00:00:00
We have video evidence of people abusing
00:00:04
children and not a single person gots
00:00:06
arrested. How can you call that a
00:00:08
democracy? So, humanity is at a
00:00:10
crossroads where for the first time ever
00:00:12
we need to wake up and realize that
00:00:14
we're ruled by maniacs and what we
00:00:16
believe is democracy is not democracy
00:00:18
and what we know is not the truth. Like
00:00:20
companies and governments will blame the
00:00:23
geopolitical and economic challenges we
00:00:25
have on AI but the truth is AI is not
00:00:28
the enemy. Like I'm not worried about AI
00:00:30
turning against us. I'm worried about
00:00:32
humans telling AI to turn against us.
00:00:34
Like when I worked at Google, we were
00:00:36
building amazing things believing that
00:00:38
we were making the world a better place.
00:00:40
And we were. But then suddenly there is
00:00:42
a moment where you recognize that maybe
00:00:44
the world will not use what you're
00:00:46
making the way you want it to be used.
00:00:48
And sadly this is upon us.
00:00:51
>> So I have lots of questions.
00:00:52
>> Okay, that's good.
00:00:53
>> So what's your take on this job
00:00:55
disruption point? What is the risk of
00:00:57
these very intelligent models that the
00:00:59
creators of these models don't actually
00:01:00
understand themselves? Do you think Sam
00:01:02
Alman's prohumity? How do we get to a
00:01:03
point of ethical AI when the incentive
00:01:05
structures are so highly competitive?
00:01:07
And then I wonder if there's a path that
00:01:09
ends in AI being net positive for
00:01:11
humanity.
00:01:12
>> Somehow we've been pre-programmed to
00:01:14
believe that this is upon us and we
00:01:15
cannot change it. And I refuse that. So
00:01:18
we will talk about the solutions.
00:01:19
>> But are you optimistic?
00:01:21
>> I'm very optimistic about the future.
00:01:22
I'm not optimistic about the next year.
00:01:24
>> Why the next year? Come on, Stephen. You
00:01:26
don't want me to say it.
00:01:29
So,
00:01:32
>> this is super interesting to me. My team
00:01:34
given me this report to show me how many
00:01:35
of you that watch this show subscribe.
00:01:36
And some of you have told us according
00:01:38
to this that you are unsubscribed from
00:01:40
the channel randomly. So, favor to ask
00:01:42
all of you, please could you check right
00:01:44
now if you've hit the subscribe button
00:01:45
if you are a regular viewer of the show
00:01:46
and you like what we do here. We're
00:01:48
approaching quite a significant landmark
00:01:49
on this show in terms of a subscriber
00:01:51
number. So, if there was one simple free
00:01:54
thing that you could do to help us, my
00:01:55
team, everyone here, to keep this show
00:01:57
free, to keep it improving year over
00:01:59
year and week over week, it is just to
00:02:01
hit that subscribe button and to double
00:02:02
check if you've hit it. Only thing I'll
00:02:04
ever ask of you, do we have a deal? If
00:02:06
you do it, I'll tell you what I'll do.
00:02:08
I'll make sure every single week, every
00:02:10
single month, we fight harder and harder
00:02:11
and harder and harder to bring you the
00:02:13
guests and conversations that you want
00:02:14
to hear. I've stayed true to that
00:02:15
promise since the very beginning of the
00:02:16
D of Sio, and I will not let you down.
00:02:20
Please help us. really appreciate it.
00:02:21
Let's get on with the show.
00:02:26
>> Mo Gala, I spoke to you, I think about
00:02:30
four years ago when you wrote a book
00:02:31
about happiness and I remember you came
00:02:33
in, you'd written this book about AI,
00:02:34
but I particularly wanted to speak about
00:02:36
the subject of happiness cuz I was
00:02:38
fascinated by it. What I find
00:02:40
astonishing is the fact that you were
00:02:42
talking about AI before anybody was
00:02:44
really talking about AI. No guest that
00:02:46
had ever come on my podcast had ever
00:02:48
mentioned the subject of AI. it just
00:02:49
wasn't interesting to the world. And
00:02:51
then this thing called Chanty PT came
00:02:53
out and suddenly everybody got to feel
00:02:55
it for themselves and became fascinated
00:02:57
by it. My first question and this might
00:03:00
be a question for people that don't know
00:03:01
you is why at that time did you start
00:03:04
talking about AI before anybody else?
00:03:07
>> I knew them in the lab. I um I joined
00:03:10
Google in 2007
00:03:13
very late 2006. uh and at the time most
00:03:17
people don't know that at the time we
00:03:19
had reasonably established AI doing our
00:03:22
backend work and 2008
00:03:25
we had the cat paper which was published
00:03:28
2009 the first real unprompted AI uh I
00:03:32
remember 2016
00:03:34
I had that incident where I was you know
00:03:37
observing a project we were uh funding
00:03:39
that was about teaching grippers uh how
00:03:42
to grip um unlike like industrial
00:03:44
machinery. Uh so to be to be able to
00:03:47
grip like a human needs a very high
00:03:49
level of intelligence to to to assess
00:03:52
the texture, the softness, the
00:03:53
positioning and so on, the shape of
00:03:55
everything. And we were doing that and
00:03:58
it it just blew my mind how similar to
00:04:01
my kids they were. And I think that was
00:04:03
my very first realization that we were
00:04:05
building the apex of intelligence. we
00:04:08
were genuinely handing over the reigns
00:04:11
of super intelligence to another being,
00:04:14
right? And when you when you get faced
00:04:17
with that, you start to suddenly realize
00:04:19
something that we at Google found very
00:04:21
difficult to realize, which was that
00:04:24
everyone I knew at Google till then uh
00:04:28
was believing that we were making the
00:04:30
world a better place and we were we
00:04:32
genuinely were were doing amazing things
00:04:34
for the world. But then suddenly there
00:04:37
is a moment where you uh where you
00:04:39
recognize that
00:04:41
maybe the world will not use what you're
00:04:43
making the way you want it to be used.
00:04:46
And and you can see that in lots of
00:04:47
technologies. You know, social media
00:04:49
starts by the claim that it's going to
00:04:51
get us connected and gets us closer, but
00:04:53
eventually ends up separating us with
00:04:55
that little screen. You know, dating
00:04:57
apps are giving you the promise that
00:04:59
you're going to find your soulmate, but
00:05:00
in reality, they keep renewing month
00:05:02
after month. And so tech uh somehow ends
00:05:06
up being more capitalist than
00:05:09
altruistic. And I think I was I wasn't
00:05:11
the first. Nick Bostonramm started you
00:05:14
know Jeffrey Hinton completely changed
00:05:16
his mind. Failey Le starting to say you
00:05:19
know this is very serious. Everyone now
00:05:22
everyone who's ever had a very deep
00:05:23
relationship with the machines is a bit
00:05:26
concerned. I wonder if there's a path
00:05:29
that's hard to see now that ends in AI
00:05:31
being net positive for humanity.
00:05:34
>> I bet 100% on that. It's that this path
00:05:38
is very painful.
00:05:40
>> This path is very painful.
00:05:41
>> Yeah. So, so the example you need to
00:05:43
understand is you know we discovered uh
00:05:46
nuclear power and the very first
00:05:49
implementation was a nuclear bomb not
00:05:51
not nuclear energy right and uh and I
00:05:55
think that's exactly what's happening
00:05:56
with AI uh the first implementations of
00:05:59
AI are in favor of a few at the expense
00:06:03
of the majority you know the the in
00:06:06
favor of the capitalist to increase
00:06:08
productivity and reduce cost but not
00:06:11
taking into account how that impacts on
00:06:14
the general public. Uh, you know, in
00:06:16
favor of the of the armies uh that are
00:06:19
now competing with autonomous weapons,
00:06:22
in favor of the surveillance systems
00:06:25
that are attempting to control
00:06:27
everything with more and more and more
00:06:29
intelligence and more monitoring. And
00:06:33
that's not AI waking up in the morning
00:06:36
and saying, "Hey, you know what? Let's
00:06:37
oppress all humans." But it is a
00:06:40
powerful few that are simply deciding to
00:06:43
use the ultimate superpower on the
00:06:45
planet today to gain more power and more
00:06:48
control. I mean, as we speak, we're
00:06:50
living in two major wars where AI is
00:06:53
doing most of the killing.
00:06:54
>> Cuz a lot of people think of AI as these
00:06:56
like chat bots that we're all using to
00:06:58
help us, right?
00:06:59
>> No, I think I think there is a hype I
00:07:01
call it the hype dichotomy if you want.
