00:00:00
are you uncomfortable talking about this
00:00:02
yeah I mean it's pretty wild right
00:00:04
Mustafa suan the billionaire founder of
00:00:07
Google's AI technology he's played a key
00:00:10
role in the development of AI from its
00:00:12
first critical steps in 2020 I moved to
00:00:15
work on Google's chat box it was the
00:00:17
ultimate technology we can use them to
00:00:19
turbocharge our knowledge unlike
00:00:21
anything else why didn't they release it
00:00:23
we were nervous we were nervous every
00:00:26
organization is going to race to get
00:00:29
their hands on intelligence and that's
00:00:31
going to be incredibly destructive this
00:00:33
technology can be used to identify
00:00:35
cancerous tumors as it can to identify a
00:00:38
Target on the battlefield a tiny group
00:00:41
of people who wish to cause harm are
00:00:43
going to have access to tools that can
00:00:45
instantly destabilize our world that's
00:00:47
the challenge how to stop something that
00:00:50
can cause harm or potentially kill
00:00:52
that's where we need containment do you
00:00:54
think that it is containable it has to
00:00:56
be possible why it must be possible why
00:00:58
must it be because otherwise it contains
00:01:00
us yet you chose to build a company in
00:01:03
this space why did you do that because I
00:01:06
want to design an AI that on your side I
00:01:09
honestly think that if we succeed
00:01:12
everything is a lot cheaper it's going
00:01:13
to power New forms of transportation
00:01:15
reduce the cost of healthcare but what
00:01:18
if we fail the really painful answer to
00:01:20
that question is
00:01:23
that do you ever get sad about
00:01:25
it yeah it's intense
00:01:30
I think this is fascinating I looked at
00:01:33
the back end of our YouTube channel and
00:01:35
it says that since this channel started
00:01:37
69.9% of you that watch it frequently
00:01:40
haven't yet hit the Subscribe button so
00:01:43
I have a favor to ask you if you've ever
00:01:45
watched this Channel and enjoyed the
00:01:46
content if you're enjoying this episode
00:01:47
right now please could I ask a small
00:01:49
favor please hit the Subscribe button
00:01:50
helps this channel more than I can
00:01:51
explain and I promise if you do that to
00:01:54
return the favor we will make the show
00:01:57
better and better and better and better
00:01:59
and better that's the promise I'm
00:02:00
willing to make you if you hit the
00:02:01
Subscribe button do we have a
00:02:03
[Music]
00:02:09
deal everything that's going on with
00:02:11
artificial intelligence now and um this
00:02:14
new wave and all these terms like AGI
00:02:16
and saw another term in your your book
00:02:18
called ACI first time I'd heard that
00:02:20
term how do you feel about it
00:02:22
emotionally if you had to incapsulate
00:02:24
how you feel emotionally about what's
00:02:25
going on in this moment how would you do
00:02:28
what words would you use I would say
00:02:30
say in the past it would have been
00:02:35
petrified and I think that over
00:02:39
time as you really think through the
00:02:42
consequences and the pros and cons and
00:02:44
the trajectory that we're on you adapt
00:02:49
and you understand that actually there
00:02:52
is
00:02:53
something incredibly inevitable about
00:02:55
this trajectory and that we have to wrap
00:02:59
our arms around it and guide it and
00:03:01
control it as a collective species as a
00:03:04
as humanity and I think the more you
00:03:08
realize how much influence we
00:03:11
collectively can have over this outcome
00:03:14
the more empowering it is because on the
00:03:17
face of it this is really going to be
00:03:19
the tool that helps us tackle all the
00:03:22
challenges that we're facing as a
00:03:24
species right we need to fix water
00:03:28
desalination we need to grow food 100x
00:03:31
cheaper than we currently do we need
00:03:33
renewable energy to be you know
00:03:35
ubiquitous and everywhere in our lives
00:03:38
we need to adapt to climate change
00:03:40
everywhere you look in the next 50 years
00:03:43
we have to do more with less and there
00:03:46
are very very
00:03:48
few proposals let alone practical
00:03:52
solutions for how we get there training
00:03:55
machines to help us as AIDS scientific
00:03:59
research Partners inventors creators is
00:04:03
absolutely essential and so the upside
00:04:06
is phenomenal it's enormous but AI isn't
00:04:10
just a thing it's not an inevitable
00:04:13
whole its form isn't inevitable right
00:04:18
its form the exact way that it manifests
00:04:21
and appears in our everyday lives and
00:04:23
the way that it's governed and who it's
00:04:24
owned by and how it's trained that is a
00:04:28
question that is up to us collectively
00:04:31
as a species to figure out over the next
00:04:33
decade because if we don't Embrace that
00:04:36
challenge then it happens to us and
00:04:39
that's really what I'm I have been
00:04:42
wrestling with for 15 years of my career
00:04:44
is how to intervene in a way that this
00:04:48
really does benefit everybody and those
00:04:51
benefits far far outweigh the potential
00:04:54
risks at what stage were you
00:04:57
petrified so I founded Deep Mind in
00:05:02
2010 and you know over the course of the
00:05:06
first few years our progress was fairly
00:05:09
modest but quite quickly in sort of 2013
00:05:14
as the Deep learning Revolution began to
00:05:16
take off I could see
00:05:19
glimmers of very early versions of AIS
00:05:22
learning to do really clever things so
00:05:25
for example one of our big initial
00:05:28
achievements was to Teach an AI to play
00:05:31
the Atari games so remember Space
00:05:34
Invaders and and pong where you batter a
00:05:36
ball from left to right and we trained
00:05:39
this initial AI to purely look at the
00:05:42
raw pixels screen by screen flickering
00:05:45
or moving in front of the AI and then
00:05:48
control the actions up down left right
00:05:51
shoot or not and it got so good at
00:05:55
learning to play this simple game simply
00:05:57
through attaching a value between the
00:06:00
reward like it was it was getting score
00:06:03
and taking an action that it learned
00:06:06
some really clever strategies uh to play
00:06:08
the game really well that Us games
00:06:12
players and humans hadn't really even
00:06:14
noticed at least people in the office
00:06:16
hadn't noticed it some professionals did
00:06:19
um and that was amazing to me because I
00:06:22
was like wow this simple system that
00:06:26
learns through a set of stimuli Plus a
00:06:29
reward to take some actions can actually
00:06:33
discover many strategies clever tricks
00:06:36
to play the game well that us humans
00:06:40
hadn't occurred to us right and that to
00:06:43
me is both thrilling because it presents
00:06:47
the opportunity to invent new knowledge
00:06:51
and Advance our
00:06:53
civilization but of course in the same
00:06:55
measure is also petrifying
00:07:00
was there a particular moment when you
00:07:01
were you were at Deep Mind where you go
00:07:03
where you had that kind of Eureka Moment
00:07:05
Like a day when something happened and
00:07:08
and it caused
00:07:09
that that Epiphany I guess was it yeah
00:07:14
it it it was actually a moment even
00:07:17
before 2013 where I remember standing in
00:07:20
the office and watching a very early
00:07:23
prototype of one of these image
00:07:25
recognition image generation models that
00:07:29
had um was trained to generate new
00:07:32
handwritten black and white digits so
00:07:34
imagine 0er to 1 2 3 4 5 6 789 all in
00:07:39
different style of handwriting on a tiny
00:07:42
grid of like 300 pixels by 300 pixels in
00:07:44
black and white and we were trying to
00:07:47
train the AI to generate a new version
00:07:51
of one of those digits a number seven in
00:07:53
a new handwriting sounds so simplistic
00:07:56
today given the incredible
00:07:58
photorealistic images that are being
00:08:00
generated right um and I just remember
00:08:03
so clearly it it took sort of 10 or 15
00:08:07
seconds and it just resolved it the the
00:08:09
number appeared it went from complete
00:08:12
Black to like slowly gray and then
00:08:14
suddenly these were like white pixels
00:08:16
appeared out of the the black darkness
00:08:18
and it revealed a number seven and that
00:08:21
sounds so simplistic in hindsight but it
00:08:24
was amazing I was like wow the model
00:08:29
kind of understands the representation
00:08:31
of a seven well enough to generate a new
00:08:34
example of a number seven an image of a
00:08:37
number seven you know and you roll
00:08:39
forward 10 years and our predictions
00:08:43
were correct in fact it was quite
00:08:45
predictable in hindsight the trajectory
00:08:48
that we were on more compute plus vast
00:08:51
amounts of data has enabled us within a
00:08:55
decade to go from predicting black and
00:08:57
white digits generating new versions of
00:09:00
those images to now generating
00:09:04
unbelievable
00:09:06
photorealistic not just images but
00:09:08
videos novel videos with a simple
00:09:13
natural language instruction or a prompt
00:09:16
what has surprised you you said you
00:09:18
referred to that as predictable but what
00:09:20
has surprised you about what's happened
00:09:21
over the last decade
00:09:24
so I think what was predictable to me
00:09:27
back then was the generation of images
00:09:30
and of audio um because the structure of
00:09:36
an image is locally contained so pixels
00:09:39
that are near one another create
00:09:41
straight lines and edges and corners and
00:09:43
then eventually they create eyebrows and
00:09:45
noses and eyes and faces and entire
00:09:47
scenes and I could just intuitively in a
00:09:50
very simplistic way I could get my head
00:09:52
around the fact that okay well we're
00:09:54
predicting these number sevens you can
00:09:56
imagine how you then can expand that out
00:09:58
to enre images maybe even to videos
00:10:02
maybe you know to audio too you know
00:10:04
what I said you know a couple seconds
00:10:06
ago is connected in phon space in the
00:10:10
spectrogram but what was much more
00:10:12
surprising to me was that those same
00:10:15
methods for Generation applied in the
00:10:18
space of language you know language
00:10:21
seems like such a different abstract
00:10:24
space of ideas when I say like the cat
00:10:27
sat on the
00:10:29
most people would probably predict Matt
00:10:32
right but it could be table car chair
00:10:36
tree it could be Mountain Cloud I mean
00:10:39
there's a gazillion possible next word
00:10:43
predictions and so the space is so much
00:10:46
larger the ideas are so much more
00:10:48
abstract I just couldn't wrap my
00:10:50
intuition around the idea that we would
00:10:53
be able to create the incredible large
00:10:56
language models that you see today
00:10:59
your chat
00:11:01
gpts chat GPT Google B the Google's Bard
00:11:06
inflection my new company has an AI
00:11:08
called Pi pi. a which stands for
00:11:10
personal intelligence and it's as good
00:11:12
as chat GPT but much more emotional and
00:11:15
empathetic and kind so it's just super
00:11:18
surprising to me that just growing the
00:11:21
size of these large language models as
00:11:24
we have done by 10x every single year
00:11:29
for the last 10 years we've been able to
00:11:31
produce this and that that that's just
00:11:33
an amazingly large number if you just
00:11:36
kind of pause for a moment to Grapple
00:11:39
with the numbers here in 2013 when we
00:11:42
trained the Atari AI that I mentioned to
00:11:44
you at Deep Mind that used two Peta
00:11:49
flops of computation so peta
00:11:52
peta stands for a million billion
00:11:56
calculations a flop is a calculation so
00:12:00
2 million
00:12:01
billion right which is already an insane
00:12:04
number of calculations lost me at two
00:12:06
it's totally crazy yeah just two of
00:12:07
these units that are already really
00:12:09
large and every year since then we've
00:12:13
10x the number of calculations that can
00:12:15
be done such that today the biggest
00:12:19
language model that we train at
00:12:21
inflection uses 10 billion Peta flops so
00:12:26
10 billion million billion calculations
00:12:28
I mean it's just unfathomably large
00:12:31
number and what we've really observed is
00:12:35
that scaling these models by 10x every
00:12:38
single year produces this magical
00:12:42
experience of talking to an AI that
00:12:44
feels like you're talking to a human
00:12:46
that is super knowledgeable and super
00:12:49
smart there's so much that's happened in
00:12:52
public conversation around AI um and
00:12:54
there's so many questions that I have
00:12:55
I've I've been speaking to a few people
00:12:57
about artificial intelligence and
00:12:59
understand it and I'm I think where I am
00:13:02
right now is I feel quite
00:13:05
scared um but when I get scared I don't
00:13:08
get it's not the type of scared that
00:13:09
makes me anxious it's not like an
00:13:11
emotional scared it's a very logical
00:13:12
scared it's my very logical brain hasn't
00:13:15
been able to figure out how the
00:13:17
inevitable outcome that I've arrived at
00:13:20
which is that humans become the less
00:13:22
dominant species on this planet um how
00:13:25
that is to be avoided in any way the
00:13:27
first chapter of your book The Coming
00:13:28
way wave is a is is a is titled
00:13:33
appropriately to how I feel containment
00:13:36
is not possible you you say in that
00:13:38
chapter the widespread emotional
00:13:40
reaction I I was observing is something
00:13:42
I've come to call the pessimism aversion
00:13:44
trap correct what is the pessimism
00:13:47
aversion trap
00:13:50
well so all of us being included feel
00:13:54
what you just described when you first
00:13:57
get to grips with the idea of this new
00:13:59
new coming wave it's scary it's
00:14:01
petrifying it's threatening is it going
00:14:03
to take my job is my daughter or son
00:14:06
going to fall in love with it you know
00:14:08
what does this mean what does it mean to
