:12.637
and makes the world better not worse."

01:27:12.720 --> 01:27:15.157
That's a great outcome from our book.

01:27:15.240 --> 01:27:19.977
>> The age of AI and uh
what a powerhouse of

01:27:20.060 --> 01:27:23.957
>> [snorts] >> the trio,
the co-authors involved. It's

01:27:24.040 --> 01:27:26.357
tremendously important and I'm thrilled

01:27:26.440 --> 01:27:29.077
that we've had a chance to

01:27:29.160 --> 01:27:34.517
deep dive into so many facets of this
subject that has been of great interest

01:27:34.600 --> 01:27:38.357
to me, but has come along with great
insecurity because I come at it from a

01:27:38.440 --> 01:27:42.797
lay perspective. Before we
go, I must ask, how has it

01:27:42.880 --> 01:27:45.277
been to start your own podcast?

01:27:45.360 --> 01:27:48.797
>> I enjoyed it. It turns out
to be harder than I thought.

01:27:48.880 --> 01:27:50.677
>> [laughter] >> Because I
actually had to prepare and I

01:27:50.760 --> 01:27:54.597
had to get context and so forth, but it
has had an enough of an impact that I

01:27:54.680 --> 01:27:57.077
will continue. >> Reimagine
with Eric Schmidt. You're an

01:27:57.160 --> 01:28:00.757
excellent conversationalist.
I'm continually impressed. I don't need to

01:28:00.840 --> 01:28:05.157
do much. That's the key to good
interviewing is pick your subject.

01:28:05.240 --> 01:28:07.117
>> That's [laughter] right.
It's all about the subject, though.

01:28:07.200 --> 01:28:10.997
>> I think it's all about
the the the It's a dual dual process.

01:28:11.080 --> 01:28:14.277
>> But [laughter] I think it's
worth it's worth thinking about

01:28:14.360 --> 01:28:17.117
in terms of what you and I do.

01:28:17.200 --> 01:28:20.837
You're a fan of Richard Feynman.
When will there be a computer as smart as

01:28:20.920 --> 01:28:26.157
Richard Feynman? And the answer
is a very long time from now.

01:28:26.240 --> 01:28:29.477
And so we always like to focus on the
lawyers who will lose their jobs and the

01:28:29.560 --> 01:28:33.557
politicians who will lose their jobs,
but that's not how it actually works.

01:28:33.640 --> 01:28:35.917
What really happens is AI is going to be

01:28:36.000 --> 01:28:39.517
used to eliminate repetitive jobs,

01:28:39.600 --> 01:28:43.597
jobs which are boring and so forth.
I mentioned vision.

01:28:43.680 --> 01:28:47.357
Most of the military's
activities are watching things.

01:28:47.440 --> 01:28:49.797
I'd much rather have
the computer watching,

01:28:49.880 --> 01:28:53.077
and then when there's an exception, say,
"Hey, something happened." and alert a

01:28:53.160 --> 01:28:57.277
human to take a look at. It's a better
use of both the human and the computer.

01:28:57.360 --> 01:29:00.877
And so the good news
is that what you do and

01:29:00.960 --> 01:29:05.437
what I do is not going to be
replaced by computers soon.

01:29:05.520 --> 01:29:08.677
>> [laughter] >> I'll take it.
I will absolutely take it

01:29:08.760 --> 01:29:13.157
and I encourage people to read this book.

01:29:13.240 --> 01:29:17.797
The Age of AI, pick it up. You can find
Eric on Twitter @ericschmidt and then

01:29:17.880 --> 01:29:21.597
certainly @ericschmidt.com. We'll
link to everything including the book

01:29:21.680 --> 01:29:25.277
and all resources, companies,
technologies mentioned in the show notes

01:29:25.360 --> 01:29:28.480
at tim.blog/podcast.

01:29:28.720 --> 01:29:33.637
Eric, thank you so much. This has
been extremely, extremely enjoyable and

01:29:33.720 --> 01:29:36.517
educational for me. So,
I appreciate you taking the time.

01:29:36.600 --> 01:29:40.807
>> Thank you, Tim. >> Hey guys,
this is Tim again. Just one

01:29:40.890 --> 01:29:45.477
[music] more thing before you take off
and that is Five Bullet Friday. Would

01:29:45.560 --> 01:29:49.277
you enjoy getting a short email from me
every Friday that provides a little fun

01:29:49.360 --> 01:29:53.117
before the weekend? Between 1 and 1/2
and 2 million people subscribe to my

01:29:53.200 --> 01:29:57.157
free newsletter, my super short
newsletter called Five Bullet Friday.