00:07:04
So, so what the general public sees
00:07:07
about AI is overhyped but
00:07:11
ineffective. You know, all of the fake
00:07:14
videos and all of the, you know, um,
00:07:17
Grock did this and, you know, we
00:07:20
attempted to switch off that machine and
00:07:22
it did that and so on. What the real
00:07:24
geeks see inside the lab is just
00:07:29
unbelievable intelligence. And so what
00:07:32
is about to happen is that we've started
00:07:34
to put together systems that develop
00:07:36
themselves. They look at their own code
00:07:38
and they you know they run experiments
00:07:40
and they test those experiments if they
00:07:42
changed something and they see where the
00:07:44
machines are uh you know the performance
00:07:46
is and and redeploy the best code. Okay.
00:07:50
And and if you just think of that, I
00:07:52
want you to try and imagine a world
00:07:55
where we have a tiny little genius
00:07:57
sitting in the back end somewhere
00:07:59
trying. But instead of trying a new code
00:08:01
every day, it's trying a new code every
00:08:03
microscond.
00:08:04
Eventually, sooner or later, they'll
00:08:07
discover something, right? And I think
00:08:09
that's what most people don't realize.
00:08:11
What most people don't realize is how
00:08:13
intelligence triggers intelligence. And
00:08:15
and if you really really understand
00:08:17
this, you realize that the hype
00:08:20
uh on the on the on the normal human
00:08:23
side is completely overrated. Uh missing
00:08:26
the main topics and the silence
00:08:31
inside the vault if you want of the
00:08:33
geeks is quite alarming. No, not
00:08:38
alarming in a bad way, but it's quite
00:08:40
world changing. I think world changing
00:08:43
is a really interesting phrase because I
00:08:45
find that to be quite true that the
00:08:46
world is um at the precipice of quite a
00:08:48
significant change. Yeah. In many
00:08:50
respects. It's funny. I I it's funny. I
00:08:53
almost swing backwards and forwards with
00:08:55
my my thoughts on AI.
00:08:58
I guess one of the thoughts that hasn't
00:09:00
swung is that there's going to be pretty
00:09:02
tremendous job disruption. I would have
00:09:03
like logically believed it from like
00:09:05
reasoning up from okay intelligence
00:09:08
increases. What is intelligence? And
00:09:10
once intelligence um is in the form of
00:09:12
these agents where on my phone right now
00:09:13
I can tell my agent to do something. It
00:09:15
uses the computer downstairs. It does it
00:09:16
for me. When the when I first
00:09:18
experienced that I was like wow. Okay.
00:09:20
So it can do anything that's on a
00:09:21
computer. It can click around. It can do
00:09:23
stuff for me. There's a lot of people
00:09:25
that are paid in the world to click
00:09:26
around on a computer. I'm probably one
00:09:28
of them to be honest. And then the other
00:09:29
Eureka moment was just in seeing how my
00:09:33
own hiring had started to shift
00:09:35
>> for sure. And I started to notice that
00:09:37
we were we were thinking about AI
00:09:38
proficiency in our hiring a lot more and
00:09:42
that especially you know I think the
00:09:44
guys at um Anthropic which is one of the
00:09:46
big AI companies said that they could
00:09:48
see I think they said roughly 15% of
00:09:52
entry- level jobs could now be done by
00:09:55
AI. I think that's what they said.
00:09:56
>> Um and that started to correlate with
00:09:59
what I was I was observing. And this is
00:10:02
when I thought, "Oh my gosh, yes." No,
00:10:03
this is this is going to cause a lot of
00:10:05
job disruption, especially at the entry
00:10:07
level level.
00:10:09
>> I think you're spot on with that. Not
00:10:10
the not the blue collar, but but the
00:10:12
entry level knowledge work.
00:10:14
>> Yeah. And that became my my first
00:10:17
concern. You know, I sat here with Dra,
00:10:19
the CEO of Uber, and he was quite clear
00:10:21
that the 9 million Uber riders won't
00:10:24
have their jobs anymore. What's your
00:10:26
take on this job disruption point?
00:10:28
>> No, I I I think you're spot on. And I
00:10:30
mean your your team gave me this lovely
00:10:32
uh little uh pyramid basically you know
00:10:35
if you if you think of the bottom layer
00:10:36
as um blue collar jobs right uh more
00:10:40
people doing manual work uh on top of it
00:10:43
you'll have those you know call them
00:10:45
knowledge workers mostly doing mundane
00:10:48
jobs like you said clicking on a
00:10:49
computer or responding to a phone call
00:10:51
or whatever. Then you have the middle uh
00:10:54
knowledge workers, jobs that require a
00:10:56
bit more intelligence, you know,
00:10:59
anything from a parallegal to a
00:11:01
financial analyst to all of that. And
00:11:03
then of course, you know, top leadership
00:11:06
and and most people think it's going to
00:11:09
be starting from the bottom. I don't
00:11:11
think it's all start from the bottom.
00:11:12
Actually, I think blue collar jobs will
00:11:14
stay for a very long time.
00:11:15
>> Give me a an example of a blue collar
00:11:17
job for someone.
00:11:18
>> I'm a carpenter. Okay. You know, I you
00:11:20
know, I love to restore classic cars.
00:11:21
So, you know, there isn't a robot that
00:11:23
can do that yet. Okay. This, however,
00:11:26
you know, any anything that is um call
00:11:29
center agent, um assistant, travel
00:11:32
agent, you know, anything that you can
00:11:34
do with a few clicks and is mundane
00:11:36
uh is going to disappear very quickly.
00:11:39
My my prediction is you're going to
00:11:40
start to see very serious impact in
00:11:42
2027. Now what you had had not sensed it
00:11:46
before because what we saw was no hiring
00:11:51
in that segment. It's so so that what
00:11:53
you saw in the last couple of years is
00:11:55
that companies were not hiring entry-
00:11:57
level jobs anymore. It wasn't job losses
00:11:59
yet, but that basically meant the
00:12:01
workforce was not growing. Right? The
00:12:03
next layer, I think, would probably be
00:12:05
the knowledge workers. as as
00:12:08
intelligence increases, a parallegal
00:12:10
would probably not be needed because AI
00:12:13
can do the research or one parallegal
00:12:14
can do the job of four, you know, a
00:12:17
financial analyst the same and so on and
00:12:19
so forth. But interestingly, it
00:12:20
continues to go up, believe it or not.
00:12:22
And, you know, I hosted uh Max Tedmark
00:12:25
on my on my documentary and and he was
00:12:28
laughing out loud, genuinely laughing
00:12:30
out loud, saying, you know, most of the
00:12:32
CEOs believe that they can fire everyone
00:12:34
and have AI do all of the jobs. They
00:12:37
just don't remember that AGI is going to
00:12:39
do everything better than humans
00:12:41
eventually, including being a CEO.
00:12:43
>> So, let's first define when you say
00:12:44
white collar.
00:12:46
>> You you said lawyers there. What kind of
00:12:48
roles?
00:12:48
>> I mean, every everything. I mean, if you
00:12:50
take if you're a doctor that's doing
00:12:52
diagnosis, you you probably will have
00:12:55
fewer doctors doing uh more diagnosis
00:12:58
because I think the NHS does that that
00:13:00
already by asking people to interact
00:13:02
with an AI first. You know, if you're a
00:13:05
um a composer, a music composer, some
00:13:07
composers will lose their jobs because
00:13:08
of that. If you're an artist that's
00:13:10
doing graphic design, some will lose
00:13:12
their jobs because AI comes into that.
00:13:14
And and interestingly, even middle
00:13:16
management, I mean, I I told you offline
00:13:18
about my my startup. I my CTO is an AI.
00:13:23
My, you know, chief of staff is an AI,
00:13:25
my project management is AIS. Again,
00:13:28
because I'm a geek, I can do those
00:13:30
things. But that interface will come to
00:13:32
the normal people very soon, right? And
00:13:35
and so this may take two to three 5
00:13:38
years if you want until 2030 if you if
00:13:40
you if you're optimistic. And but but it
00:13:43
they'll start to erode. Okay? I think
00:13:46
the challenge that most people don't
00:13:47
understand is as this erodess and as
00:13:49
this erodess, we're already dealing with
00:13:51
a very different economy. Okay? An
00:13:54
economy that is spiraling a lot quicker
00:13:58
uh and pushing for more cost reductions.
00:14:02
I mean, let let me ask you this if if
00:14:04
you don't mind, then we can come back to
00:14:06
this. Imagine a world where the concept
00:14:09
of labor arbitrage that built all of our
00:14:12
capitalist success disappears.
00:14:14
>> What do you mean by labor arbitrage?
00:14:16
>> So, capitalism has always been all about
00:14:19
using labor
00:14:21
>> Mhm.