00:14:10
be human in a world where there's these
00:14:12
other humanlike things that aren't human
00:14:15
how do I make sense of that it's super
00:14:18
scary and a lot of people over the last
00:14:21
few years I think things have changed in
00:14:24
the last six months I have to say but o
00:14:26
over the last few years I would say
00:14:29
the default reaction has been to avoid
00:14:33
the pessimism and the fear right to just
00:14:36
kind of recoil from it and pretend that
00:14:39
it's like either not happening or that
00:14:41
it's all going to work out to be Rosy
00:14:43
it's going to be fine we don't have to
00:14:44
worry about it people often say well
00:14:46
we've always created new jobs we've
00:14:48
never permanently displaced jobs we've
00:14:50
only ever seen new jobs be created
00:14:53
unemployment is at an all-time low right
00:14:55
so there's this default optimism bias
00:14:58
that we have and I think it's less about
00:15:00
a need for optimism and more about a
00:15:02
fear of P pessimism and so that trap
00:15:06
particularly in Elite circles means that
00:15:09
often we aren't having the tough
00:15:11
conversations that we need to have in
00:15:14
order to respond to the coming
00:15:17
wave are you scared in part about having
00:15:20
those tough conversations because of how
00:15:22
it might be received um not so much
00:15:26
anymore so I've spent most of my career
00:15:30
trying to put those tough questions on
00:15:34
the policy table right I've been raising
00:15:38
these questions the ethics of AI safety
00:15:41
and questions of containment for as long
00:15:44
as I can remember with governments and
00:15:45
civil societies and all the rest of it
00:15:46
and so I've become used to talking about
00:15:49
that and you know I
00:15:53
think it's essential that we have the
00:15:56
honest conversation because we can't let
00:15:59
let it happen to us we have to openly
00:16:01
talk about
00:16:02
it is I mean this is a this is a big a
00:16:06
big question
00:16:08
but as you sit here now do you think
00:16:11
that it is containable because
00:16:14
I I I can't see
00:16:16
how I can't see how it can be
00:16:19
contained chapter 3 is the containment
00:16:21
problem or you give give the example of
00:16:24
how Technologies are often invented for
00:16:27
good reasons and for certain use cases
00:16:30
like the hammer you know which is used
00:16:32
you know maybe to build something but it
00:16:33
also can be used to kill people
00:16:36
um and you say in history we haven't
00:16:38
been able to ban a technology ever
00:16:41
really it has always found a way into
00:16:44
society um because of other societies
00:16:47
have an incentive to have it even if we
00:16:49
don't and then we need we need it like
00:16:51
the nuclear bomb because if they have it
00:16:52
then we don't then we're at a
00:16:54
disadvantage so are you optimistic
00:17:01
honestly I don't think an optimism or a
00:17:05
pessimism frame is the right one because
00:17:09
the E both are equally biased in ways
00:17:12
that I think distract us as I say in the
00:17:16
book on the face of it it does look like
00:17:19
containment isn't possible we haven't
00:17:21
contained or permanently banned a
00:17:24
technology of this type in the past
00:17:27
there are some that we have done right
00:17:30
so we banned cfc's for example because
00:17:32
they were producing a hole in the ozone
00:17:34
layer we've banned certain weapons
00:17:37
chemical and biological weapons for
00:17:39
example or blinding lasers Believe It or
00:17:41
Not There are such things as lasers that
00:17:43
will instantly blind you you know so we
00:17:45
have stepped back from the frontier in
00:17:49
some cases but that's largely where
00:17:51
there's either cheaper or you know
00:17:54
equally effective Alternatives that are
00:17:56
quickly adopted in this case
00:17:59
these Technologies are Omni use so the
00:18:02
same core technology can be used to
00:18:06
identify you know cancerous tumors in
00:18:09
chest x-rays as it can to identify a
00:18:12
Target on the battlefield for an aerial
00:18:14
strike so that mixed use or Omni use is
00:18:18
going to drive the proliferation because
00:18:21
there's huge commercial incentives
00:18:22
because it's going to deliver a huge
00:18:23
benefit and do a lot of good and that's
00:18:27
the challenge that we have to figure out
00:18:29
is how to stop something which on the
00:18:31
face of it is so good but at the same
00:18:33
time can be used in really bad ways too
00:18:37
do you think we will I do think we will
00:18:41
so I think that nation states Remain the
00:18:46
backbone of our civilization we have
00:18:50
chosen to concentrate power in a single
00:18:55
Authority the nation state and we pay
00:18:58
our tax is and we've given the nation
00:19:00
state a monopoly over the use of
00:19:03
violence and now the nation state is
00:19:06
going to have
00:19:07
to update itself quickly to be able to
00:19:11
contain this technology because without
00:19:14
that kind of essentially oversight both
00:19:18
of those of us who are making it but
00:19:20
also crucially of the open
00:19:22
source then it will proliferate and it
00:19:25
will spread but regulation is still a
00:19:27
real tool and we we can use it and we
00:19:29
must what does what does the world look
00:19:31
like in um let's say 30 years if that
00:19:35
doesn't happen in your
00:19:38
view people because people the average
00:19:40
person can't really gra grapple their
00:19:42
head around artificial intelligence when
00:19:43
they think of it they think of like
00:19:45
these large Lang large language models
00:19:47
that you can chat to and ask it about
00:19:49
your
00:19:50
homework that's like the average
00:19:52
person's understanding of artificial
00:19:53
intelligence because that's all they've
00:19:54
ever been exposed to of it you have a
00:19:56
different view because of the work
00:19:58
you've spent the last decade doing so to
00:20:01
try and give Dave who's I don't know an
00:20:04
Uber driver in Birmingham an idea who's
00:20:07
listening to this right now what
00:20:08
artificial intelligence intelligence is
00:20:10
and its potential capabilities if you
00:20:13
know there's no there's no containment
00:20:16
what does it what does the world look
00:20:17
like in 30
00:20:18
years so I think it's going to feel
00:20:22
largely like another human so think
00:20:25
about the things that you can do not
00:20:28
again in the physical world but in the
00:20:31
digital world 2050 I'm thinking of I'm
00:20:33
in
00:20:34
2050 2050 we will have robots 2050 we
00:20:39
will definitely have robots I mean more
00:20:40
than that 2050 we will have new
00:20:43
biological beings as well because the
00:20:45
same trajectory that we've been on with
00:20:49
hardware and software is also going to
00:20:52
apply to the platform of biology are you
00:20:56
uncomfortable talking about this yeah I
00:20:58
mean it's pretty wild right don't know
00:21:00
you crossed your arms
00:21:01
and no I always I always look I always I
00:21:04
always use that as as a cue for someone
00:21:05
when when a subject matter is
00:21:07
uncomfortable and it's interesting
00:21:10
because I know you know so much more
00:21:11
than me and about this and I know youve
00:21:14
spent way more hours thinking off into
00:21:16
the future about the consequences of
00:21:17
this I mean you've written a book about
00:21:19
it so P like you spent 10 years at the
00:21:21
very deep mind is one of the the
00:21:23
Pinnacle companies the Pioneers in this
00:21:24
whole Space so you know you know some
00:21:28
stuff and it's funny because when I was
00:21:29
I watched an interview with Elon Musk
00:21:31
and he was asked a question similar to
00:21:32
this I know he speaks in certain certain
00:21:35
tone of voice but he said that he he's
00:21:37
almost he's gotten to the point where he
00:21:38
thinks he's living in suspended
00:21:41
disbelief where he thinks that if he
00:21:43
spent too long thinking about it he
00:21:44
wouldn't understand the purpose of what
00:21:45
he's doing right now and he he says that
00:21:48
it's more dangerous than nuclear weapons
00:21:50
um and that it's too late too late to
00:21:51
stop it there this's one interview
00:21:53
that's chilling and I was filming
00:21:55
Dragons Den the other day and I showed
00:21:56
the dragons the clip I was said look
00:21:58
what El musk said when he was asked
00:21:59
about what his child what advice he
00:22:01
should give to his children in a world
00:22:03
of in an an inevitable world of
00:22:05
artificial intelligence it's the first
00:22:07
time I've seen Elon Musk stop for like
00:22:08
20 seconds and not know what to say
00:22:10
stumble stumble stumble stumble
00:22:13
stumble and then conclude that he's
00:22:15
living in suspended
00:22:17
disbelief yeah I mean I think it's a
00:22:19
great phrase that is the moment we're in
00:22:22
we have to it's what I said to you about
00:22:24
the pessimism verion trap and we have to
00:22:27
confront the probab ility of seriously
00:22:30
dark outcomes and we have to spend time
00:22:33
really thinking about those consequences
00:22:36
because the competitive nature of
00:22:39
companies and of nation states is going
00:22:42
to mean that every organization is going
00:22:45
to race to get their hands on
00:22:49
intelligence intelligence is going to be
00:22:51
a new form of of capital right just as
00:22:54
there was a grab for land or there's a
00:22:56
grab for oil there's a grab for anything
00:22:59
that enables you to do more with less
00:23:01
faster better smarter right and we can
00:23:05
clearly see the predictable trajectory
00:23:08
of the exponential improvements in these
00:23:09
Technologies and so we should expect
00:23:12
that wherever there is power there's now
00:23:15
a new tool to amplify that power
00:23:18
accelerate that power turbocharge it
00:23:21
right and you know in 2050 if you ask me
00:23:25
to look out there I mean of of course it
00:23:28
makes me Grimace that's why I was like
00:23:30
oh my God
00:23:32
it's it really does feel like a new
00:23:35
species and and that has to be brought
00:23:38
under control we cannot allow ourselves
00:23:43
to be dislodged from our position as the
00:23:48
dominant species on this planet we
00:23:50
cannot allow that you mentioned robots
00:23:53
so these are sort of adjacent
00:23:55
technologies that Rising with artificial
00:23:57
intelligence robots you mentioned um
00:24:00
biological new biological
00:24:04
species give me some light on what you
00:24:06
mean by
00:24:07
that well so so far the dream of
00:24:11
Robotics hasn't really come to fruition
00:24:14
right I mean we we still have the most
00:24:17
we have now are sort of drones and
00:24:19
little bit of self-driving
00:24:21
cars but that is broadly on the same
00:24:25
trajectory as these other Technologies
00:24:27
and I think that over for the next 30
00:24:29
Years you know we are going to have
00:24:32
humanoid robotics we're going to have um
00:24:35
you know
00:24:37
physical tools within our everyday
00:24:41
system that we can rely on that will be
00:24:44
pretty good that would be pretty good to
00:24:46
do many of the physical tasks and that's
00:24:48
a little bit further out because I think
00:24:49
it you know there's a lot of tough
00:24:51
problems there but it's still coming in
00:24:53
the same way and likewise with Biology
00:24:56
you know we can now see sequence a
00:24:59
genome for a millionth of the cost of
00:25:02
the first genome which took place in
00:25:04
2000 so 20ish years ago the cost has
00:25:08
come down by a million times and we can
00:25:11
now increasingly synthesize that is
00:25:14
create or
00:25:16
manufacture new bits of DNA which
00:25:19
obviously give rise to life in every
00:25:22
possible form and we're starting to
00:25:24
engineer that DNA to either remove
00:25:27
traits
00:25:29
uh or capabilities that we don't like or
00:25:31
indeed to add new things that we want it
00:25:33
to do we want you know fruit to last
00:25:36
longer or we want meat to have higher
00:25:39
protein etc etc synthetic meat to have
00:25:42
higher protein
00:25:44
levels and what's the implications of
00:25:48
that potential
00:25:51
implications I think the the darkest
00:25:54
scenario there is that people will
00:25:56
experiment with pathogens
00:26:00
engineered you know synthetic pathogens
00:26:03
that might end up accidentally or
00:26:05
intentionally being more
00:26:07
transmissible I.E they they can spread
00:26:10
faster um or more lethal I.E you know
00:26:15
they cause more harm or potentially kill
00:26:17
like a pandemic like a pandemic um
00:26:21
and that's where we need containment
00:26:24
right we have to limit access to the
00:26:27
tool TOs and the knowhow to carry out
00:26:31
that kind of experimentation so one
00:26:34
framework of thinking about this with
00:26:36
respect to making containment possible
00:26:39
is that we really are experimenting with
00:26:43
dangerous
00:26:44
materials and Anthrax is not something
00:26:47
that can be bought over the Internet
00:26:50
that can be freely experimented with and
00:26:53
likewise the very best of these tools in
00:26:55
a few years time are going to be capable
00:26:58
of creating you know new synthetic um
00:27:02
pandemic pathogens and so we have to
00:27:05
restrict access to those things that
00:27:06
means restricting access to the compute
00:27:09
it means restricting access to the
00:27:11
software that runs the models to the
00:27:14
cloud environments that provide apis
00:27:16
provide you access to experiment with
00:27:18
those things um and of course on the
00:27:21
biology side it means restricting access
00:27:23
to some of the substances and people
00:27:24
aren't going to like this people are not
00:27:27
going to like that claim because it
00:27:30
means that those who want to do good
00:27:32
with those tools those who want to
00:27:35
create a start up the small guy the
00:27:38
little developer that struggles to
00:27:40