01:29:57.240 --> 01:29:59.147
Easy to sign up, easy to cancel.

01:29:59.230 --> 01:30:02.357
>> [music] >> It is basically
a half page that I send

01:30:02.440 --> 01:30:06.277
out every Friday to share the coolest
things I've found or discovered or have

01:30:06.360 --> 01:30:10.117
started exploring over that week. It's
kind of like my diary of cool things. It

01:30:10.200 --> 01:30:14.517
often includes articles I'm reading,
books I'm reading, albums perhaps,

01:30:14.600 --> 01:30:18.797
gadgets, gizmos, all sorts of tech
tricks and so on that get sent to me

01:30:18.880 --> 01:30:23.477
[music] by my friends, including a lot
of podcast guests. And these strange

01:30:23.560 --> 01:30:27.837
esoteric things end up [music] in my
field, and then I test them, and then I

01:30:27.920 --> 01:30:32.597
share them with you. So, if that sounds
fun, again, it's very short, a little

01:30:32.680 --> 01:30:36.037
tiny bite [music] of goodness before you
head off for the weekend, something to

01:30:36.120 --> 01:30:40.077
think about. If you'd like to try it out,
just go to tim.blog/friday. [music]

01:30:40.160 --> 01:30:43.520
Type that into your browser,
tim.blog/friday.

01:30:44.120 --> 01:30:48.437
Drop in your email, and you'll get
the very next one. Thanks for listening.

01:30:48.520 --> 01:30:52.217
This episode is brought to you by
ButcherBox. ButcherBox makes it easy for

01:30:52.300 --> 01:30:55.917
[music] you to get high-quality,
humanely raised meat that you can trust.

01:30:56.000 --> 01:31:00.597
They deliver delicious, 100% grass-fed,
grass-finished [music] beef, free-range

01:31:00.680 --> 01:31:05.957
organic chicken, heritage breed pork, and
wild-caught seafood directly to your door.

01:31:06.040 --> 01:31:07.917
>> [music] >> For me,
in the past few weeks, I've

01:31:08.000 --> 01:31:11.837
cooked a ton of their salmon, as well
as two delicious barbecue rib racks in the

01:31:11.920 --> 01:31:14.757
oven. Super simple. They were the most
delicious [music] pork ribs I've ever

01:31:14.840 --> 01:31:19.437
prepared. My freezer is full
of ButcherBox. When you become a member,

01:31:19.520 --> 01:31:22.637
you're joining a community focused
on doing what's better for all. That means

01:31:22.720 --> 01:31:26.037
caring about the lives of animals,
livelihoods of farmers, treating our

01:31:26.120 --> 01:31:29.517
planet with respect, and enjoying
better meals [music] together. ButcherBox

01:31:29.600 --> 01:31:33.117
partners with folks, small farmers
included, who believe in going above and

01:31:33.200 --> 01:31:36.157
beyond when it comes to caring
for animals, the environment, and

01:31:36.240 --> 01:31:39.957
sustainability. And none of their meat
is ever given antibiotics or added

01:31:40.040 --> 01:31:43.757
hormones. So, how does it work? It's
pretty simple. You choose your box and

01:31:43.840 --> 01:31:48.717
your delivery frequency. They offer five
boxes, four curated box options, as well

01:31:48.800 --> 01:31:52.917
as the popular custom box, so you get
exactly what you and or your family

01:31:53.000 --> 01:31:56.277
love. Box options and delivery
frequencies can be customized [music]

01:31:56.360 --> 01:32:00.797
to fit your needs. You can cancel at any
time with no penalty. ButcherBox ships

01:32:00.880 --> 01:32:05.557
your order frozen for freshness and packed
in an eco-friendly 100% recyclable box.

01:32:05.640 --> 01:32:09.477
>> [music] >> It's easy,
it's fast, it's convenient. I

01:32:09.560 --> 01:32:12.357
really, really enjoy it. And best of all,
looking at the average cost, it

01:32:12.440 --> 01:32:16.517
works out to be less than $6 [music]
per meal. Skip the lines for your

01:32:16.600 --> 01:32:18.037
Thanksgiving turkey this holiday.