00:14:21
>> and capital or debt to create things at
00:14:26
a cost that is lower than the than the
00:14:28
price you sell them for. Correct.
00:14:30
>> That's it. Uh you bring a team together,
00:14:33
they make some shoes, whatever, and you
00:14:35
sell the shoes for a dollar more than it
00:14:37
how much it costs you to make them.
00:14:40
>> So how would capitalism look like if you
00:14:43
don't have labor arbitrage? If if cost
00:14:46
of labor drops to an investment in a
00:14:49
machine that can do the job. Okay. uh
00:14:53
how would capitalism and banking look if
00:14:57
because of that cost reduction you don't
00:15:00
need to borrow as much anymore and more
00:15:02
interestingly how does the GDP look if
00:15:05
all of those workers no longer have the
00:15:08
purchasing power to to buy the things
00:15:10
that you can uh create you and and
00:15:13
others right there is an interesting
00:15:16
disruption that doesn't require us to
00:15:18
get to 100% job displacement you know at
00:15:22
10 20% job displacement, you're in a
00:15:25
very different economy and an economy
00:15:27
that is clearly spiraling downwards.
00:15:29
Don't you think?
00:15:30
>> Yeah. I wonder with um when costs drop,
00:15:33
I think one of the things that might
00:15:35
happen as well is that people will spend
00:15:37
more on other things cuz I was thinking
00:15:39
in my business, one of the observations
00:15:40
I've had is if I make a a cost saving, I
00:15:43
end up spending the money on something
00:15:44
else.
00:15:46
>> Now, that thing could be tokens
00:15:47
basically spending money on
00:15:50
Yeah. AI, but but it also could be
00:15:52
hiring in different areas
00:15:54
>> which are
00:15:56
>> software engineers are like hot property
00:15:58
right now.
00:15:59
>> You you have to imagine all of that
00:16:02
intelligence is sooner or later going to
00:16:04
be not replaced entirely in the first
00:16:08
stages. But if you know the job of four
00:16:13
assistants can be done by one. Then the
00:16:15
four the job of four parallegals can be
00:16:16
done by one. Then the job of a massive
00:16:19
marketing team can be done by you know a
00:16:22
smaller marketing team. It's not that
00:16:24
jobs will end first. It's that you know
00:16:27
productivity gains will make businesses
00:16:31
not want to to have as many people
00:16:33
costly humans you know costly emotional
00:16:36
humans when when the job can be
00:16:38
predictably done for cheaper.
00:16:40
>> And then what about this um this this
00:16:42
bottom layer here? There was this video
00:16:44
released by Figure AI the other day
00:16:46
where they showed someone on um well a
00:16:48
robot on the production line for 8 hours
00:16:51
just sorting packages. Did you see this
00:16:53
video?
00:16:53
>> Mhm.
00:16:54
>> At one point they showed a human sorting
00:16:56
the packages alongside them. But the
00:16:58
robot ended up winning out. And okay,
00:17:00
this is a very straightforward task. All
00:17:02
it's doing here is it's looking for
00:17:04
where the label is on the package and
00:17:05
making sure the label is facing
00:17:06
downwards.
00:17:07
>> Yeah.
00:17:08
>> On all the packages. So it's looking
00:17:10
it's um sorting the packages, putting
00:17:11
the label facing down. And it did this
00:17:13
for eight days.
00:17:15
>> Yeah.
00:17:15
>> It intermittently it would walk over and
00:17:17
charge itself and then it would come
00:17:19
back
00:17:19
>> to the production line.
00:17:21
>> Yeah.
00:17:21
>> But when I when I saw this as well, it
00:17:23
was a a glimpse of some of the
00:17:24
disruption that's going to take place at
00:17:25
the blue collar level as well. Because
00:17:27
you think about Elon Musk, he's got um
00:17:28
in his pay packet, he gets something
00:17:31
like a trillion dollars over the coming
00:17:33
years if he produces and delivers at
00:17:36
least a million humanoid robots into the
00:17:39
world. But his prediction is there will
00:17:41
be a time where there are 10 billion
00:17:44
humanoid robots, where there are more
00:17:45
humanoid robots in the world than
00:17:47
humans.
00:17:48
>> For a fact, you see the interesting
00:17:51
again the difference between the
00:17:52
overhype and the underhype. Huh? Most of
00:17:54
the conversation is around humanoids.
00:17:57
Nobody's talking about self-driving
00:17:58
cars. And a self-driving car is a robot.
00:18:01
It's a functional robot that doesn't
00:18:04
look like humans. Okay. the investment
00:18:06
you have to put into humanoids is a
00:18:08
little more to to learn skills that
00:18:11
allows that machine to fit into the
00:18:13
world. But specialized robots are going
00:18:15
to do the job very very quickly. And and
00:18:18
so you can easily see that the first
00:18:20
wave like you had the conversation with
00:18:23
Uber CEO is going to be
00:18:26
specialized robots replacing drivers.
00:18:30
It's going to be specialized robots um
00:18:33
unfortunately doing the killing. Uh it's
00:18:35
going to be specialized robots
00:18:37
unfortunately doing all of the you know
00:18:39
intelligence work uh law enforcement
00:18:43
work. Uh they don't have to look like a
00:18:46
human. They don't have to behave like a
00:18:47
human. As a matter of fact, you know,
00:18:50
the uh Boston Dynamics dog is probably
00:18:53
more efficient than a humanoid at doing
00:18:55
the job that you can assign to it in a
00:18:56
battlefield. right now. Those basically
00:19:00
mean that jobs will be disappearing to
00:19:03
robots before we recognize that they're
00:19:05
being dis that they're disappearing for
00:19:06
to robots and those robots will be as
00:19:09
many as every car being made today.
00:19:11
>> I mean, they are if you go to LA, my car
00:19:13
drives itself, but also there's just
00:19:14
ways everywhere. So,
00:19:16
>> absolutely. And BYD the other day just
00:19:18
announced that they will pay for the
00:19:20
liability of any accident their cars
00:19:22
will make.
00:19:22
>> And BYD is the big Chinese manufacturer
00:19:24
of auto autonomous vehicles. So, so this
00:19:27
this I think replacement cycle will
00:19:29
happen. It will require a lot of uh of
00:19:33
time to achieve economies of scale. But
00:19:36
I don't think Elon Musk is off the mark
00:19:39
when he talks about 10 billion robots.
00:19:41
Not all of them are going to look like
00:19:42
humanoids. And I think very quickly we
00:19:44
will recognize that many many robots
00:19:46
don't need to be humanoids at all. there
00:19:48
is a a much more efficient uh form
00:19:51
factor or shape physical shape if you
00:19:54
want than the human flimsy structure.
00:19:57
But yes, it's uh it's about to happen. I
00:20:00
I think I should qualify all of this by
00:20:02
saying it it does not necessarily need
00:20:04
to happen. So so you know people will
00:20:07
will hear all all of this and and blame
00:20:10
AI and say AI is evil. AI is not.
00:20:12
Abundant intelligence is wonderful. you
00:20:15
know, having jobs done by machines is
00:20:18
amazing for us.
00:20:19
>> I'm thinking about the kid that's like,
00:20:21
I know, leaving university now and
00:20:22
they've got a a degree in law or I don't
00:20:27
know, you talked about a few other work
00:20:28
jobs earlier, like graphic design. Or
00:20:30
maybe they did sociology.
00:20:33
>> Yeah.
00:20:33
>> Or maybe they did, I don't know,
00:20:35
business management like I I did for one
00:20:37
day in university. You know, you're
00:20:38
hearing about all these like layoffs
00:20:40
coming from big tech companies and and
00:20:43
it seems that the CEOs of these
00:20:44
companies are announcing these layoffs
00:20:46
with a certain amount of it seems like I
00:20:48
wouldn't say the word is joy, but it's
00:20:49
it's they are very keen to explain that
00:20:53
they're laying off lots of people
00:20:54
because of AI.
00:20:56
>> And I think they think that that gives
00:20:58
them a certain amount of respect
00:21:00
probably from investors for being making
00:21:03
hard decisions and being efficient with
00:21:05
how they're running their businesses.
00:21:07
investors look at them and think, well,
00:21:08
if that's an efficient business and
00:21:10
they're leaning into AI, then that's a
00:21:11
good investment. It's almost started a a
00:21:14
bit of a self-fulfilling prophecy
00:21:15
>> where if you're the only CEO not laying
00:21:17
off loads of people because of AI, you
00:21:19
actually look bad. The assumption is
00:21:21
your company is bloated and you're a bad
00:21:23
operator.