comply with all the regulations they're
00:27:42
going to be pissed off understandably
00:27:44
right but that is the age we're in deal
00:27:48
with it like we have to confront that
00:27:51
reality that means that we have to
00:27:53
approach this with the precautionary
00:27:55
principle right never before in the
00:27:58
invention of a technology or in the
00:27:59
creation of a regulation have we
00:28:02
proactively said we need to go slowly we
00:28:05
need to make sure that this first does
00:28:07
no harm the precautionary principle and
00:28:10
that is just an unprecedented moment no
00:28:13
other Technology's done that right
00:28:15
because I think we collectively in the
00:28:18
industry those of us who are closest to
00:28:20
the work can see a place in 5 years or
00:28:23
10 years where it could get out of
00:28:26
control and we have to get on top of it
00:28:27
now now and it's better to forgo like
00:28:30
that is give up some of those potential
00:28:33
upsides or benefits until we can be more
00:28:36
sure that it can be contained that it
00:28:39
can be controlled that it always serves
00:28:41
our Collective interests and I I think
00:28:44
about that so I think about what you've
00:28:45
just said there about being able to
00:28:47
create these pathogens these diseases
00:28:49
and viruses Etc that you know could
00:28:51
become weapons or whatever else but with
00:28:53
artificial intelligence and the power of
00:28:55
that intelligence with these um
00:28:59
pathogens you could theoretically ask
00:29:02
one of these systems to create a virus
00:29:06
that a very deadly
00:29:09
virus um you could ask the artificial
00:29:11
intelligence to create a very deadly
00:29:12
virus that has certain
00:29:15
properties um maybe even that mutates
00:29:18
over time in a certain way so it only
00:29:19
kills a certain amount of people kind of
00:29:21
like a nuclear bomb of of viruses that
00:29:23
you could just pop hit an enemy with now
00:29:26
if I'm if I hear that and I go okay
00:29:27
that's powerful I would like one of
00:29:30
those you know there might be an
00:29:31
adversary out there that goes I would
00:29:32
like one of those just in case America
00:29:34
get out of hand and America's thinking
00:29:36
you know I want one of those in case
00:29:37
Russia gets out of hand and so okay you
00:29:40
might take a precautionary approach in
00:29:43
the United States but that's only going
00:29:44
to put you on the back foot when China
00:29:47
or Russia or one of your adversaries
00:29:50
accelerates forward in that in that path
00:29:51
and this was the same with the the
00:29:52
nuclear bomb and you know you nailed it
00:29:57
I mean that is is the race condition we
00:30:00
refer to that as the race condition the
00:30:03
idea that if I don't do it the other
00:30:07
party is going to do it and therefore I
00:30:10
must do it but the problem with that is
00:30:12
that it creates a self-fulfilling
00:30:14
prophecy so the default there is that we
00:30:16
all end up doing it and that can't be
00:30:19
right because there is a opportunity for
00:30:23
massive cooperation here there's a
00:30:26
shared that is between us and China and
00:30:29
every other quote unquote them or they
00:30:31
or enemy that we want to create we've
00:30:35
all got a shared interest in advancing
00:30:39
the collective health and well-being of
00:30:42
humans and Humanity how well have we
00:30:44
done at promoting shared interest well
00:30:47
in the development of Technologies over
00:30:48
the years even at like a corporate level
00:30:51
even you
00:30:52
know you know the nuclear
00:30:54
nonproliferation treaty has been
00:30:55
reasonably successful there's only nine
00:30:58
nuclear states in the world today we've
00:31:00
stopped many like three countries
00:31:03
actually gave up nuclear weapons because
00:31:05
we incentivize them with sanctions and
00:31:07
threats and economic rewards um small
00:31:11
groups have tried to get access to
00:31:12
nuclear weapons and so far have largely
00:31:14
failed it's expensive though right and
00:31:16
hard to like uranium as a as a chemical
00:31:19
to keep it stable and to to buy it and
00:31:21
to house it I mean I couldn't just put
00:31:22
it in the shed you certainly couldn't
00:31:24
put it in a shed you can't download
00:31:26
uranium 235 off the Internet it's not
00:31:28
available open source that is totally
00:31:30
true so it's got different
00:31:32
characteristics for sure but a kid in
00:31:34
Russia could you know in his bedroom
00:31:36
could download something onto his
00:31:38
computer that's incredibly harmful in
00:31:41
the artificial intelligence Department
00:31:44
right I think that that will be possible
00:31:47
at some point in the next five years
00:31:49
it's true because there's a weird Trend
00:31:51
that's going on here on the one hand
00:31:55
You've Got The Cutting Edge AI models
00:31:58
that are built by Google and open Ai and
00:32:01
my company inflection and they cost
00:32:04
hundreds of millions of dollars and
00:32:05
there's only a few of them but on the
00:32:07
other hand the what was cutting edge a
00:32:10
few years ago is now open source today
00:32:14
so gpt3 which came out in the summer of
00:32:18
2020 is now reproduced as an open-
00:32:22
Source model so the code and the weights
00:32:24
of the model the design of the model and
00:32:26
the actual implementation code is
00:32:29
completely freely available on the web
00:32:32
and it's tiny it's like 60 times or 60
00:32:36
70 times smaller than the original model
00:32:39
which means that it's cheaper to use and
00:32:41
cheaper to run and that's as you know
00:32:44
we've said earlier like that's the
00:32:45
natural trajectory of technologies that
00:32:48
become useful they get more efficient
00:32:50
they get cheaper and they spread further
00:32:52
and so that's the containment challenge
00:32:55
that's really the essence of what I'm
00:32:57
sort of trying to raise in my book is to
00:33:00
frame the challenge of the next 30 to 50
00:33:03
years as around containment um and
00:33:06
around confronting
00:33:08
proliferation do you believe because
00:33:11
we're both going to be alive unless this
00:33:12
you know there some robot kills us but
00:33:14
we're both going to be alive in 30 years
00:33:16
time I hope so maybe the podcast will
00:33:19
still be going unless AI is now taking
00:33:21
my
00:33:23
job it's very possible so I'm I'm going
00:33:26
to sit you here and you know when you're
00:33:28
you you'll be what 60 68 years old I'll
00:33:30
be
00:33:33
60 um and I'll
00:33:37
say at that point when we have that
00:33:39
conversation do you think we would have
00:33:41
been successful in containment at a
00:33:43
global level I think we have to be I
00:33:47
can't even think that we're not why
00:33:59
because I'm fundamentally a humanist and
00:34:02
I think that we have to make a choice to
00:34:05
put our species first
00:34:10
and I think that that's what we have to
00:34:12
be defending for the next 50 years
00:34:15
that's what we have to defend because
00:34:17
look it it's it's certainly possible
00:34:21
that we invent these
00:34:23
agis in such a way that they are always
00:34:26
going to be
00:34:29
provably um
00:34:31
subservient uh to humans and take
00:34:36
instructions you know from their human
00:34:38
controller every single time but enough
00:34:41
of us think that we can't be sure about
00:34:44
that that I don't think we should take
00:34:47
the gamble basically
00:34:51
so that's why I think that we should
00:34:53
focus on containment and non
00:34:54
proliferation because some people if
00:34:56
they do have access of the technology
00:34:58
will want to take those risks and they
00:35:00
will just want to see like what's on the
00:35:03
other side of the door you know and they
00:35:05
might end up opening Pandora's Box and
00:35:07
that's a decision that affects all of us
00:35:10
and that's the challenge of the
00:35:11
networked age you know we live in this
00:35:13
globalized world and we use these words
00:35:15
like globalization and we you sort of
00:35:17
forget what globalization means this is
00:35:19
what globalization is this is what a
00:35:21
networked world is it means that someone
00:35:24
taking one small action can suddenly
00:35:27
spread
00:35:28
everywhere instantly regardless of their
00:35:31
intentions when they took the action it
00:35:33
maybe you know unintentional like you
00:35:35
say may be that they're never they
00:35:37
weren't ever meaning to do
00:35:41
harm well I think I asked you when I
00:35:44
said it you 30 years time you said that
00:35:45
there will be like human level
00:35:48
intelligence you'll be interacting with
00:35:50
you know this new species but the
00:35:52
species for me to think the the species
00:35:54
will want to interact with me is feels
00:35:57
like wishful thinking because what will
00:35:59
I be to them you know like I've got a
00:36:02
French Bulldog Pablo and I can't imagine
00:36:05
our IQ is that far apart like like you
00:36:08
know in relative terms my the IQ between
00:36:11
me and my dog Pablo I can't imagine
00:36:12
that's that far apart even when I think
00:36:14
about is it like the orangutang where we
00:36:16
only have like 1% difference in DNA or
00:36:18
something crazy and yet they throw their
00:36:20
poop around and I'm sat here
00:36:22
broadcasting around the world there's
00:36:24
quite a difference in that 1% you know
00:36:27
and then I think about this new species
00:36:29
where as you write in your book in
00:36:31
chapter
00:36:33
4 there seems to be no upper limit to
00:36:35
ai's potential
00:36:37
intelligence why would such an
00:36:39
intelligence want to interact with me
00:36:42
well it depends how you design it
00:36:45
so I think that our goal one of the
00:36:48
challenges of containment is to design
00:36:52
AIS that we want to interact with that
00:36:55
want to interact with us right if you
00:36:58
set an objective function for an AI a
00:37:00
goal for an AI by its design which you
00:37:04
know inherently disregards or
00:37:07
disrespects you as a human and your
00:37:09
goals then it's going to wander off and
00:37:10
do a lot of strange things what if it
00:37:12
has kids and the kids are you know what
00:37:15
I mean what if it replicates in a way
00:37:17
where because because I've I've heard
00:37:19
this this conversation around like it
00:37:20
depends how we design it but you
00:37:23
know I think about
00:37:28
it's kind of like if I have a kid and
00:37:30
the kid grows up to be a thousand times
00:37:32
more intelligent than me to think that I
00:37:35
could have any influence on it on it
00:37:37
when it's a thinking
00:37:39
sentient
00:37:40
developing species again feels like I'm
00:37:45
overestimating my version of
00:37:46
intelligence and importance and
00:37:48
significance in the face of something
00:37:50
that is incomprehensibly like a even a
00:37:52
hundred times more intelligent than me
00:37:54
and the speed of its computation is a
00:37:56
thousand times what my
00:37:58
the meat in my skull can do yeah like
00:38:01
how how how is it gonna how how do I
00:38:03
know it's going to respect me or care
00:38:04
about me or understand you know that I
00:38:06
me you
00:38:08
know I think that comes back down to the
00:38:10
containment challenge I think that if we
00:38:12
can't be confident that it's going to
00:38:15
respect you and understand you and work
00:38:19
for you and us as a species overall then
00:38:24
that's where we have to adopt the
00:38:26
precautionary principle I don't think we
00:38:28
should be taking those kinds of risks in
00:38:31
experimentation and design and now I'm
00:38:33
not saying it's possible to design an AI
00:38:38
that doesn't have those self-improvement
00:38:41
capabilities in the limit in like 30 or
00:38:43
50 years I think it you know that's kind
00:38:45
of what I was saying is like it seems
00:38:47
likely that if you have one like that
00:38:50
it's going to take advantage of infinite
00:38:52
amounts of data and infinite amounts of
00:38:54
computation and it's going to kind of
00:38:55
outstrip our ability to act and so I
00:38:59
think we have to step back from that
00:39:01
precipice that's what the containment
00:39:04
problem is is that it's it's actually
00:39:07
saying no sometimes it's saying no and
00:39:11
that's a different sort of muscle that
00:39:14
we've never really exercised as a
00:39:16
civilization and and that's obviously
00:39:18
why containment appears not to be
00:39:20
possible because we've never done it
00:39:22
before we've never done it before and
00:39:24
every inch of our you know Comm and
00:39:28
politics and our war and all of our
00:39:30
instincts are just like Clash compete
00:39:32
Clash compete prophit profit grow beat
00:39:36
exactly dominate you know fear them be
00:39:39
paranoid like now all this nonsense
00:39:41
about like China being this new evil
00:39:44
like it how does that slip into our
00:39:46
culture how are we suddenly all shifted
00:39:49
from thinking it's the the the Muslim
00:39:50
terrorists about to blow us all up to
00:39:53
now it's the Chinese who are about to
00:39:55
you know blow up Kansas it's just like
00:39:57
what are we talking about that like we
00:39:59
really have to pair back the paranoia
00:40:02
and the fear and the othering um because
00:40:06
those are the incentive dynamics that
00:40:08
are going to drive us to you know cause
00:40:11
self harm to humanity thinking the worst
00:40:14
in each other the the there's couple of
00:40:16
key moments when in my understanding of
00:40:18
artificial intelligence that have been
00:40:19
kind of Paradigm Paradigm shifts for me
00:40:21
because I think like many people I
00:40:23
thought of artificial intelligence
00:40:25
as you know like a like a child I was
00:40:28
Raising and I would program I would code
00:40:31
it to do certain things so I would code
00:40:32
it to play chess and I would tell it the
00:40:36
moves that are conducive with being
00:40:39
successful in chess and then I remember
00:40:41