01:32:18.120 --> 01:32:21.597
Butcher Box is proud to give new
members a free turkey. Just go to

01:32:21.680 --> 01:32:23.757
butcherbox.com/tim

01:32:23.840 --> 01:32:27.837
to sign up. That's butcherbox.com/tim

01:32:27.920 --> 01:32:32.277
to receive a free 10 to 14 lb
turkey [music] in your first box.

01:32:32.360 --> 01:32:34.237
This episode is brought to you by Pique

01:32:34.320 --> 01:32:37.797
Tea. That's p i q u e. I have had so

01:32:37.880 --> 01:32:40.237
much tea in my life. I've been to China.

01:32:40.320 --> 01:32:44.877
I've lived in China and Japan. I've done
tea tours. [music] I drink a lot of tea.

01:32:44.960 --> 01:32:48.197
And 10 years plus of physical
experimentation and

01:32:48.280 --> 01:32:52.557
tracking has shown me many things. Chief
among them, that gut health is critical

01:32:52.640 --> 01:32:55.077
to just about everything. [music]
And you'll see where tea is going to tie

01:32:55.160 --> 01:32:58.317
into this. It affects immune function,
weight management, mental performance,

01:32:58.400 --> 01:33:02.237
emotional health, you name it.
I've been drinking fermented puerh tea

01:33:02.320 --> 01:33:05.477
specifically [music] pretty
much every day for years now.

01:33:05.560 --> 01:33:09.157
Puerh tea delivers more polyphenols and
probiotics [music] than you can shake a

01:33:09.240 --> 01:33:13.317
stick at. It's like providing
the optimal fertilizer to your microbiome.

01:33:13.400 --> 01:33:17.117
The problem with good puerh [music]
is that it's hard to source. It's hard to

01:33:17.200 --> 01:33:21.157
find real puerh that
hasn't been exposed to

01:33:21.240 --> 01:33:24.637
pesticides and other nasties, which
is super common. That's why Pique's

01:33:24.720 --> 01:33:28.277
fermented puerh tea crystals [music]
have become my daily go-to. It's so

01:33:28.360 --> 01:33:31.877
simple. They have so many benefits
that I'm going to get [music] into. And I

01:33:31.960 --> 01:33:36.197
first learned about them through
my friends Dr. Peter Attia and Kevin Rose.

01:33:36.280 --> 01:33:40.357
Pique crystals are cold extracted
[music] using only wild harvested leaves

01:33:40.440 --> 01:33:44.717
from 250-year-old tea trees. I often
kickstart my mornings with their puerh

01:33:44.800 --> 01:33:48.277
green tea, their puerh [music] black
tea, and I alternate between the two.

01:33:48.360 --> 01:33:52.477
The rich earthy flavor of the black
specifically is amazing. It's very, very

01:33:52.560 --> 01:33:55.517
It's like a a delicious barnyard.

01:33:55.600 --> 01:34:00.157
Very peaty if you like whiskey and stuff
like that. They triple toxin screen all

01:34:00.240 --> 01:34:03.877
of their products for heavy metals,
pesticides, and toxic mold contaminants

01:34:03.960 --> 01:34:06.157
commonly found in tea. There's also zero

01:34:06.240 --> 01:34:09.077
prep or brewing required as the crystals

01:34:09.160 --> 01:34:13.317
dissolve in seconds. So, you can just
drop it into your hot tea or I also make

01:34:13.400 --> 01:34:16.357
iced tea and that saves
a ton of time and hassle.

01:34:16.440 --> 01:34:20.357
So, Pique is offering 15% off their
Purity Teas for the very first time

01:34:20.440 --> 01:34:22.837
exclusive to you, my listeners. This is

01:34:22.920 --> 01:34:24.997
a sweet offer. Simply visit pique

01:34:25.080 --> 01:34:28.877
tea.com/tim. [music] That's p i q u e t

01:34:28.960 --> 01:34:33.477
e a dot com forward slash tim. This
promotion is only available to listeners

01:34:33.560 --> 01:34:37.437
of this podcast. That's pique tea.com/tim.

01:34:37.520 --> 01:34:40.717
The discount is automatically applied
when you use that URL. You also have a

01:34:40.800 --> 01:34:44.437
30-day satisfaction guarantee, so
your purchase is risk-free. One more time,

01:34:44.520 --> 01:34:46.477
check it out. Pique tea.com/tim. That's

01:34:46.560 --> 01:34:52.120
p i q u e t e a dot com forward slash tim.