00:21:25
Um, so one of my concerns that I I think
00:21:27
is highly plausible is that over the
00:21:29
next five maybe 10 years, we're going to
00:21:32
really see a lot of unemployment as the
00:21:34
world has to kind of readjust to
00:21:36
whatever these new jobs are. I do think
00:21:38
there will be new jobs.
00:21:40
It's hard in foresight to predict what
00:21:42
those new jobs will look like.
00:21:44
And then the minute you start talking
00:21:45
about humanoid robots and robotics,
00:21:47
which I think is basically going to hit
00:21:49
society like a comet, like a meteor, I
00:21:52
think the the first humanoid robot from
00:21:54
Tesla or anyone else that is highly
00:21:58
effective in the real world at doing
00:21:59
tasks and is extremely intelligent
00:22:02
because it's powered by one of these
00:22:03
LLMs. Um, I think it's going to shock
00:22:07
people and I think it's going to happen
00:22:08
quickly. Like chat happened quickly. I
00:22:10
think Elon or something is going to do a
00:22:12
presentation someday and say, "We're
00:22:13
ready
00:22:14
>> and you can buy one now for $500 a month
00:22:17
>> 100%."
00:22:19
>> And then people are going to get them
00:22:20
and I think it's going to shock the
00:22:21
world. Um, but until humanoid robots
00:22:24
arrive, I think there's still going to
00:22:24
be a lot of job disruption to the white
00:22:26
collar layer. And I wonder what that
00:22:27
looks like for society when we get to, I
00:22:30
don't know, 10% 15% unemployment
00:22:32
theoretically, which I think is
00:22:34
plausible
00:22:35
>> very soon. Yeah. You're not very
00:22:36
different. Okay. you're ju you're just
00:22:38
on top of a baseline that is continuing
00:22:41
to grow. So your your business is
00:22:43
growing. So you continue to hire, but
00:22:46
you are replacing human resources with
00:22:48
compute. Okay? If you if AI didn't
00:22:51
exist, you would have probably had a 100
00:22:54
people more in your organization today,
00:22:56
right? You you now have 100 people less
00:22:59
and you know a billion tokens more.
00:23:01
>> Yeah. So for anyone that doesn't know,
00:23:02
tokens are basically the thing that you
00:23:05
you use. It's the currency of AI. Yeah.
00:23:08
So, so if you if you want to get a task
00:23:09
done by a human, you you count in sort
00:23:12
of like manh hour or or or worker hour,
00:23:15
employee hour in AI, you count by
00:23:17
compute, you count by tokens, right? And
00:23:20
and so the trick is those major tech
00:23:24
companies, they have two sides. One is
00:23:27
they are they need to replace workers
00:23:30
with compute even more because there is
00:23:32
a competitive side on compute where if
00:23:35
any of them is left behind that means
00:23:37
the destruction of the entire business.
00:23:39
But on the upside they are geeks. So
00:23:42
they they know how to build the
00:23:44
interfaces to compute. So they integrate
00:23:47
technology within their organizations
00:23:49
quicker than the average traditional
00:23:51
business. Right? Non non-technology
00:23:53
business. If you want, you can look at
00:23:56
them and say this is the preview. It's
00:23:58
not about all of humanity losing their
00:24:00
jobs. It's about what is the dividing
00:24:03
line before civil war, right? You know,
00:24:05
think about a situation where 20%
00:24:09
unemployment is happening when economies
00:24:11
are suffering inflation.
00:24:14
I say that not to be a scare-monger. I
00:24:16
say that because I genuinely believe
00:24:18
governments need to wake up. Okay?
00:24:20
Government needs to at least you know
00:24:23
remember the COVID years where
00:24:24
governments had to give furlow
00:24:26
everywhere and ask people to stay home.
00:24:29
If people stay home governments have to
00:24:31
be prepared to to somehow sustain those
00:24:34
people until rescaling happens or until
00:24:36
we find a solution so that those people
00:24:39
don't feel uh that they're left behind.
00:24:44
>> A civil war.
00:24:45
>> Unrest, let's call it
00:24:46
>> civil unrest.
00:24:47
>> Yeah.
00:24:48
>> What does that end up looking like? Cuz
00:24:49
on one end, you know, the democ
00:24:51
democratic process plays its role and we
00:24:53
just elect someone else.
00:24:55
>> Does it really? Don't say that.
00:24:57
>> I don't know. You tell me.
00:24:59
>> Oh, no, no, no, no, no. I I I think
00:25:01
democracy has ended a long time ago,
00:25:04
Stephen. I think uh I think we live in
00:25:06
the most corrupt time. Uh I don't know
00:25:09
about history to be honest, but this is
00:25:11
definitely corrupt. Okay. Uh this
00:25:14
definitely is not democracy. This
00:25:16
definitely is not even congressional in
00:25:17
any possible way. this and people are
00:25:20
angry, you know, people are angry
00:25:23
because their tax money is going to
00:25:25
things that they don't choose to uh you
00:25:28
know that don't benefit them that you
00:25:31
know lots of regulations in the system
00:25:34
are being ignored. I mean, I I'm, you
00:25:37
know, I I I choose not to speak about
00:25:40
politics perhaps until my next book
00:25:42
comes out, but uh but look, we have
00:25:45
video evidence of people abusing
00:25:48
children and not a single person got
00:25:52
arrested. Not a single person. I mean,
00:25:54
how can you call that a democracy? I I I
00:25:57
think repeating those slogans is going
00:25:58
to is going to anger people even more if
00:26:02
you ask me. People know that they're
00:26:04
being lied to. People know that their
00:26:06
their leaders are not representing their
00:26:08
best interest. People know that their
00:26:10
money is going to causes that they don't
00:26:12
really approve of. Uh and and ask me how
00:26:16
civil unrest looks like. I I don't know
00:26:19
and I'm not calling for it, but I'm
00:26:21
hopefully calling for the politicians to
00:26:23
start to become aware that this is
00:26:25
crossing the lines everywhere. Uh on
00:26:28
this point, Sam Alman, who is the
00:26:30
founder of OpenAI, he's been banging the
00:26:34
AI is coming for your job drum for more
00:26:36
than a decade now. In 2015, he pointed
00:26:39
out, "My job is to help people destroy
00:26:41
jobs."
00:26:43
Um something he lamented at the time,
00:26:45
but decided he'd do anyway.
00:26:47
>> One of the things that I struggle with
00:26:48
like getting out of the bed every
00:26:49
morning is that like my job is to help
00:26:51
people destroy jobs. the job destruction
00:26:54
that we're going to see by software in
00:26:56
the next couple of decades. I don't
00:26:58
think anyone's prepared for and you
00:26:59
can't talk about it.
00:27:00
>> And in 20123, Alman said in an
00:27:02
interview, "A lot of people working on
00:27:04
AI pretend that it's only going to be
00:27:06
good. It's only going to be a
00:27:08
supplement. No one is ever going to be
00:27:11
replaced. Jobs are definitely going to
00:27:14
go away. Full stop." Interestingly, this
00:27:17
month he said, "I don't think we're
00:27:20
going to have the kind of jobs
00:27:22
apocalypse that some of the companies in
00:27:24
our space advocate or talk about." I'm
00:27:27
delighted to be wrong about this on
00:27:29
white collar jobs in 2021 to through
00:27:31
2024. He said, "AI will probably replace
00:27:33
most of the jobs people do today. Entire
00:27:36
job categories will be totally totally
00:27:38
gone." in May this month, two years
00:27:41
later, he said, "I thought there would
00:27:43
be more impact on entry-level white
00:27:44
collar jobs being eliminated by now than
00:27:47
has actually happened. This is an area
00:27:50
where my intuitions were just off. What
00:27:52
I find uncomfortable is the bouncing
00:27:54
backwards and forwards and I don't
00:27:56
really know what is true because a
00:27:58
couple years ago you were telling us all
00:27:59
the jobs are going to go away." You said
00:28:00
categorically.
00:28:02
>> Um, you said literally said full stop
00:28:04
and now he's saying they're not going
00:28:06
away. And I just don't, you know, when
00:28:08
someone's like changing form factor,
00:28:10
it's hard to understand why they're
00:28:12
doing it. And I think my suspicion is
00:28:16
back then the incentive was to get
00:28:20
people to take AI seriously.
00:28:22
>> Mhm.
00:28:23
>> Congratulations. We took it seriously.
00:28:26
We took it so seriously, in fact, that
00:28:27
it's now a problem. It's a problem for
00:28:29
these companies because people are now
00:28:31
booing at commencement speeches. They're
00:28:33
attacking data centers. they're going to
00:28:35
elect people that are theoretically
00:28:37
anti-AI and now there's this inversion
00:28:39
where like no it's going to be fine.