watching that like Alpha go documentary
00:40:43
right which I think was deep deep mind
00:40:44
wasn't it that was us yeah you guys so
00:40:45
you programmed this this um artificial
00:40:47
intelligence to play the game go which
00:40:49
is kind of like just think of it kind of
00:40:50
like a chess or a black am or whatever
00:40:52
and it eventually just beats the best
00:40:53
player in the world of all time and it
00:40:55
and the way it learned how to beat the
00:40:57
best player in the world of all time the
00:40:58
world champion who was by the way
00:41:00
depressed when he got beat um was just
00:41:03
by playing itself right and then there's
00:41:06
this moment I think in is it game four
00:41:08
or something where right it does this
00:41:10
move that no one could have predicted a
00:41:12
move that seemingly makes absolutely no
00:41:15
sense
00:41:16
right in those moments where no one
00:41:20
trained it to do that and it did
00:41:21
something unexpected Beyond where humans
00:41:24
are trying to figure it out in hindsight
00:41:26
this is where I go how do you how do you
00:41:27
train it if it's doing things we didn't
00:41:30
anticipate right like how do you control
00:41:32
it when it's doing things that humans
00:41:34
couldn't anticipate it doing where we're
00:41:35
looking at that move it's called like
00:41:37
move 37 or something correct yeah is it
00:41:39
move 37 it is look at my intelligence F
00:41:41
nice work yeah I'm I'm going to survive
00:41:43
a bit longer than I thought it's like
00:41:45
move 37 you at least another decade in
00:41:48
you um move 37 does this crazy thing and
00:41:51
you see everybody like lean in and go
00:41:53
why has it done that and it turns out to
00:41:55
be brilliant that humans couldn't
00:41:57
couldn't forecast the commentator
00:41:58
actually thought it was a mistake yeah
00:42:00
he was a pro and he was like this this
00:42:02
definitely a mistake you know it's the
00:42:03
the alpha go lost the game but it was so
00:42:05
far ahead of us that it knew something
00:42:07
we didn't right right that's when that's
00:42:10
when I lost hope in this whole idea of
00:42:11
like oh train it to do what we want like
00:42:13
a dog like sit pour roll over right well
00:42:18
the real challenge is that we actually
00:42:22
want it to do those things like when it
00:42:25
discovers a new strategy
00:42:27
or it invents a new idea or it helps us
00:42:30
find like you know a cure for some
00:42:32
disease like that's why we're building
00:42:34
it right because we're reaching the
00:42:37
limits of what we as you know humans can
00:42:41
invent and solve right especially with
00:42:43
what we're facing of you know in terms
00:42:45
of population growth over the next 30
00:42:47
Years and how climate change is going to
00:42:49
affect that and so on like we really
00:42:52
want these tools to turbocharge us right
00:42:55
and yet like it's that creativity and
00:42:57
that invention which obviously makes us
00:43:00
also
00:43:01
feel well maybe it it is really going to
00:43:05
do things that we don't like for sure
00:43:07
right so
00:43:11
interesting how do you contend with all
00:43:13
of this how do you contend with the the
00:43:16
clearup side and then you must like Elon
00:43:19
must be completely aware of the the
00:43:23
horrifying existential risk at the same
00:43:26
time and and you're building a big
00:43:28
company in this space which I think is
00:43:30
valued at 4 billion now inflection AI
00:43:32
which has got this its own model called
00:43:36
Pi so you're building in this space you
00:43:40
understand the incentives at both a
00:43:41
nation state level and a corporate level
00:43:43
that we're we're going to keep planing
00:43:44
forward even if the US stops there's
00:43:46
going to be some other country that sees
00:43:47
that as a huge Advantage their economy
00:43:49
will swell because they did if this
00:43:51
company stops then this one's going to
00:43:53
get a get a huge advantage and their
00:43:54
shareholders are you know
00:43:58
everyone's investing in AI full steam
00:44:00
ahead but you feel you can see this huge
00:44:03
existential risk is it suspended is that
00:44:04
the pathw suspended
00:44:06
disbelief I mean just to kind of like
00:44:09
just know that it's I feel like I know
00:44:11
that it's going to happen no one's been
00:44:13
able to tell me
00:44:15
otherwise but just don't think too much
00:44:18
about it and you'll be okay I think you
00:44:21
can't give up right I think that in some
00:44:25
ways you're realization exactly what
00:44:28
you've just described like weighing up
00:44:31
two conflicting and horrible truths
00:44:33
about what is likely to happen those
00:44:37
contradictions that is a kind of honesty
00:44:40
and a wisdom I think that we need all
00:44:43
collectively to realize because the only
00:44:47
path through this is to be straight up
00:44:51
and embrace you know the risks and
00:44:54
embrace the default trajectory of all
00:44:56
these competing incentives driving
00:44:58
forward to kind of make this feel like
00:45:01
inevitable and if you put the blinkers
00:45:03
on and you kind of just ignore it or if
00:45:05
you just be super Rosy and it's all
00:45:06
going to be all right and if you say
00:45:07
that we've always figured it out anyway
00:45:09
then you we're not going to get the
00:45:11
energy and the dynamism and engagement
00:45:13
from everybody to try to figure this out
00:45:16
and that's what gives me like reason to
00:45:19
be hopeful because I think that we make
00:45:22
progress by getting everybody paying
00:45:25
attention to this it isn't going to be
00:45:28
about those who are currently the AI
00:45:30
scientists or those who are the
00:45:32
technologists you know like me or the
00:45:34
Venture capitalists or just the
00:45:36
politicians like all of those people no
00:45:38
one's got answers so that's what we have
00:45:40
to confront there are no obvious answers
00:45:44
to this profound question and I've
00:45:48
basically written the book to say prove
00:45:51
that I'm wrong you know containment must
00:45:54
be
00:45:55
possible and I it must be it must be
00:45:59
possible it has to be possible it has to
00:46:01
be you want it to be I I desperately
00:46:04
want it to be yeah why must it be
00:46:08
because otherwise I think you're in the
00:46:10
camp of believing that this is the
00:46:13
inevitable evolution of humans the
00:46:17
transhuman kind of view you know some
00:46:21
people would argue like what is okay
00:46:23
let's part let's let's stretch the
00:46:24
timelines out okay so let's not talk
00:46:27
about 30 years let's talk about 200
00:46:31
years like what is this going to look
00:46:34
like in
00:46:38
2200 you tell me you're smarter than
00:46:40
me I mean it's mindblowing it's
00:46:43
mind-blowing we'll have quantum
00:46:45
computers by then what's a quantum
00:46:48
computer a quantum computer is a
00:46:51
completely different type of computing
00:46:54
architecture which in simple
00:46:57
terms basically allows you
00:47:00
to those those calculations that I
00:47:03
described at the beginning billions and
00:47:05
billions of flops those billions of
00:47:07
flops can be done in a single
00:47:10
computation so everything that you see
00:47:13
in the digital world today relies on
00:47:16
computers processing information and and
00:47:19
the speed of that processing is a
00:47:21
friction it kind of slows things down
00:47:24
right you remember back in the day old
00:47:27
School modems 56k modem the dialup sound
00:47:30
and the image pixel loading like pixel
00:47:33
by pixel that was because the computers
00:47:35
were slow and we're getting to a point
00:47:37
now where the computers are getting
00:47:39
faster and faster and faster and Quantum
00:47:41
Computing is like a whole new leap like
00:47:44
way way way Beyond where we where we
00:47:47
currently are and so by analogy how
00:47:50
would I understand that so like if my
00:47:52
I've got my dialup modem over here and
00:47:55
then Quantum Computing over here
00:47:57
right what's the how do I what's the
00:48:00
difference well I don't know what it's
00:48:02
really difficult to exp a billion times
00:48:04
faster oh it's it's it's like it's like
00:48:06
billions of billions times faster it's
00:48:08
it's it's much more than that I mean one
00:48:10
way of think about it is
00:48:12
like a floppy disc which I guess most
00:48:16
people remember 1.4 megabytes a physical
00:48:19
thing back in the day in
00:48:22
1960 or so that was basically an entire
00:48:27
pallets worth of computer that was moved
00:48:30
around by a forklift truck right which
00:48:33
is insane today you know you have
00:48:37
billions and billions of times that
00:48:39
floppy disc in your smartphone in your
00:48:43
pocket tomorrow you're going to have
00:48:47
billions and billions of smartphones in
00:48:51
minuscule wearable devices there'll be
00:48:54
cheap fridge magnets that you know are
00:48:56
constantly on everywhere sensing all the
00:48:59
time monitoring processing analyzing
00:49:02
improving
00:49:03
optimizing you know and they'll be super
00:49:06
cheap so it's super unclear what do you
00:49:10
do with all of that knowledge and
00:49:12
information I mean it's ultimately
00:49:15
knowledge creates value when you know
00:49:18
the relationship between things you can
00:49:20
improve them you know make it more
00:49:22
efficient and so more data is what has
00:49:26
enabled us to build all the value of you
00:49:28
know online in the last 25 years and so
00:49:32
what does that look like in 150 years I
00:49:35
can't really even imagine to be honest
00:49:36
with you it's very hard to say I don't
00:49:39
think everybody is going to be
00:49:41
working why would we yeah what we
00:49:45
wouldn't be working in that kind of
00:49:46
environment I mean look the other
00:49:48
trajectory to add to this is the cost of
00:49:53
energy
00:49:54
production you know AI if it really
00:49:58
helps us
00:50:00
solve battery
00:50:02
storage which is the missing piece I
00:50:05
think to really tackle climate change
00:50:08
then we will be able to Source basically
00:50:11
source and store infinite energy from
00:50:15
the Sun and I think in 20 or so years
00:50:18
time 20 30 years time that is going to
00:50:21
be a cheap and widely available if not
00:50:23
completely freely available resource and
00:50:26
if you think about it everything in
00:50:29
life has the cost of energy built into
00:50:32
its production value and so if you strip
00:50:35
that out everything is likely to get a
00:50:38
lot cheaper we'll be able to desalinate
00:50:41
water we'll be able to grow crops much
00:50:44
much cheaper we'll be able to grow much
00:50:46
higher quality food right it's going to
00:50:48
power New forms of transportation it's
00:50:50
going to reduce the cost of drug
00:50:52
production and Healthcare right so all
00:50:55
of those gains
00:50:57
obviously there'll be a huge commercial
00:50:58
incentive to drive the production of
00:51:00
those gains but the cost of producing
00:51:02
them is going to go through the floor I
00:51:03
think that's one key thing that a lot of
00:51:05
people don't realize that is a reason to
00:51:07
be hugely hopeful and optimistic about
00:51:09
the future everything is going to get
00:51:12
radically cheaper in 30 to 50
00:51:18
years it's a 200 years time we have no
00:51:20
idea what the world looks like it's uh
00:51:22
this goes back to the point about being
00:51:24
is it did you say transhumanist
00:51:26
right what does that
00:51:28
mean
00:51:30
transhumanism I mean it's a group of
00:51:32
people
00:51:34
who basically believe
00:51:37
that you that that humans and our soul
00:51:41
and our being will one day transcend or
00:51:44
move beyond our biological substrate
00:51:49
okay so our physical body our brain our
00:51:51
biology is just an enabler for your
00:51:55
intelligence and who you are as a person
00:51:59
and there's a group of kind of crack
00:52:03
Bots basically I think who think that
00:52:06
we're going to be able to upload
00:52:09
ourselves to a silicon substrate right a
00:52:13
computer that can hold the essence of
00:52:16
what it means to be Stephen so you in
00:52:18
200 in 20 uh in in
00:52:22
2200 will could well still be You by
00:52:25
their reasoning
00:52:27
but you'll live on a server somewhere
00:52:30
why are they wrong I think about all
00:52:31
these adjacent Technologies like
00:52:33
biological um biological advancements
00:52:36
did you call it like biosynthesis or
00:52:38
something was yeah synthetic biology syn
00:52:40
synthetic biology um I think about the
00:52:44
nanotechnology development right think
00:52:46
about Quantum Computing the the progress
00:52:48
in artificial intelligence everything
00:52:50
becoming cheaper and I think why why are
00:52:51
they
00:52:53
wrong it's hard to say precisely
00:52:57
but broadly speaking I haven't seen any
00:53:00
evidence yet that we're able to extract
00:53:03
the essence of a being from a brain
00:53:06
right it's that that that that kind of
00:53:08
dualism that you know there is a mind
00:53:11
and a body and a spirit that is a I I
00:53:16
don't think I don't see much evidence
00:53:18
for that even in Neuroscience um that
00:53:20
actually it's much more one and the same
00:53:22
so I don't think you know you're going
00:53:24
to be able to emulate the entire brain
00:53:26
so their thesis is that well some of
00:53:28
them cryogenically store their brain
00:53:30
after death Jesus so they they have it
00:53:33
they' they they wear these like you know
00:53:35
how you have like an organ donor tag or
00:53:37
whatever so they have a cryogenically
00:53:40
freeze me when I die tag and so they
00:53:43
there's like a special like ambulance
00:53:45
services that will come pick you up
00:53:47
because obviously you need to do it
00:53:48
really quickly the moment you die you
00:53:49
need to get put into a cryogenic freezer
00:53:52
to preserve your you know brain forever
00:53:55
I personally think this is this is is
00:53:56
nuts but you know their belief is that
00:53:58
you'll then be able to reboot that
00:54:01
biological brain and then transfer you
00:54:03
over
00:54:05
um it it doesn't seem plausible to me
00:54:08
when you said at the start of this this
00:54:09
little topic here that you it must be
00:54:12
possible to contain it said it must be
00:54:15
possible um the the reason why I I
00:54:18
struggle with that is because in chapter
00:54:19
7 you say line in your book that AI is
00:54:22
more autonomous than any other
00:54:23
technology in
00:54:24
history for centuries the idea that
00:54:26
technology is is somehow running out of
00:54:29
control a self-directed and
00:54:30
self-propelling force beyond the Realms
00:54:32
of human agency remained a fiction not
00:54:38
anym and this idea of autonomous
00:54:41
technology that
00:54:43
is acting
00:54:47
uninstructed um and is intelligent and
00:54:50
then you say we must be able to contain
00:54:52
it it's kind of like a massive dog like
00:54:54
a big rottweiler yeah
00:54:56
that is you know a thousand times bigger
00:54:59
than me and me looking up at it and
00:55:00
going I'm going to get take you for a
00:55:02
walk y yeah and then it's just looking
00:55:05
down at me and
00:55:07
just stepping over me or stepping on me
00:55:10
well that's actually a good example
00:55:12
because we have actually contained
00:55:15
Rottweilers before we've contained
00:55:17
gorillas and you know tigers and
00:55:20
crocodiles and pandemic pathogens and
00:55:23
nuclear weapons and so you know it's
00:55:26
easy to be you know a hater on what
00:55:29
we've achieved but this is the most
00:55:32
peaceful moment in the history of our
00:55:34
species this is a moment when our
00:55:36
biggest problem is that people eat too
00:55:38
much think about that we've spent our
00:55:42
entire evolutionary
00:55:44
period running around looking for food
00:55:47
and trying to stop you know our enemies
00:55:49
throwing rocks at us and we've had this
00:55:53
incredible period of 500 years
00:55:57
where you know each year things have
00:56:00
broadly well maybe each each Century
00:56:03
let's say there's been a few ups and
00:56:04
downs but things have broadly got better
00:56:07
and we're on a trajectory for you know
00:56:10
lifespans to increase and quality of
00:56:12
life to increase and health and
00:56:14
well-being to improve and I think that's
00:56:17
because in many ways we have succeeded
00:56:20
in containing forces that appear to be
00:56:23
more powerful than ourselves it just
00:56:25
requires unbelievable creativity and
00:56:29
adaptation it requires compromise and it
00:56:33
requires an a new tone right a much more
00:56:37
humble tone to governance and politics
00:56:41
and and how we run our world not this
00:56:43
kind of like hyper aggressive
00:56:45
adversarial paranoia tone that we talked
00:56:47
about previously but one that is like
00:56:50
much more wise than that much more
00:56:53
accepting that we are unleashing this
00:56:55
force that does have that that potential
00:56:56
to be the Rott riler that you described
00:56:59
but that we must contain that as our
00:57:02
number one priority that has to be the
00:57:04
thing that we focus on because otherwise
00:57:06
it contains
00:57:08
us i' I've been thinking a lot recently
00:57:10
about cyber security as well just
00:57:11
broadly on an individual level in a
00:57:14
world where there are these kinds of
00:57:16
tools which seems to be quite close um
00:57:18
large language models brings up this
00:57:21
whole new question about cyber security
00:57:23
and cyber safety and you know in a world
00:57:25
where there's these ability to generate
00:57:28
audio and language and videos that seem
00:57:31
to be real um what can we trust and you
00:57:34
know I was watching a video of a of a of
00:57:37
a young girl whose grandmother was
00:57:39
called up by a voice that was made to
00:57:42
sound like her son saying he'd been in a
00:57:44
car accident and asking for money and
00:57:47
her nearly sending the money or this
00:57:49
whole you know because this really
00:57:50
brings into Focus that we our lives are
00:57:52
build on built on trust trusting the
00:57:54
things we see here in
00:57:57
watch and in in and now we're at feels
00:58:01
like a a a moment where we're no longer
00:58:04
going to be able to trust what we see on
00:58:06
the internet on the
00:58:09
phone what what what advice do you do we
00:58:11
you have for people who were worried
00:58:13
about
00:58:15
this
00:58:16
so skepticism I think is healthy and
00:58:20
necessary and I think that we're going
00:58:22
to need it um even more than than we
00:58:26
ever did right and so if you think about
00:58:29
how we've adapted to the first wave of
00:58:32
this which was spammy email scams um
00:58:35
everybody got them and over
00:58:38
time people learned to identify them and
00:58:42
be skeptical of them and reject them
00:58:44
likewise you know I'm sure many of us
00:58:46
get like text messages I certainly get
00:58:48
loads of text messages trying to fish me
00:58:50
and ask me to meet up or do this that
00:58:52
and the other and we've adapted right
00:58:55
now I think we should all know and
00:58:59
expect that criminals will use these
00:59:02
tools to manipulate us just as you've
00:59:05
described I mean you know the voice is
00:59:08
going to be humanlike the Deep fake is
00:59:11
going to be super convincing and there
00:59:15
are actually ways around those things so
00:59:18
for example the reason why the banks
00:59:21
invented OTP um one-time passwords where
00:59:24
they send you a text message with a
00:59:26
special code um is precisely for this
00:59:29
reason so that you have a 2fa a two
00:59:32
Factor authentication increasingly we
00:59:34
will have a three or four Factor
00:59:37
authentication where you have to
00:59:38
triangulate between multiple separate
00:59:42
independent sources and it won't just be
00:59:45
like call your bank manager and release
00:59:46
the funds right
00:59:49
so this is where we need the creativity
00:59:52
and energy and attention of everybody
00:59:54
because
00:59:56
defense the kind of defensive measures
00:59:59
have to evolve as quickly as the
01:00:01
potential offensive measures the attacks
01:00:04
that are
01:00:05
coming I heard you say this that you
01:00:08
think um some people are for many of
01:00:10
these problems we're going to need to
01:00:12
develop AIS to defend us from the
01:00:15
AIS right we kind of already have that
01:00:18
right so we have automated ways of
01:00:21
detecting spam online these days you
01:00:24
know most of the time there are um
01:00:27
machine Learning Systems which are
01:00:29
trying to identify when your credit card
01:00:31
is used in a fraudulent way that's not a
01:00:34
human sitting there looking at patterns
01:00:35
of spending traffic in real time that's
01:00:38
an AI that is like flagging that
01:00:40
something looks off um likewise with
01:00:43
data centers or security cameras a lot
01:00:47
of those security cameras these days are
01:00:49
you know have tracking algorithms that
01:00:51
look for you know surprising sounds or
01:00:55
like if a if a glass window is is
01:00:57
smashed that'll be detected by an AI
01:01:01
often that is you know listening on the
01:01:03
security camera so you know that's kind
01:01:06
of what I mean by that is that
01:01:07
increasingly those AIS will get more
01:01:09
capable and we'll want to use them for
01:01:11
defensive purposes and that's exactly
01:01:14
what it looks like to have good healthy
01:01:16
well-functioning controlled AIS that
01:01:18
serve us I went on one of these large
01:01:20
language models and and said to me give
01:01:22
I said to the large language model give
01:01:23
me an example where in artificial
01:01:25
intelligence takes over the world or
01:01:26
whatever and just and results in the
01:01:29
destruction of humanity and then tell me
01:01:31
what we'd need to do to prevent
01:01:33
it and it said it gave me this wonderful
01:01:36
example of this AI called Cynthia that
01:01:39
threatens to destroy the world and it
01:01:41
says the way to defend that would be a
01:01:43
different AI which had a different name
01:01:46
and it said that this one would be
01:01:47
acting in human interests and we'd
01:01:49
basically be fighting one AI with
01:01:51
another Ai and of and of course of
01:01:55
course of course that level if Cynthia
01:01:57
started to wreak hav havoc on the world
01:01:59
and take control of the nuclear weapons
01:02:00
and infrastructure and all that we would
01:02:02
need an equally
01:02:04
intelligent weapon to fight
01:02:07
it although one of the interesting
01:02:09
things that we found um over the last
01:02:12
few decades is that it so far tended to
01:02:15
be the AI plus the human that has that
01:02:19
is still dominating that's the case in
01:02:21
chess uh in go and other games um
01:02:26
that in go it's still yeah so there was
01:02:28
a paper that came out a few months ago
01:02:31
two months ago that showed that a human
01:02:33
was actually able to beat The Cutting
01:02:35
Edge go program um even one that was
01:02:38
better than Alpha go with a new strategy
01:02:40
that they had discovered um you know so
01:02:44
obviously it's not just a sort of game
01:02:47
over environment where the AI just
01:02:48
arrives and it gets better like humans
01:02:50
also adapt they get super smart they
01:02:53
like I say get more cynical ask get more
01:02:55
more skeptical ask you know good
01:02:58
questions invent their own things use
01:02:59
their own AIS to adapt and that's the
01:03:02
evolutionary nature of what it means to
01:03:05
have a technology right I mean
01:03:06
everything is a technology like your
01:03:08
pair of glasses made you smarter in a
01:03:12
way like before there were glasses and
01:03:14
people got bad eyesight they weren't
01:03:15
able to read you know suddenly those who
01:03:18
did adopt those Technologies were able
01:03:20
to read for you know longer in their
01:03:22
lives or under low light conditions and
01:03:24
they were able to consume more
01:03:25
information and got smarter and so that
01:03:27
is the trajectory of Technology it's
01:03:29
this iterative interplay between you
01:03:33
know human and machine that makes us
01:03:36
better over time you know the potential
01:03:39
um consequences if if we don't reach a
01:03:42
point of containment yet you chose to
01:03:43
build a company in this space
01:03:47
yeah why why that why did you do that
01:03:51
because I believe that the best way to
01:03:56
uh demonstrate how to build safe and and
01:04:00
contained AI is to actually experiment
01:04:04
with it in practice and I think that if
01:04:08
we are just Skeptics or critics and we
01:04:11
stand back from The Cutting Edge then we
01:04:14
give up that opportunity to shape
01:04:16
outcomes to you know all of those other
01:04:20
actors that we referred to whether it's
01:04:22
like China and the US going at each
01:04:23
other's throats uh you know you know or
01:04:26
other big companies that are purely
01:04:27
pursuing profit at all costs and so it
01:04:31
doesn't solve all the problems of course
01:04:33
it's super hard and again it's full of
01:04:35
contradictions but I honestly think it's
01:04:38
the right way for everybody to proceed
01:04:41
you know if experiment at the front yeah
01:04:43
if you're afraid Russia Putin understand
01:04:47
right what reduces fear is deep
01:04:50
understanding spend time playing with
01:04:52
these models look at their weaknesses
01:04:54
they're not superhuman yet they make
01:04:56
tons of mistakes they're crappy in lots
01:04:58
of ways they're actually not that hard
01:05:00
to make the more you've experimented has
01:05:03
it has that correlated with a reduction
01:05:05
in
01:05:08
fear cheeky question no but that's yes
01:05:12
and no you're totally right yes it has
01:05:14
in the sense that you know the problem
01:05:17
is the more you learn the more you
01:05:19
realize yeah that's what I'm saying I
01:05:22
was fine before I started talking about
01:05:24
Ai and now more I've talked about
01:05:27
it it's true it's true it's it's sort of
01:05:30
pulling on a thread
01:05:32
which it's a crazy spiral um yeah I mean
01:05:37
like I think in the short term It's Made
01:05:39
Me way less afraid because I I don't see
01:05:42
that kind of existential harm that we've
01:05:45
been talking about in the next decade or
01:05:47
two but longer term that's that's where
01:05:49
I struggle to wrap my head around how
01:05:51
things play out in 30
01:05:53
years some people say
01:05:57
government regulation will sorted
01:05:59
out you discussed this in Chapter 13 of
01:06:02
your book where you which is titled
01:06:04
containment must be possible I love how
01:06:07
you didn't say is yeah containment must
01:06:11
be containment must be possible um what
01:06:14
do you say to people that say government
01:06:15
regulation will sorted out I had rishy
01:06:16
sunak did some announcement and he's got
01:06:18
a cobra committee coming together
01:06:20
they'll handle it that's right and the
01:06:23
EU have a huge piece of regulation
01:06:26
called the EU AI act um um you know Joe
01:06:30
President Joe Biden has you know gotten
01:06:32
his own you know set of proposals and um
01:06:36
you know we've been working with with
01:06:38
both you know Rishi sunak and and Biden
01:06:40
and you know trying to contribute and
01:06:42
shape it in the best way that we can
01:06:44
look it isn't going to happen without
01:06:47
regulation so regulation is essential is
01:06:50
critical um again going back to the
01:06:53
precautionary principle but at the same
01:06:56
time regulation isn't enough you know I
01:06:59
often hear people say well we'll just
01:07:01
regulate it we'll just stop we'll just
01:07:04
stop we'll just stop we'll slow down um
01:07:08
and the problem with that is that it
01:07:10
kind of ignores the fact that the people
01:07:14
who are putting together the regulation
01:07:17