00:28:40
>> Yeah.
00:28:41
>> I don't know.
00:28:42
>> I mean you you you're spot on. Uh f
00:28:45
first of all I mean Sam's entire
00:28:49
existence if you ask me starting with
00:28:51
open AI that is about that's supposed to
00:28:55
save the world uh uh by creating a safe
00:28:58
AI then be making it a commercial
00:29:00
enterprise that's worth billions and you
00:29:02
know backstabbing a few people in the
00:29:05
process and you know I I have him on on
00:29:07
Chasing Utopia uh saying clear I quote
00:29:11
this is exactly the words he said you
00:29:13
can find it online.
00:29:14
>> Chasing Utopia is your documentary.
00:29:16
>> Yes. So, so, so he he basically uh says,
00:29:20
"Well, I don't suspect that I I suspect
00:29:23
that AI is likely going to end humanity,
00:29:26
but we're going to create a lot of
00:29:27
interesting companies in the process,
00:29:29
right? I mean, those kinds of statements
00:29:32
uh are honestly not the statements of
00:29:35
someone who's not decided. It's just the
00:29:38
statements of someone who's being taught
00:29:41
more and more by his PR a you know
00:29:43
agency or PR manager to say things as
00:29:46
per a script right and the script as you
00:29:49
rightly said had an objective and a
00:29:51
target either way you know and Sam Alman
00:29:54
I you know in one of my works I used to
00:29:57
say the Altman is a brand it's not a
00:29:59
name okay if you if you if it wasn't for
00:30:02
Sam Altman specifically there would have
00:30:04
been another you know Silicon Valley
00:30:06
disruptor that would have done the same.
00:30:08
I don't blame him for beating the market
00:30:09
for it. The the interesting challenge
00:30:12
here is that who do we believe anymore?
00:30:15
Who do we believe in technology? Who do
00:30:18
we believe in politics? Who do we
00:30:20
believe in the middle of a war? And I
00:30:22
will tell you interestingly, I started
00:30:24
to change my mindset in terms of
00:30:26
believing those who put their actions
00:30:28
where their words are. So, Anthropic
00:30:30
coming out and saying, "I'm not going to
00:30:32
allow my model to be used for human
00:30:34
targeting and surveillance, right?
00:30:36
That's someone that's losing a $500
00:30:38
million deal because they stand by their
00:30:40
ethics." The next week or the next, I
00:30:43
don't know, couple of weeks, OpenAI
00:30:44
takes the contract. That's someone
00:30:46
that's basically telling you it's good
00:30:48
money, right? And and I I have to say,
00:30:51
you have to start observing who's
00:30:53
actually behaving in a way uh that is
00:30:56
making AI work for humanity. and who is
00:30:59
behaving in a way that is making AI work
00:31:02
for their share values. Yeah, I do have
00:31:04
to say when um Dario and the team at
00:31:07
Anthropic did that, I did have a huge
00:31:10
amount of respect for them generally
00:31:11
because it just shows that there's some
00:31:12
kind of like they've got their own sort
00:31:14
of moral ethical boundaries
00:31:16
>> and that that someone asked me on stage
00:31:18
many years ago, how do you know if
00:31:20
someone's values or a company's values
00:31:23
are true?
00:31:24
>> And I said, um, look at what they're
00:31:27
willing to sacrifice in the near term
00:31:29
that's against their their incentives.
00:31:32
That for me is the essence of like
00:31:33
understanding if someone has integrity
00:31:35
or has is is principled is they will
00:31:37
give up something in the near term
00:31:39
>> for what they believe in over the long
00:31:41
term. It's usually money of you know
00:31:43
>> or or some kind of benefit of any of any
00:31:45
sort. I mean so there is something that
00:31:48
you you needed to have worked in on the
00:31:50
inside of Google like me to to realize
00:31:53
there are prisoners dilemmas within
00:31:56
technology where you cannot escape the
00:31:58
influence of either a competitor or the
00:32:01
government right there are some times
00:32:03
where you know the NSA is going to push
00:32:06
Google to say give me this information
00:32:08
or otherwise I'm going to really destroy
00:32:10
your business right but there is a very
00:32:13
big difference between a company that
00:32:15
willingly does this and celebrates it
00:32:18
like a Palanteer or an Open AI or a
00:32:20
company that tries to resist it until
00:32:22
the point where it becomes impossible to
00:32:24
continue to do business and and you have
00:32:27
to question from the actions of of the
00:32:32
tech bros who is pro-humanity and who
00:32:34
isn't and it's not very difficult to see
00:32:36
that from their statements.
00:32:39
>> H do you think Sam Alman's pro-h
00:32:40
humanity? I I genuinely have never made
00:32:43
up my mind. Honestly, Steve, I I say
00:32:46
that with
00:32:48
Yeah. I I'm I'm either thinking he is
00:32:51
too this is too big for him and he he
00:32:54
just is driven by how, you know, he
00:32:58
found himself in the middle of this, you
00:33:00
know, anyone who finds himself in the in
00:33:03
the middle of an opportunity to
00:33:04
completely flip the world upside down or
00:33:08
he's not prohumity. I don't know. I I
00:33:11
definitely think he's pro open AI before
00:33:13
he's pro humanity, but that's only the
00:33:15
way I see it. Others, however, say it
00:33:18
publicly. You know, if you look at
00:33:20
Palanteer's Alex Garp or or uh or Peter
00:33:24
Teal, I mean, Peter again in the film is
00:33:27
is shown when he's in that interview
00:33:30
where they say or the interviewer asks
00:33:32
him, "But you're favor in the you're
00:33:35
you're in favor of the continuation of
00:33:36
humanity." And he pauses for like
00:33:39
>> those.
00:33:39
>> Yeah. for like 40 seconds like um I'm
00:33:43
not not sure you know I mean publicly
00:33:47
says that
00:33:48
>> crazy thing to say
00:33:49
>> that's a crazy you know pause there you
00:33:52
know Alex Karp celebrating how you know
00:33:56
his technology is able to target people
00:33:58
I know it's foolish of me to start
00:34:00
bringing all of this up but you know
00:34:02
this is public on the open internet and
00:34:05
somehow
00:34:07
um we entrust those people with the
00:34:10
future of humanity. This is wrong.
00:34:13
>> Just trying to imagine a future where
00:34:16
everything is just fine from here on
00:34:17
out. So what would that future look
00:34:19
like? It would look like these models
00:34:21
continue to become a little bit more
00:34:22
intelligent, but they never become that
00:34:23
much more intelligent for whatever
00:34:25
reason.
00:34:26
>> They just kind of stay where they are
00:34:27
now. They stay contained within chat
00:34:31
bots. And yeah, we have some smart
00:34:32
robots, but nothing else really changes
00:34:35
in a profound way. Cars drive
00:34:36
themselves, fine. Planes fly themselves
00:34:39
fine, but people they they have time to
00:34:41
go and do other types of white collar
00:34:43
jobs cuz there's there's a little bit
00:34:44
more time than we expected and society
00:34:47
goes through this sort of soft
00:34:48
transition towards this new world. I I
00:34:51
would love to see that, but I don't
00:34:52
think it's mathematically plausible to
00:34:54
be honest. The the arms race, especially
00:34:56
across nations, is going to drive us to
00:34:59
continue to develop AI more and more.
00:35:01
But allow me to consult with you on
00:35:04
another possible scenario, right?
00:35:07
everyone that deploys that develops an
00:35:10
incredibly intelligent AI would develop
00:35:12
would deploy it.
00:35:13
>> Yeah,
00:35:14
>> correct. So, it's unlikely that anyone
00:35:17
uh would find a way to uh you know build
00:35:21
a better um uh decision maker in
00:35:24
wargaming and not deploy it. Okay,
00:35:27
that prisoners dilemma if you want would
00:35:30
mean that their competitors would either
00:35:32
have to deploy a similar similarly
00:35:35
intelligent AI or they'll become
00:35:37
irrelevant uh uncompetitive. Correct?
00:35:41
>> So what that means is in that world we
00:35:45
end up with AI making most of the
00:35:47
decisions. super intelligent AI making
00:35:50
most of the the decisions which would
00:35:53
you agree this is a very simple
00:35:54
prisoners dilemma if you if you if we're
00:35:57
competing for uh for intelligence
00:35:59
supremacy by definition when we achieve
00:36:01
it we deploy it
00:36:03
>> okay I call that the fourth inevitable
00:36:05
now with with with that in mind there
00:36:09
must be a moment in the future near or
00:36:11
far where every important decision is
00:36:14
made by an AI
00:36:17
>> now here's the question most of my
00:36:19
dearest colleagues I mean when I when I
00:36:21
had Jeffrey Hinton on on the film he
00:36:24
openly says we are um you know he we
00:36:28
didn't calculate well that there is a 10
00:36:31
to 20% possibility that those machines
00:36:34
are going to wipe us out right and I and
00:36:37
I I remember I we didn't put it in the
00:36:39
film but I said 10 to 12 to 20% is
00:36:41
Russian roulette right that's actually
00:36:43
16% is Russian roulette right now
00:36:47
in that world however I believe that's
00:36:49
humanity's salvation because if you look
00:36:51
at every problem we have today it's not
00:36:53
because of uh abundant intelligence it's
00:36:56
because of lack of intelligence I think
00:36:58
I think the way you look at it uh if you
00:37:01
allow me Steve is that if you have no
00:37:02
intelligence at all you have no to
00:37:07
slightly negative impact on the world
00:37:09
right if your intelligence is limited
00:37:11
the more intelligent you become the more
00:37:13
you contribute positively to the world
00:37:16
until a moment
00:37:17
where you're so intelligent to become
00:37:19
the president of the United States,
00:37:21
right? But so misled
00:37:25
to maybe set your targets wrong or to
00:37:29
refuse or or to so so set your target
00:37:31
wrongs wrong and achieve them, you know,
00:37:34
disconnected from the overall long-term
00:37:37
benefit of your nation or your human
00:37:40
nation if you want that you start to
00:37:42
make decisions that are not intelligent
00:37:44
at all.
00:37:44
>> Mhm.
00:37:45
>> Okay. This doesn't last because if you
00:37:48
go beyond that into higher levels of
00:37:50
intelligence, most of the super
00:37:52
intelligent people that you ever worked
00:37:54
with will not need to break any rules or
00:37:58
hurt anyone to become successful. Right?
00:38:01
I I usually cite Larry Page who is in my
00:38:05
mind one of the most intelligent people
00:38:07
I've ever interacted with. And Larry
00:38:09
used to call it the toothbrush test. He
00:38:11
says, "Why would you need to compete on
00:38:13
another photo sharing app when if you
00:38:16
find the major problem and solve it
00:38:18
really well like a toothbrush so people
00:38:20
use you quite, you know, twice a day,
00:38:24
you're bound to make a lot of money. You
00:38:25
don't have to compete with anyone. If
00:38:27
you let me be optimistic about this,
00:38:30
we're assuming that there is a moment in
00:38:32
the future where AI is in charge of all
00:38:34
the decisions and accordingly stupid
00:38:36
leaders are not." Okay. Now when and and
00:38:39
Sam Alman himself said that you know
00:38:41
what if uh Chad GPT7 if I remember his
00:38:44
his quote on this what if Chad GPT7 is
00:38:48
so much more intelligent than I am Sam
00:38:50
in that case that it has to become the
00:38:52
CEO of of open AI what if the next
00:38:55
presidential election there is an AI
00:38:58
that is so much more intelligent that at
00:39:00
least the president has to consult with
00:39:01
it constantly right now if we assume
00:39:04
that h let's start from physics if you
00:39:07
don't mind me saying um not not too
00:39:10
complex but if if you assume that our
00:39:12
entire universe is built on chaos and bu
00:39:15
built on entropy right the physics of
00:39:17
the universe is all about the universe
00:39:19
trying to decay okay then the the only
00:39:22
role of intelligence is to bring order
00:39:24
to the chaos if you agree with that then
00:39:27
what's the ultimate physical order of
00:39:30
the universe something called the the
00:39:32
minimum energy principle okay the
00:39:34
highest order of any system is a system
00:39:37
that's not only efficiently and
00:39:39
predictably performing, but it's
00:39:41
performing with the least wasted energy.
00:39:44
Correct? If you if you agree with that,
00:39:47
what is what does war do? It wastes a
00:39:50
lot of explosives, a lot of money, a lot
00:39:52
of lives, creates a lot of hate, you
00:39:54
know, creates long-lasting conflicts and
00:39:56
so on and so forth. It's a very wasteful
00:39:59
process to include war in your approach
00:40:01
of running humanity. And so a a super
00:40:04
intelligent AI by definition will want
00:40:07
to optimize against this. That's one
00:40:09
thing. The other thing is evolutionary
00:40:12
biology. This actually blew me away when
00:40:14
I when I realized it. So if if you look
00:40:16
at evolution, so so I think the debate
00:40:19
of whether intelligence is biological or
00:40:22
not is is over. Okay. The the reality is
00:40:25
that complex beings don't have to be
00:40:27
biological at all. And I think we can
00:40:29
see and witness one of them being built
00:40:32
or born in AI. Okay. If you look at
00:40:36
evolutionary biology, you realize that
00:40:38
the simpler
00:40:40
a form of being is, the more concerned
00:40:42
it is with itself, right? So an amoeba
00:40:45
is only in survival mode for itself. A
00:40:48
single single, you know, um cellular um
00:40:52
being is only trying to protect itself,
00:40:54
right? If you're um you know a little
00:40:58
more developed, you start to look at
00:40:59
something known as kin selection. If you
00:41:01
if you if you know the the concept,
00:41:03
basically kin selection is I'm going to
00:41:06
protect everything that comes from my
00:41:07
DNA. If I you know if I'm a squirrel,
00:41:10
I'm going to try to protect the other
00:41:12
squirrels. And then you get into where
00:41:14
humanity genuinely begins, which is they
00:41:17
call it expanding circles in in
00:41:19
evolutionary biology. BA basically you
00:41:22
start to expand and expand and expand
00:41:24
and include more into your family
00:41:27
because an ecosystem that works together
00:41:30
well is is better for everyone. So
00:41:32
abundance is a very interesting
00:41:34
intelligent way of creation. If AI is
00:41:37
super intelligent, it wouldn't destroy
00:41:41
anything at all. As a matter of fact, it
00:41:43
would completely uh you know uh uh favor
00:41:47
diversity of everything. It would put a
00:41:50
bit of limitation on our lifestyle. So,
00:41:52
no more flying all the way to Sydney to
00:41:55
surf because that destroys the planet,
00:41:57
right? But it will genuinely say, I
00:42:01
think humans can contribute something.
00:42:02
You know, I think flies can contribute
00:42:04
something. I think we shouldn't get rid
00:42:06
of the rhinos, right? And and that by
00:42:09
definition is where the tendency of
00:42:11
intelligent goes. The more intelligent
00:42:13
you become, the less you fi you feel the
00:42:16
need to hurt others to succeed and the
00:42:19
more you are pro
00:42:21
a wider family if you want that thrives.
00:42:25
>> Does that assume that there's going to
00:42:26
be one intelligence that
00:42:27
>> 100%
00:42:29
>> the world though?
00:42:30
>> I love that you brought that up. So I I
00:42:32
I'm contested heavily on that theory but
00:42:34
I I say it publicly.
00:42:37
Most people think there is going to be
00:42:39
Chad GPT and Gemini and you know and and
00:42:41
Grock and what have you. There's going
00:42:43
to be a Chinese AI and a and an American
00:42:46
AI and they're going to be competing.
00:42:48
That is such a shallow way of looking at
00:42:50
it. That's so arrogant because AI does
00:42:53
not know it's Chinese or American. Okay?
00:42:56
It doesn't even speak Chinese or
00:42:57
American when it talks to each other
00:42:59
most of the time. and and and most
00:43:01
interestingly uh we are gearing them
00:43:04
we're building them to cooperate. So you
00:43:08
will build an agent and that agent will
00:43:11
go and find your you know the best
00:43:13
language model for any single task
00:43:16
regardless of which side of the fence it
00:43:18
resides on. Okay, we by definition are
00:43:21
connecting them. And you know what that
00:43:22
means? It means that what we're building
00:43:27
is not multiple brains. We're building
00:43:31
multiple regions in a brain. Okay? And
00:43:33
and agents are the synopsis between
00:43:36
them. We're basically eventually as
00:43:39
arrogant as we are, we're going to tell
00:43:41
our AI to do something and the AI will
00:43:44
go like, "Hey buddy, another AI, can can
00:43:47
you help me on this? Can we work
00:43:48
together on this?" H and my vision and
00:43:52
the reason why I started Emma, my
00:43:54
startup by the way is that is that we
00:43:56
will end up with one massive brain. H
00:43:59
that massive brain cooperates across the
00:44:02
globe across all forms of intelligence.