don't really understand enough about the
01:07:19
detail today you know in their defense
01:07:22
they're rapidly trying to wrap their
01:07:24
head around it especially in in the last
01:07:25
6 months and that's a great relief to me
01:07:28
cuz I feel the burden is now
01:07:29
increasingly shared and you know just
01:07:32
from a personal perspective I'm like I
01:07:35
feel like I've been saying this for
01:07:36
about a decade and just in the last six
01:07:38
months now everyone's coming at me and
01:07:40
saying like you know what's going on I'm
01:07:42
like great this is the conversation we
01:07:44
need to be having because everybody can
01:07:46
start to see the glimmers of the future
01:07:49
like what will happen if a chat GPT like
01:07:52
product or a piie like product really
01:07:54
does improve over the next 10 years and
01:07:58
so when I say you know regulation is not
01:08:00
enough what I mean is it needs movement
01:08:03
it needs culture it needs people who are
01:08:06
actually building and making you know in
01:08:09
like modern creative critical ways not
01:08:11
just like giving it up to you know
01:08:14
companies or small groups of people
01:08:16
right we need lots of different people
01:08:18
experimenting with strategies for
01:08:19
containment isn't it predicted that this
01:08:21
industry is A1 15 trillion dollar
01:08:23
industry or something like that yeah
01:08:26
I've heard that it is a lot so if I'm
01:08:28
rishy and I know that I'm going to be
01:08:30
chucked out of office Rish is the prime
01:08:33
minister of the UK If I'm going to be
01:08:34
trucked out of office in two years
01:08:35
unless this economy gets good I don't
01:08:38
want to do anything to slow down that
01:08:39
$15 trillion dollar bag that I could be
01:08:42
on the receiving end of I would I would
01:08:44
definitely not want to slow that 15
01:08:45
billion trillion dollar bag and give it
01:08:47
to like America or Canada or some other
01:08:50
country I'd want that $15 trillion doll
01:08:53
windfall to be on my country right so I
01:08:58
have I have no other than the long-term
01:09:00
you know health and success of humanity
01:09:03
in my four-year election window I've got
01:09:05
to do everything I can to boost these
01:09:07
numbers right and get us looking good so
01:09:10
I I could give you lip
01:09:12
service but but but listen I'm not going
01:09:15
to be here unless these numbers look
01:09:18
good right exactly that's another one of
01:09:22
the problems short-termism is everywhere
01:09:26
who is responsible for thinking about
01:09:29
the 20-year
01:09:31
future who is it I mean that's a deep
01:09:34
question right I mean we we we the world
01:09:36
is happening to us on a decade by decade
01:09:39
time scale it's also happening hour by
01:09:41
hour so change is just ripping through
01:09:44
us and this arbitrary window of
01:09:46
governance of like a four-year election
01:09:49
cycle where actually it's not even four
01:09:51
years because by the time you've got in
01:09:53
you do some stuff for six months and
01:09:54
then by month you know 12 or 18 you're
01:09:58
starting to think about the next cycle
01:09:59
and are you going to pull you know this
01:10:01
just like the short-termism is killing
01:10:03
us right and we don't have an
01:10:06
Institutional body whose responsibility
01:10:10
is
01:10:11
stability you could think of it as like
01:10:13
a you
01:10:15
know like a global technology stability
01:10:19
function what is the global strategy for
01:10:23
containment that has the ability to to
01:10:25
introduce friction when necessary to
01:10:28
implement the precautionary principle
01:10:31
and to basically keep the
01:10:33
peace that I think is the missing
01:10:36
governance piece which we have to invent
01:10:39
in the next 20 years and it's insane
01:10:41
because I'm basically
01:10:43
describing the UN Security Council plus
01:10:47
the World Trade Organization all these
01:10:50
huge you know Global institutions which
01:10:53
formed after you know the horrors of the
01:10:55
second world war have actually been
01:10:58
incredible they've created
01:11:00
interdependence and alignment and
01:11:02
stability right obviously there's been a
01:11:04
lot of bumps along the way in the last
01:11:05
70 years but broadly speaking it's an
01:11:08
unprecedented period of peace and when
01:11:11
there's peace we can create prosperity
01:11:14
and that's actually what we're lacking
01:11:16
at the moment is that we don't have an
01:11:18
international mechanism for coordinating
01:11:20
among competing Nations competing
01:11:24
corporations um to drive the peace in
01:11:26
fact we're actually going kind of in the
01:11:29
opposite direction we're resorting to
01:11:30
the old school language of a clash of
01:11:33
civilizations with like China is the new
01:11:36
enemy they're going to come to dominate
01:11:37
us we have to dominate them it's a it's
01:11:39
a battle between two polls China's
01:11:41
taking over Africa China's taking over
01:11:43
the Middle East we have to count I mean
01:11:46
it's just like that can only lead to
01:11:49
conflict that just assumes that conflict
01:11:51
is inevitable and so when I say
01:11:54
regulation is not enough no amount of
01:11:56
good regulation in the UK or in Europe
01:11:59
or in the US is going to deal with that
01:12:01
Clash of civilizations language which we
01:12:04
seem to have been become addicted to if
01:12:07
we need that Global collaboration to be
01:12:09
successful here are you optimistic now
01:12:13
that we'll we'll get it because the same
01:12:14
incentives are at play with climate
01:12:16
change in AI you know why would I want
01:12:17
to reduce my carbon emissions when it's
01:12:19
making me loads of money or why you know
01:12:22
why would I want to reduce my AI
01:12:24
development when it's going to make us
01:12:25
15 trillion yeah so the the the really
01:12:29
painful answer to that question is that
01:12:32
we've only really ever driven extreme
01:12:36
compromise and consensus in two
01:12:40
scenarios one off the back of
01:12:44
unimaginable catastrophe and suffering
01:12:46
you know Hiroshima Nagasaki and the
01:12:49
Holocaust and World War II which drove
01:12:52
10 years of consensus and new political
01:12:55
structures right and then the second is
01:13:00
um we did fire the bullet though didn't
01:13:02
we we fired a couple of those nuclear
01:13:05
bombs exactly and that that's why I'm
01:13:07
saying the brutal truth of that is that
01:13:09
it takes a catastrophe to
01:13:13
trigger the need for alignment right so
01:13:16
that that's one the second is where
01:13:19
there is an obvious mutually assured
01:13:22
destruction um you know
01:13:25
Dynamic where both parties are afraid
01:13:29
that this would trigger nuclear meltdown
01:13:32
right and that means suicide and when
01:13:35
there was few parties exactly when there
01:13:38
was just nine people exactly you could
01:13:40
get all nine but in in when we're
01:13:42
talking about artificial technology
01:13:43
there's going to be more than nine
01:13:44
people right that have P access to the
01:13:47
full sort of power of that technology
01:13:49
for NE various reasons I don't think it
01:13:51
has to be like that I think that's the
01:13:54
challenge of containment
01:13:55
is to reduce the number of actors that
01:13:58
have access to the existential threat
01:14:01
Technologies to an absolute minimum and
01:14:03
then use the existing military and
01:14:07
economic incentives which have driven
01:14:09
World Order and peace so far um to to
01:14:12
prevent the proliferation of access to
01:14:14
these super intelligences or these agis
01:14:17
a quick word on hu as you know they're a
01:14:19
sponsor of this podcast and I'm an
01:14:20
investor in the company and I have to
01:14:22
say it's moments like this in my life
01:14:24
where I'm extreme busy and I'm flying
01:14:26
all over the place and I'm recording TV
01:14:28
shows and I'm recording shows in America
01:14:30
and here in the UK that hu is a
01:14:33
necessity in my life I'm someone that
01:14:36
regardless of external circumstances or
01:14:38
professional demands wants to stay
01:14:40
healthy and nutritionally complete and
01:14:42
that's exactly where heal fits in my
01:14:43
life it's enabled me to get all of the
01:14:45
vitamins and minerals and nutrients that
01:14:48
I need in my diet to be aligned with my
01:14:50
health goals while also not dropping the
01:14:52
ball on my professional goals because
01:14:54
it's convenient and because I can get it
01:14:56
online in Tesco in supermarkets all over
01:14:58
the country if you're one of those
01:15:00
people that hasn't yet tried hu or you
01:15:01
have before but for whatever reason
01:15:03
you're not a Hu consumer right now I
01:15:06
would highly recommend giving hu a go
01:15:09
and Tesco have now increased the
01:15:11
listings with hu so you can now get the
01:15:13
RTD ready to drink in Tesco expresses
01:15:15
all across the UK 10 areas of focus for
01:15:18
containment you're the first person I've
01:15:20
met that's really hazarded a laid out a
01:15:23
blueprint for the things that need to be
01:15:24
done
01:15:26
um cohesively to try and reach this
01:15:28
point of containment so I super excited
01:15:29
to talk to you about these the first one
01:15:30
is about safety and you mentioned there
01:15:33
that's kind of what we talked about a
01:15:33
little bit about there being AIS that
01:15:35
are currently being developed to help
01:15:37
contain other
01:15:40
AIS two
01:15:42
audits um which is being able to from
01:15:45
what I understand being able to audit
01:15:47
what's being built in the these open
01:15:48
source models three choke points what's
01:15:52
that yeah so choke point refers to
01:15:56
points in the supply chain where you can
01:15:59
throttle who has access to what okay so
01:16:02
on the internet today everyone thinks of
01:16:04
the internet as an idea this kind of
01:16:07
abstract Cloud thing that hovers around
01:16:09
above our heads but really the internet
01:16:12
is a bunch of cables those cables you
01:16:15
know are physical things that transmit
01:16:18
information you know under the sea and
01:16:22
you know the those points the end points
01:16:24
can be
01:16:25
stopped and you can monitor traffic you
01:16:27
can control basically what traffic moves
01:16:31
back and forth and then the second choke
01:16:33
point is access to chips so the gpus
01:16:38
graphics processing units which are used
01:16:41
to train these super large clusters I
01:16:44
mean we now have the second largest
01:16:46
supercomputer in the world today uh at
01:16:50
least you know just for this next six
01:16:51
months we will other people will catch
01:16:53
up soon but we're ahead of the curve
01:16:54
we're very luy
01:16:55
cost a billion dollars and those chips
01:16:59
are really the raw commodity that we use
01:17:02
to build these large language models and
01:17:05
access to those chips is something that
01:17:07
governments can should and are um you
01:17:11
know restricting that's a choke point
01:17:13
you spent a billion dollars on a
01:17:14
computer we did yeah it's bit more than
01:17:16
that actually about
01:17:20
1.3 a couple of years time that'll be
01:17:23
the price of an iPhone
01:17:25
that's the problem everyone's going to
01:17:27
have
01:17:28
it number six is quite curious you say
01:17:31
that um the need for governments to put
01:17:32
increased taxation on AI companies to be
01:17:35
able to find um fund the massive changes
01:17:37
in society such as paying for reskilling
01:17:40
and education yeah
01:17:43
um you put massive tax on over here I'm
01:17:45
going to go over
01:17:47
here if you tax it if I'm an AI company
01:17:49
and you're taxing me heavily over here
01:17:51
I'm going to Dubai yep or Portugal yep
01:17:56
so if it's that much of a competitive
01:17:58
disadvantage I will not build my company
01:18:00
where the taxation's high right
01:18:04
right so the way to think about this is
01:18:07
what are the strategies for containment
01:18:09
if we're agreed that long-term we want
01:18:12
to contain that is close down slow down
01:18:16
control both the proliferation of these
01:18:18
Technologies and the way the really big
01:18:20
AIS are used then the way to do that is
01:18:24
to tax things tax things taxing things
01:18:27
slows them down and that's what you're
01:18:29
looking for provided you can coordinate
01:18:31
internationally so you're totally right
01:18:34
that you know some people will move to
01:18:35
Singapore or to Abu Dhabi or Dubai or
01:18:38
whatever the reality is that at least
01:18:41
for the next you know sort of period I
01:18:43
would say 10 years or so the
01:18:45
concentrations of intellectual you know
01:18:48
horsepower will remain the big mega
01:18:52
cities right you know I I moved from
01:18:54
from London in 2020 to go to Silicon
01:18:57
Valley and I started my new company in
01:18:59
Silicon Valley because the concentration
01:19:01
of talent there is overwhelming all the
01:19:03
very best people are there on in in Ai
01:19:06
and software engineering so I think it's
01:19:09
quite likely that that's going to remain
01:19:11
the case for the foreseeable future but
01:19:13
in the long term you're totally right
01:19:15
how do you it's another coordination
01:19:16
problem how do we get nation states to
01:19:19
collectively agree that we want to try
01:19:22
and contain that we want to slow down
01:19:24
because
01:19:25
as we've discussed with the
01:19:26
proliferation of dangerous materials or
01:19:28
on the military side there's no use one
01:19:30
person doing it or one country doing it
01:19:33
if others race ahead and that's the
01:19:35
conundrum that we face I am I don't
01:19:38
consider myself to be a pessimist in my
01:19:39
life I consider myself to be an optimist
01:19:42
generally I think and I always I think
01:19:44
that as you've said I think we have no
01:19:45
choice but to be optimistic and I have
01:19:47
faith in humanity we've done so much so
01:19:49