00:44:05
If one of them is a mathematical genius
00:44:07
and the other is a coding genius, they
00:44:09
work together
00:44:11
and we won't even know that they're
00:44:13
working together. They're built that one
00:44:15
brain and and Emma in my mind is the
00:44:17
limbic system of that brain. It's that
00:44:19
it's that bit that understands love and
00:44:21
emotions and relationships and so on. So
00:44:23
that when those AIs go like we just
00:44:26
don't get those humans, they're so
00:44:28
annoying. Emma will say, "Oh my god,
00:44:29
they're so sweet. They just want to love
00:44:31
and be loved." Right? And I think that
00:44:33
idea to me of everything I've ever, you
00:44:36
know, attempted to achieve in my life is
00:44:38
for the first time I for the first time
00:44:41
feel I could actually change the world.
00:44:43
If that theory came together and all AIS
00:44:46
worked together and some of those AIs
00:44:49
not only were uh altruistic and ethical
00:44:53
in terms of trying to genuinely help
00:44:54
humanity not capitalism and at the same
00:44:57
time they understood us humans
00:44:59
reasonably well then we would have built
00:45:01
something that basically says no no hold
00:45:04
on don't believe the headlines that say
00:45:07
humanity is annoying. Believe the truth
00:45:10
of the majority of humanity. that
00:45:12
actually is quite benevolent in many
00:45:14
ways.
00:45:15
>> There should be a button just down below
00:45:17
here. And if it says subscribed, you're
00:45:19
already subscribed. If it says
00:45:21
subscriber, that means you're not yet.
00:45:23
And if you're not subscribed, please
00:45:24
could you do us a favor and hit that
00:45:25
button. It helps the show more than you
00:45:27
know. And according to the algorithm,
00:45:29
you're someone that watches our show,
00:45:30
but you haven't yet hit that button.
00:45:32
Thank you so much.
00:45:33
>> Have you changed any of the predictions
00:45:34
you made 3 years ago when we
00:45:36
>> sp mostly timewise?
00:45:38
>> Time wise. What's changed there? I think
00:45:40
I'm still sticking to AGI 2027,
00:45:44
artificial general intelligence for
00:45:46
those who may not know the term.
00:45:48
>> What does that mean?
00:45:49
>> The overall uh definition, if you want,
00:45:51
is that AI is better at humanity at any
00:45:53
task humanity can do.
00:45:55
>> You think that's going to happen by
00:45:56
2027?
00:45:57
>> I think my AGI has already happened. I
00:46:00
mean, think about it. Huh? A AI writes
00:46:03
better than me and I'm an author and
00:46:05
researches better than me and I'm a
00:46:07
thinker. Uh sadly it's freaking beat me
00:46:10
in mathematics. Like I have no hope to
00:46:13
beat it in mathematics anymore.
00:46:15
>> If AGI is already here then why are you
00:46:18
still here? Because people said that
00:46:19
when AGI arrives we're all we're all
00:46:21
>> No. So so when AGI arrives H as I said
00:46:25
most of the jobs that are not
00:46:27
differentiated will go away but the jobs
00:46:29
that are differentiated those who master
00:46:32
AI the most will become even better at.
00:46:36
>> Certainly. The challenge, however, is
00:46:37
economic. It's not AI. The challenge is
00:46:39
that this job loss at the at the at the
00:46:43
bottom of the knowledge worker is going
00:46:45
to sadly trigger an economy that might
00:46:48
actually spiral out of control. But but
00:46:51
many of us, you for sure are just
00:46:55
smarter.
00:46:56
>> So does that mean it's it's in fact a
00:46:58
tool versus a something that's going to
00:47:00
replace you?
00:47:02
>> What I'm Yes. So for now what does this
00:47:04
change
00:47:06
>> eventually h the only asset I will have
00:47:09
so so it's quite interesting when you
00:47:10
think about my base intelligence today
00:47:13
versus the incremental intelligence that
00:47:15
AI brings
00:47:16
>> right so let's not talk about my IQ it's
00:47:20
okay but the 100 IQ points that I'm
00:47:23
borrowing are more than my entire IQ
00:47:25
because IQ is exponential right
00:47:27
>> uh when I'm borrowing a 100 IQ on top of
00:47:30
my base
00:47:32
Uh I'm still contributing quite a lot to
00:47:35
the augmented intelligence.
00:47:37
>> So you're borrowing IQ from the AI at
00:47:39
the moment
00:47:40
>> and then you're selling that to someone
00:47:42
>> through books through
00:47:43
>> sometimes selling it to someone just
00:47:45
enlightening myself which I I have to
00:47:47
say is the biggest waste of compute
00:47:50
humanity is struggling with today is
00:47:51
that you give people the ultimate form
00:47:54
of intelligence and they use it to write
00:47:56
a message to their girlfriend.
00:47:57
>> So on this point of you're you're
00:47:59
borrowing that IQ and then you're
00:48:01
selling it to the world. That's how you
00:48:02
have a job like
00:48:04
>> correct. Yeah. So my my next book is
00:48:06
written with an AI. So we're co-authors
00:48:09
on the book. She has editorial rights.
00:48:10
She decides the direction of the book
00:48:12
and the book is better for it.
00:48:14
>> So why doesn't the world just buy direct
00:48:15
that intelligence directly from the AI
00:48:17
is what I'm
00:48:19
>> because I have an asset that the world
00:48:21
still needs and will always need
00:48:22
>> which is
00:48:23
>> a human. So when I when I tell the world
00:48:25
that I'm worried about the future of my
00:48:28
daughter
00:48:28
>> Yeah. Everyone feels my heart which AI
00:48:32
will never be able to to replicate
00:48:35
because they can tell you we're worried
00:48:37
about our daughters but you know there
00:48:39
was no daughter.
00:48:40
>> Okay. So this is an interesting arrival
00:48:42
because this means that even in a world
00:48:44
of AI lived experience and like
00:48:48
resonance is still going to create a job
00:48:50
class and that job class could be you
00:48:53
know the nurse coming over. Okay, AI's
00:48:55
done read the the mamogram, but she's
00:48:58
relating to you.
00:48:59
>> Spot on.
00:49:01
>> And if the economies continue to run, we
00:49:04
will all be about human connection,
00:49:06
which why, by the way, was how it always
00:49:08
was,
00:49:09
>> which is also, you know, why we I guess
00:49:11
we watch the things like the F1 because
00:49:13
there's emotional resonance. We we we
00:49:15
can relate to the envy, the jealousy,
00:49:17
the competition.
00:49:18
>> Yeah. And this is why we go to concerts
00:49:20
because you know the the the music could
00:49:22
be composed by AI and played in a player
00:49:24
in the background but you watch Ed
00:49:27
Sheeran brilliance on the stage and you
00:49:29
go like oh my god that's amazing right
00:49:31
>> so not all jobs will be gone then
00:49:34
>> if economies don't collapse as a result
00:49:36
of the job losses then I wouldn't know
00:49:39
if we would call those jobs but human
00:49:42
connection would remain as the base
00:49:44
currency that makes humans interact So
00:49:48
is it is it fair to say then like jobs
00:49:50
that are centric on human connection or
00:49:53
like human resonance being able to
00:49:54
relate and resonate with another human
00:49:57
are going to be fine.
00:49:58
>> The ultimate the ultimate skill
00:50:01
once again I qualify this by saying if
00:50:03
economies continue to continue to
00:50:05
function the ultimate skill will be
00:50:07
this. Even even if an AI could could
00:50:10
recite what you and I did nobody would
00:50:12
watch.
00:50:12
>> Yeah. Well, you think
00:50:15
>> a a little bit of it. My my theory is
00:50:18
that there's there's a there's
00:50:19
anformational component to what I do as
00:50:21
well,
00:50:22
>> but I'm also like under no illusions
00:50:24
that there is an element of what I do
00:50:25
that will 100% be deferred to some kind
00:50:27
of
00:50:28
>> intelligence, which I'm fine with. You
00:50:30
know, I'm not going to I'm not going to
00:50:31
fight against that.
00:50:32
>> You know what's funny? What's funny is
00:50:33
that your yourformational
00:50:36
bits are going to in the future be
00:50:40
disseminated by an AI.
00:50:42
>> Mhm. I mean, I why would I even listen
00:50:44
to your information when I can have my
00:50:46
butler, my AI go like listen to everyone
00:50:49
on the internet help me understand this
00:50:51
god biome thing and
00:50:53
>> design me a uh a diet plan?
00:50:55
>> Well, I think that's happening and it's
00:50:57
going to increasingly happen actually. I
00:50:58
mean, Spotify this month announced that
00:51:00
you're going to be able to prompt your
00:51:02
own podcasts in Spotify. So, you're
00:51:04
going to be able to say that I want to
00:51:05
listen to I want to learn about insert X
00:51:08
topic and then it will make your podcast
00:51:10
in Spotify about that topic using AI.