many incredible things and so overcome
01:19:50
so many things and I also think I'm
01:19:52
really logical as in I'm the type of
01:19:54
person that needs evidence to change my
01:19:56
beliefs either way um when I look at all
01:19:59
of the whole picture having spoken to
01:20:01
you and several others on this subject
01:20:03
matter I see more reasons why we won't
01:20:06
be able to contain than reasons why we
01:20:07
will especially when I dig into those
01:20:10
incentives um you talk about incentives
01:20:13
at length in your book um at different
01:20:15
different points and it's clear that all
01:20:17
the incentives are pushing towards a
01:20:20
lack of containment especially in the
01:20:22
short and Midterm which tends to happen
01:20:23
with new technology in the short and
01:20:25
Midterm it's like a land grab the gold
01:20:27
is in the Stream we all rush to get the
01:20:29
the shovels and the you know the cves
01:20:31
and stuff and then we realize the
01:20:33
unintended consequences of that
01:20:35
hopefully not before it's too
01:20:37
late in chapter 8 you talk about
01:20:40
Unstoppable incentives at play here the
01:20:43
coming wave represents the greatest
01:20:44
economic prize in
01:20:47
history and scientists and technologists
01:20:50
are all too human they crave status
01:20:52
success and Legacy
01:20:55
and they want to be recognized as the
01:20:57
first and the best they're competitive
01:20:59
and clever with a carefully nurtured
01:21:01
sense of their place in the world and in
01:21:04
history
01:21:06
right I look at you I look at people
01:21:08
like Sam um from open AI
01:21:12
Elon you're all
01:21:14
humans with the same understanding of
01:21:16
your place in history and status and
01:21:19
success you all want that right
01:21:23
right there's a lot of people that maybe
01:21:25
aren't as don't have as good a track
01:21:28
record of you at doing the right thing
01:21:29
which you certainly have that will just
01:21:31
want the status and the success and the
01:21:33
money incredibly strong incentives I
01:21:36
always think about incentives as being
01:21:37
the thing that you look at when you want
01:21:39
to understand how people will behave all
01:21:41
of the incentives on a on a geopolitical
01:21:44
like on a global level suggest that
01:21:48
containment won't happen am I right in
01:21:51
that
01:21:52
assumption that all the incentives
01:21:54
suggest containment won't happen in the
01:21:56
short or midterm until there is a c a a
01:21:59
tragic event that makes us forces us
01:22:01
towards that idea of containment or if
01:22:05
there is a threat of mutually assured
01:22:08
destruction right so that and that's the
01:22:11
case that I'm trying to make is that
01:22:14
let's not wait for something
01:22:16
catastrophic to happen so it's
01:22:18
self-evident that we all have to work
01:22:20
towards containment right I mean you you
01:22:23
would have thought that the Potential
01:22:26
Threat the potential idea that
01:22:31
covid-19
01:22:33
was a side effect let's call it of a
01:22:37
laboratory in Wuhan that was exploring
01:22:40
gain of function research where it was
01:22:42
deliberately trying to basically make
01:22:45
the pathogen more transmissible you
01:22:48
would have thought that warning to all
01:22:50
of us let's let's not even debate
01:22:52
whether it was or wasn't but just the
01:22:54
fact that it's conceivable that it could
01:22:56
be that should really in my opinion have
01:23:00
forced all of us to instantly agree that
01:23:04
this kind of research should just be
01:23:05
shut down we should just not be doing
01:23:07
gain of function research on what planet
01:23:09
could we possibly persuade ourselves
01:23:12
that we can overcome the containment
01:23:14
problem in biology because we've proven
01:23:16
that we can't cuz it could have
01:23:19
potentially got out and there's a number
01:23:20
of other examples of where it did get
01:23:21
out of other diseases like foot and
01:23:23
mouth disease
01:23:25
mhm back in the '90s in the UK so but
01:23:29
that didn't change our Behavior right
01:23:32
well foot and mouth disease clearly
01:23:33
didn't cause enough harm because it only
01:23:35
killed a bunch of cattle right um and
01:23:38
the pandemic we can't seem you know
01:23:40
covid-19 pandemic we can't seem to agree
01:23:42
you know that it really was from a lab
01:23:45
and not from a bunch of bats right and
01:23:49
so that's where I struggle where you
01:23:52
know now you catch me in a moment where
01:23:53
I feel angry and sad and pessimistic
01:23:57
because to me that's like a
01:23:59
straightforwardly obvious conclusion
01:24:01
that you know this is a type of research
01:24:04
that we should be closing down and I
01:24:05
think we should be using these moments
01:24:09
to give us insight and wisdom about how
01:24:11
we handle other technology trajectories
01:24:15
in the next few decades should we should
01:24:18
we should that's what I'm advocating for
01:24:20
must that's the best I can do I want to
01:24:22
know will will I think be a
01:24:25
low I can only do my best I'm doing my
01:24:28
best to advocate for it I mean you know
01:24:30
like I'll give you an example like I
01:24:32
think autonomy is a type of AI
01:24:35
capability that we should not be
01:24:37
pursuing really like autonomous cars and
01:24:40
stuff well I I autonomous cars I think
01:24:43
are slightly different because
01:24:44
autonomous cars operate within a much
01:24:46
more constrained physical domain right
01:24:49
like you know you you really can the
01:24:52
containment strategies for autonomous
01:24:53
cars are quite reassuring right they
01:24:56
have you know GPS control you know we
01:24:59
know exactly all the Telemetry and how
01:25:01
exactly all of those you know components
01:25:03
on board a car operate and we can
01:25:06
observe repeatedly that it behaves
01:25:08
exactly as intended right whereas I
01:25:12
think with with other forms of autonomy
01:25:14
that people might be pursuing like
01:25:16
online okay you know where you have an
01:25:18
an AI that is like designed to
01:25:20
self-improve without any human oversight
01:25:23
or a battle Battlefield weapon which you
01:25:26
know like unlike a car hasn't been you
01:25:28
know over that particular moment in the
01:25:31
battlefield millions of times but is
01:25:33
actually facing a new enemy every time
01:25:35
you know every single time and we're
01:25:37
just going to go and you know allow
01:25:39
these autonomous weapons to have you
01:25:41
know the these autonomous military
01:25:43
robots to have lethal
01:25:45
Force I think that's something that we
01:25:47
should really resist I don't think we
01:25:50
want to have autonomous robots that have
01:25:53
lethal Force you're a super smart guy
01:25:55
and I I struggle to believe that you're
01:25:58
you you because you you demonstrate such
01:26:00
a clear understanding of the incentives
01:26:02
in your book that I struggle to believe
01:26:04
that you don't think the incentives will
01:26:07
win out especially in the short and near
01:26:09
term and then the problem is in the
01:26:10
short and near term as is the case with
01:26:12
most of these waves is we we wake up and
01:26:17
10 years time ago how the hell did we
01:26:19
get here right and why like and we and
01:26:21
as you say this precautionary approach
01:26:23
of we should have ranged the Bell
01:26:24
earlier we should have sounded the alarm
01:26:25
earlier but we waltzed in with optimism
01:26:28
right and with that kind of aversion to
01:26:31
confronting the realities of it and then
01:26:34
we woke up in 30 years and we're on a
01:26:36
leash right and there's a big rottweiler
01:26:38
and we're we've lost control we've lost
01:26:40
you know I
01:26:44
I I I would love to
01:26:47
know someone as smart as you I don't I
01:26:50
don't believe can be can believe that
01:26:52
containment is
01:26:55
possible and that's me just being
01:26:57
completely honest I'm not saying you're
01:26:58
lying to me but I just can't see how
01:26:59
someone as smart as you and in the know
01:27:01
as you can believe that containment is
01:27:03
going to happen well I didn't say it is
01:27:05
possible I said it must be right which
01:27:07
is this is what we keep discussing right
01:27:09
that's an important distinction is that
01:27:11
on the face of it look what I I care
01:27:13
about I care about science I care about
01:27:16
facts I care about describing the world
01:27:19
as I see it and what I've set out to do
01:27:22
in the book is des describe a set of
01:27:25
interlocking incentives which drive a
01:27:27
technology production process which
01:27:30
produces potentially really dangerous
01:27:33
outcomes and what I'm trying to do is
01:27:36
frame those outcomes in the context of
01:27:38
the containment problem and say this is
01:27:40
the big challenge of the 21st century
01:27:43
containment is the challenge and if it
01:27:45
isn't possible then we have serious
01:27:47
issues and on the face of it like I've
01:27:49
said in the book I mean the CH the first
01:27:50
chapter is called containment is not
01:27:52
possible right the last chapter is
01:27:53
called con must be possible for all our
01:27:55
sakes it must be possible but that but I
01:27:58
agree with you that I'm not I'm not
01:27:59
saying it is I'm saying this is what we
01:28:02
have to be working on we have no choice
01:28:04
we have no choice but to work on this
01:28:06
problem this is a critical
01:28:08
problem how much of your time are you
01:28:10
focusing on this problem basically all
01:28:13
my time I mean bu building and creating
01:28:15
is about understanding how these models
01:28:19
work what their limitations are how to
01:28:21
build it safely and ethically I mean we
01:28:23
have designed the structure of the
01:28:25
company to focus on the safety and
01:28:28
ethics aspects so for example we are a
01:28:30
public benefit Corporation right which
01:28:32
is a new type of corporation which gives
01:28:35
us a legal obligation to balance profit
01:28:40
making with the consequences of our
01:28:44
actions as a company on the rest of the
01:28:47
world the way that we affect the
01:28:48
environment you know the way that we
01:28:51
affect people the way that we affect
01:28:53
users that people who aren't users of
01:28:55
our products and that's a really
01:28:59
interesting I think and important New
01:29:02
Direction it's a new Evolution in
01:29:04
corporate structure because it says we
01:29:06
have a responsibility to proactively do
01:29:09
our best to do the right thing right and
01:29:12
I think that if if you were a tobacco
01:29:15
company back in the day or an oil
01:29:17
company back in the day and your legal
01:29:19
Charter said that your directors are
01:29:22
liable in if they don't meet the
01:29:25
criteria of stewarding your work in a
01:29:29
way that doesn't just optimize profit
01:29:31
which is what all companies are
01:29:32
incentivized to do at the moment talking
01:29:33
about incentives but actually in equal
01:29:36
measure attends to the importance of
01:29:38
doing good in the world to me that's a
01:29:42
incremental but important innovation in
01:29:45
how we organize society and how we
01:29:48
incentivize our work so it doesn't solve
01:29:51
everything it's it's it's not a Panacea
01:29:53
but that's my effort to try and take a
01:29:56
small step in the right direction do you
01:29:57
ever get sad about it about what's
01:29:59
happening yeah for sure for sure it's
01:30:08
intense it's intense it's a lot to take
01:30:10
in this is it's a it's a very
01:30:14
real
01:30:17
reality does that weigh on
01:30:20
you yeah it does I mean every day every
01:30:23
day I mean I've been working on this for
01:30:25
many years now and it's uh you know it's
01:30:28
it's
01:30:29
emotionally a lot to take in it's it's
01:30:32
it's hard to think about the far out
01:30:35
future and how your actions today our
01:30:39
actions collectively our weaknesses our
01:30:42
failures that you know that irritation
01:30:44
that I have that we can't learn the
01:30:47
lessons from the pandemic right like all
01:30:50
of those moments where you feel the
01:30:53
frustration governments not working
01:30:55
properly or corporations not listening
01:30:58
or some of the obsessions that we have
01:31:00
in culture where we're debating like
01:31:03
small things you know and you're just
01:31:06
like
01:31:07
Whoa We need to focus on the big picture
01:31:10
here you must feel a certain sense of
01:31:12
responsibility as well that most people
01:31:14
won't carry because you've spent so much
01:31:17
of your life at the very cutting edge of
01:31:18
this technology and you understand it
01:31:20
better than most you can speak to it
01:31:22
better than most so you have a a great
01:31:24
chance than many at
01:31:27
steering that's a
01:31:29
responsibility yeah I embrace that I try
01:31:32
to treat that as a
01:31:35
privilege I feel lucky to have
01:31:39
the opportunity to try and do that
01:31:42
there's this wonderful thing in my
01:31:44
favorite theatrical play called Hamilton
01:31:47
where he says history has its eyes on
01:31:50
you do you feel
01:31:52
that yeah I feel the I feel that I feel
01:31:55
that it's a good way of putting it I do
01:31:58
feel
01:32:01
that you're happy
01:32:04
right well what is happiness to
01:32:10
know um what's the range of emotions
01:32:13
that you you contend with on a on a
01:32:16
frequent basis if you're being
01:32:19
honest I think
01:32:24
is kind of exhausting and exhilarating
01:32:27
in equal measure because for me it is
01:32:32
beautiful to see people interact with
01:32:36
AIS and get huge benefit out of it I
01:32:39
mean you know every day now millions of
01:32:42
people have a super smart tool in their
01:32:46
pocket that is making them wiser and
01:32:48
healthier and happier providing
01:32:50
emotional support answering questions of
01:32:53
every type
01:32:54
making you more intelligent and so on
01:32:56
the face of it in the short term that
01:32:57
feels incredible it's amazing what we're
01:32:59
all
01:33:00
building but in the longer term it is
01:33:04
exhausting to keep making this argument
01:33:07
and you know have been doing it for a
01:33:10
long time and in a weird way I feel a
01:33:13
bit of a sense of relief in the last six
01:33:15
months because after chat gbt and you
01:33:19
know this this wave feels like it's
01:33:21
started to arrive and everybody gets it
01:33:24
so I feel like it's a shared problem
01:33:27
now and uh that feels nice it's not just
01:33:32
bouncing around in your head a little
01:33:33
bit it's not just in my head and a few
01:33:35
other people at Deep Mind and open Ai
01:33:37
and other places that have been talking
01:33:39
about it for a long
01:33:40
time ultimately human beings May no
01:33:43
longer be the primary planetary drivers
01:33:46
as we have become accustomed to being we
01:33:48
are going to live in an Epoch where the
01:33:50
majority of our daily interactions are
01:33:52
not with other people but with a eyes
01:33:54
page
01:33:56
284 of your
01:33:59
book The Last Page
01:34:01
[Laughter]
01:34:04
yeah think about how much of your day
01:34:08
you spend looking uh
01:34:11
screen 12 hours pretty much right
01:34:14
whether it's a phone or an iPad or a
01:34:16
desktop versus how much time you spend
01:34:19
looking into the eyes of your friends
01:34:21
and your loved
01:34:22
ones and so to me it's like we're
01:34:26
already there in a way you know what I
01:34:30
meant by that was you
01:34:33
know this is a world that we're kind of
01:34:37
already in you know the last three years
01:34:39
people have been talking about metaverse
01:34:40
metaverse metaverse and the
01:34:43
mischaracterization of the metaverse was
01:34:45
that it's over there it was this like
01:34:48
virtual world that we would all Bop
01:34:49
around in and talk to each other as
01:34:51
these little characters and but that was
01:34:54
totally wrong that was a complete
01:34:56
misframing the metaverse is already here
01:35:01
it's the digital space that exists in
01:35:04
parallel time to our everyday life it's
01:35:07
the conversation that you will have on
01:35:10
Twitter or you know the video that
01:35:12
you'll post on YouTube or this podcast
01:35:14
that will go out and connect with other
01:35:16
people it's that meta space of
01:35:19
interaction you know and I use meta to
01:35:22
mean Beyond this space not just that
01:35:26
weird other over there space that people
01:35:29
seem to point to and that's really what
01:35:33
is emerging here it's this parallel
01:35:36
digital space that is going to live
01:35:38
alongside with and in relation to our
01:35:41
physical world your kids come to you you
01:35:44
got kids no I don't have kids your
01:35:46
future kids if you ever have kids a
01:35:48
young child walks up to you and says
01:35:51
asks that question that Elon was asks
01:35:53
what should I do about with my future
01:35:55
what should I pursue in the light of
01:35:58
everything you know about how artificial
01:36:00
intelligence is going to change the
01:36:01
world and computational power and all of
01:36:03
these things what should I dedicate my
01:36:05
life to what do you say I would say
01:36:08
knowledge is power
01:36:11
Embrace understand grapple with the
01:36:14
consequences don't look the other way
01:36:16
when it feels
01:36:18
scary and do everything you can to
01:36:22
understand and part anticipate and shape
01:36:25
because it is
01:36:30
coming and if someone's listening to
01:36:32
this and they want to do something to
01:36:33
help this battle for which I think you
01:36:36
present as the solution
01:36:38
containment what can the individual
01:36:42
do read listen use the tools try to make
01:36:48
the
01:36:48
tools understand the current state of
01:36:51
Regulation see which organization are
01:36:54
organizing around it like you know
01:36:56
campaign groups activism groups you know
01:37:00
find solidarity connect with other
01:37:03
people spend time online ask these
01:37:05
questions mention it at the pub you know
01:37:09
ask your parents ask your mom how she's
01:37:11
reacting to you know talking to Alexa or
01:37:13
whatever it is that she might do pay
01:37:16
attention I think that's already enough
01:37:19
and there's no need to be more
01:37:21
prescriptive than that because I think
01:37:22
people are
01:37:24
creative and independent and will it
01:37:26
will it will be obvious to you what you
01:37:30
as an individual feel you need to
01:37:32
contribute In This Moment provided
01:37:34
you're paying
01:37:37
attention last question what if we fail
01:37:41
and what if we succeed what if we fail
01:37:43
in containment and what if we succeed in
01:37:45
containment of artificial
01:37:48
intelligence I honestly think that if we
01:37:51
succeed this is going to be the the most
01:37:55
productive and the most meritocratic
01:37:57
moment in the history of our species we
01:38:00
are about to make intelligence widely
01:38:04
available to hundreds of millions if not
01:38:07
billions of people and that is all going
01:38:09
to make us smarter and much more
01:38:11
creative and much more productive and I
01:38:13
think over the next few decades we will
01:38:16
solve many of our biggest Social
01:38:18
Challenges I really believe that I
01:38:21
really believe we're going to reduce the
01:38:22
cost of energy production storage and
01:38:24
distribution to zero marginal cost we're
01:38:26
going to reduce the cost of producing
01:38:28
healthy food and make that widely
01:38:29
available to everybody and I
01:38:33
think the same trajectory with healthc
01:38:35
care with Transportation with
01:38:37
education I think that ends up producing
01:38:42
radical abundance over a 30-year period
01:38:45
and in a world of radical abundance what
01:38:46
do I do with my day I think that's
01:38:49
another profound question and believe me
01:38:50
that is a good problem to have if we can
01:38:53
absolutely if do we don't need meaning
01:38:55
and purpose and oh man that is a better
01:38:58
problem to have than what we've just
01:38:59
been talking about for the last like 90
01:39:02
minutes yeah and I think that's
01:39:04
wonderful isn't that amazing I don't
01:39:06
know I I don't know the reason I I I'm
01:39:08
unsure is because everything that seems
01:39:11
wonderful has a has a unintended
01:39:13
consequence I'm sure it does we live in
01:39:15
a world of food abundance in the west
01:39:17
and our biggest problem is obesity right
01:39:19
so I'll take that problem in the grand
01:39:21
scheme of everything not need struggle
01:39:25
do we not need that kind of meaningful
01:39:27
voluntary struggle I think we'll create
01:39:30
new other you know opportunities to
01:39:34
Quest okay you know I I think that's an
01:39:37
easier problem to solve and I think it's
01:39:38
an amazing problem like many people
01:39:40
really don't want to work right they
01:39:42
they want to pursue their passion and
01:39:43
their Hobby and you know all the things
01:39:45
that you talk about and so on and
01:39:47
absolutely like we're now I think going
01:39:49
to be heading towards a world where we
01:39:51
can liberate people from the the Les of
01:39:53
work unless you really want to Universal
01:39:56
basic income I've long been an advocate
01:39:58
of Ubi very long time everyone gets a
01:40:01
check every month I don't think it's
01:40:03
going to quite take that form I actually
01:40:06
think it's going to be that we basically
01:40:09
reduce the cost of producing basic Goods
01:40:12
so that you're not as dependent on
01:40:14
income like imagine if you did have
01:40:16
basically free energy and food and you
01:40:20
you you could use that free energy to
01:40:21
grow your own food you could grow in a
01:40:23
desert because you would have adapted
01:40:25
seeds and so on you would have you know
01:40:28
desalination and so on that really
01:40:30
changes the structure of cities it
01:40:31
changes the structure of Nations it
01:40:33
means that you really can live in quite
01:40:35
different ways for very extended periods
01:40:38
without contact with the kind of Center
01:40:40
I mean I'm actually not a huge advocate
01:40:42
of that kind of libertarian you know wet
01:40:44
dream but like I think if you think
01:40:46
about it in theory it's kind of a really
01:40:49
interesting Dynamic that's what
01:40:50
proliferation of power means power isn't
01:40:52
just about access to intelligence it's
01:40:54
about access to these tools which allow
01:40:57
you to take control of your own destiny
01:40:59
and your life and create meaning and
01:41:01
purpose in the way that you you know
01:41:03
might Envision and that's incredibly
01:41:05
creative incredibly creative time that's
01:41:08
what success looks like to me
01:41:11
and well in some ways the downside of
01:41:14
that I think this that failure is not
01:41:17
achieving a world of radical abundance
01:41:20
in my opinion and and more more
01:41:22
importantly failure is a failure to
01:41:25
contain right what does that lead
01:41:29
to I think it leads to a mass
01:41:31
proliferation of power and people who
01:41:33
have really bad you know intentions what
01:41:37
does that lead to will potentially use
01:41:39
that power to cause harm to others this
01:41:42
is part of the challenge right a small
01:41:45
in this networked globalized World a
01:41:48
tiny group of people who wish to
01:41:51
deliberately cause harm
01:41:53
are going to have access to tools that
01:41:56
can instantly quickly have large scale
01:41:59
impact on many many other people and
01:42:02
that's the challenge of proliferation is
01:42:04
preventing those Bad actors from getting
01:42:06
access to the means to completely
01:42:09
destabilize um our world that's what
01:42:12
containment is
01:42:15
about we have a closing tradition on
01:42:17
this podcast where the last guest leaves
01:42:18
a question for the next guest not
01:42:19
knowing who they're leaving the question
01:42:21
for the question left for you is
01:42:24
what is a space or place that you
01:42:27
consider the most
01:42:32
sacred well I think one of the most
01:42:35
beautiful places I remember going to as
01:42:38
a child was um windir Lake in the Lake
01:42:43
District um and I was pretty young and
01:42:48
on a on a dingy with uh some family
01:42:52
members
01:42:53
and I just remember it being incredibly
01:42:56
Serene and beautiful and and calm I
01:42:58
actually haven't been back there since
01:43:01
but that was a pretty beautiful place
01:43:05
seems like the antithesis of the world
01:43:06
we live in right maybe I should go back
01:43:09
there and chill
01:43:11
out maybe thank you so much for writing
01:43:13
such a great book it's wonderful to to
01:43:15
to read a book on this subject matter
01:43:17
that does present Solutions because not
01:43:19
many of them do and it presents them in
01:43:21
a balanced way that appreciates both
01:43:23
sides of the argument doesn't isn't
01:43:25
tempted to just play to either what do
01:43:27
they call it playing to like the crowd
01:43:29
they call like playing to the orchestra
01:43:30
I can't remember right but just it
01:43:31
doesn't attempt to play to either side
01:43:33
or Ponder to either side in order to
01:43:34
score points it seems to be entirely
01:43:37
nuanced incredibly smart and Incredibly
01:43:41
necessary because of the stakes that the
01:43:43
book confronts um that are at play in
01:43:46
the world at the moment and and that's
01:43:49
really important it's very very very
01:43:51
important and it's important that I
01:43:52
think everybody reads this book it's
01:43:54
incredibly accessible as well and I said
01:43:56
to Jack who's the director of this
01:43:58
podcast before we started recording that
01:44:01
there's so many term there's so many
01:44:02
terms like
01:44:04
nanotechnology and um all the stuff
01:44:06
about like biotechnologies and Quantum
01:44:09
Computing that reading through the book
01:44:11
suddenly I understood what they meant
01:44:13
and these had been kind of exclusive ter
01:44:15
terms and Technologies and I also had
01:44:17
never understood the relationship that
01:44:19
all of these Technologies now have with
01:44:21
each other and how like robotics
01:44:23
emerging with artificial intelligence is
01:44:25
going to cause this whole new range of
01:44:27
possibilities that again have a good
01:44:30
side and a potential downside um It's a
01:44:32
Wonderful book and it's perfectly timed
01:44:34
it's perfectly timed wonderfully written
01:44:36
perfectly timed I'm so thankful that I
01:44:38
got to read it and I highly recommend
01:44:40
that anybody that's curious on this
01:44:41
subject matter goes and gets the book so
01:44:44
thank you Mustafa really really
01:44:45
appreciate your time and hopefully it
01:44:46
wasn't too uncomfortable for you thank
01:44:48
you this was awesome I loved it it was
01:44:50
really fun and uh thanks for such a
01:44:52
amazing amazing wide ranging
01:44:54
conversation thank
01:44:57
you if you've been listening to this
01:44:59
podcast over the last few months you'll
01:45:01
know that we're sponsored and supported
01:45:03
by Airbnb but it amazes me how many
01:45:05
people don't realize they could actually
01:45:07
be sitting on their very own Airbnb for
01:45:09
me as someone who works away a lot it
01:45:11
just makes sense to Airbnb my place at
01:45:13
home whilst I'm away if your job
01:45:15
requires you to be away from home for
01:45:17
extended periods of time why leave your
01:45:19
home empty you can so easily turn your
01:45:22
home into an Airbnb and let it generate
01:45:23
income for you whilst you're on the road
01:45:26
whether you could use a little extra
01:45:27
money to cover some bills or for
01:45:29
something a little bit more fun your
01:45:31
home might just be worth more than you
01:45:32
think and you can find out how much it's
01:45:34
worth at airbnb.co
01:45:36
/host that's
01:45:38
airbnb.co slost
01:45:41
[Music]
01:45:53
ah
01:46:00
[Music]