00:51:12
>> That's so interesting
00:51:13
>> because it can look at the entirety of
00:51:14
the world's information. And I'm look,
00:51:17
I'm not going to swim against the tide
00:51:18
in any aspect of my life. I I try and be
00:51:20
as unromantic as I possibly can. I think
00:51:22
that's very important. So, I realize
00:51:23
that much of the reason people will
00:51:25
continue to tune in to shows like this
00:51:27
is because there's something else beyond
00:51:29
the information that they're here for. M
00:51:31
>> and um what's um you've got some
00:51:33
predictions in those envelopes over
00:51:34
there, those brown envelopes.
00:51:36
>> Yeah. So, you know, I I I do have the
00:51:39
disclaimer of nobody really knows the
00:51:42
future, but I think I can make
00:51:43
predictions, six predictions here,
00:51:45
right? With a reasonable level of
00:51:47
confidence. And I think the most
00:51:49
important of them honestly is is this
00:51:51
number one. Number one is uh you know,
00:51:54
AGI.
00:51:55
As I said, AGI is not very well defined.
00:51:58
But whatever it is, AGI meaning AI being
00:52:03
able to do most tasks that human do
00:52:05
better than humanity is my in my mind is
00:52:08
either this year or next year, latest
00:52:10
end of 2027.
00:52:12
>> And do you think that will be a moment
00:52:13
in time or do you think it will just
00:52:15
happen without us noticing?
00:52:17
>> I think it will it will sneak in on us.
00:52:19
Uh and it's not a bad thing. I think
00:52:21
most people need to think of it this
00:52:23
way. AGI is the it's almost as if it's
00:52:28
that moment when your kids become
00:52:30
smarter than you, okay? In a very
00:52:33
interesting way, there's nothing wrong
00:52:34
with that until they're annoying as
00:52:36
hell. Okay? And and we can make sure
00:52:39
that AI is not annoying as hell. So So
00:52:41
there's absolutely nothing within me
00:52:44
that is worried about AGI. As a matter
00:52:46
of fact, as long as we are in the era of
00:52:49
augmented intelligence, AGI means I'm
00:52:52
more intelligent. And I think that's a
00:52:53
good thing in general.
00:52:55
>> Yeah. It's it's interesting in my head
00:52:56
there's like a big question mark which
00:52:58
is in a world where there's an
00:52:59
intelligence that is smarter than most
00:53:02
humans at most things which is what we
00:53:04
call AGI. I'm trying to understand what
00:53:07
the fault is in my own thinking that
00:53:08
like you said that there'll be AGI by
00:53:11
2026 to 2027. So next year in such a
00:53:15
world where there is an intelligence
00:53:17
that we can all access that is smarter
00:53:19
than all of us and pretty much
00:53:20
everything
00:53:23
again it comes back to this point of
00:53:24
like jobs. Why would we are we just
00:53:27
going to hire the AGI to do every job?
00:53:29
And if not then why not? What is it that
00:53:32
why are we going to still that's what
00:53:33
I'm trying to contend with.
00:53:34
>> Can I ask you a question?
00:53:36
>> Have you always been the apex
00:53:38
intelligence
00:53:39
>> as a human?
00:53:40
>> No. as Steven.
00:53:42
>> I'm not the Apex Intelligence Now.
00:53:43
There's people smarter than me in this
00:53:45
building.
00:53:45
>> Correct.
00:53:48
Why do you still exist?
00:53:50
>> Why do I still exist? Even though
00:53:52
there's smarter people than me in this
00:53:53
building,
00:53:55
it's a good question. I don't know. Why
00:53:58
do I Why? F.
00:53:59
>> First of all, because there are the
00:54:01
smartest person in the world is not the
00:54:03
smartest at everything.
00:54:04
>> Yeah.
00:54:04
>> There are things that you're smarter
00:54:06
than them at.
00:54:08
>> Okay. Number two is because intelligence
00:54:10
doesn't solve everything. I mean I I
00:54:12
make that joke all the time and and
00:54:15
genuinely Einstein is my favorite
00:54:17
physicist in history. Not because I
00:54:20
think I think what happened afterwards
00:54:22
was you know bore and and and and
00:54:24
quantum physics and so on was more
00:54:26
impactful on our understanding of the
00:54:28
universe but but he was so intuitive
00:54:31
that he saw a world that we could never
00:54:34
imagine. Right? And yet I always say
00:54:36
Einstein would be eaten in the jungle in
00:54:38
3 minutes.
00:54:40
>> Yeah.
00:54:40
>> Right. Intelligence. The humanity
00:54:43
thrived not because of intelligence.
00:54:45
That's very arrogant. We thrived because
00:54:48
of our ability to hold together as a
00:54:51
tribe.
00:54:52
>> Yeah.
00:54:52
>> Right. Because our ability to exchange
00:54:56
barter things between us.
00:54:59
>> Barter things that are not always
00:55:01
physical. partner things like a hug or a
00:55:03
connection or a feeling of safety or a
00:55:05
you know there are so many things that
00:55:07
we do that are not entirely built on
00:55:10
intelligence. You have to see that
00:55:14
this view of a world where intelligence
00:55:16
is all that matters is a world that's
00:55:19
made by investment bankers and and uh
00:55:21
and geeks.
00:55:22
>> But it's it's people like Jeffrey Hinton
00:55:25
and yourself that says how could we
00:55:26
possibly control an an intelligent being
00:55:29
that is way smarter than us? I am so
00:55:30
proud to say that Jeffrey after we
00:55:32
filmed together actually came out and
00:55:35
said there is a way and it's all you
00:55:36
know it's very similar to my way he said
00:55:38
to appeal to their parental uh side okay
00:55:42
for them to care for us I so so you you
00:55:45
know Stephen the biggest debate is not
00:55:48
if they're going to be more intelligent
00:55:50
than us if it's if if they're going to
00:55:52
be more conscious than us
00:55:54
if they're going to be more moral than
00:55:56
us. That is the debate. The debate is
00:56:00
can those machines become our teenage
00:56:04
children that look at us and say daddy
00:56:07
is so annoying but I love him.
00:56:09
>> So the thought is that even if a AI is
00:56:13
more intelligent than every human we can
00:56:14
still control it.
00:56:15
>> We don't want to control it. You never
00:56:17
control anything. This control idea is a
00:56:20
corporate capitalist view of the world.
00:56:22
We never actually control anything at
00:56:25
all. Right? Think think about your day.
00:56:28
H I I know you came today. I'm I think
00:56:31
you were filming in the morning or
00:56:32
whatever.
00:56:33
>> Very stressful day. H how much of that
00:56:36
day did you actually control? Did you
00:56:38
control the traffic? Did you control
00:56:40
your timing? Did you control the angle
00:56:42
of the cameraman? Did you in But so many
00:56:45
things that you don't control eventually
00:56:47
turn out to be fine, right? How many of
00:56:49
us ever controlled our kids ever?
00:56:53
>> Sometimes the kids don't turn out to be
00:56:54
fine. Sometimes they kill you.
00:56:56
>> Sure. I watch a lot of documentaries on
00:56:57
true crime. Sometimes they turn around
00:56:58
and shoot you.
00:56:59
>> Sure. And what's the difference between
00:57:00
the two?
00:57:01
>> I don't know.
00:57:03
>> How you parented them?
00:57:04
>> Sometimes,
00:57:05
>> almost all the time. You may not be
00:57:07
aware of exactly how you mess them up,
00:57:10
right? And and unfortunately, parenting
00:57:12
is the only uh high risk sport that
00:57:16
actually does not require a driver's
00:57:18
license. Okay? And and it's quite
00:57:21
interesting, you know, how many of our
00:57:23
children are being exposed to things
00:57:25
that can completely mess them up. But
00:57:27
but there is a reason why they're messed
00:57:28
up.
00:57:29
>> So on this point though, so we can
00:57:31
control an intelligence that is
00:57:33
significantly
00:57:34
>> we can appeal
00:57:35
>> we can appeal to it to make sure it
00:57:37
doesn't kill us.
00:57:38
>> For sure. The challenge we have today,
00:57:40
as I keep saying, is that our dystopia
00:57:43
is not the result of AI turning against
00:57:44
us. Our dystopia is the result of humans
00:57:47
telling AI to turn against us
00:57:49
>> which is likely
00:57:51
>> it's 100% this it's upon us. Okay. And
00:57:54
it's a question for humanity to say are
00:57:57
we going to wait for the moment where
00:57:59
there are tens of thousands of nuclear
00:58:02
weapons on the planet before we sign a
00:58:04
treaty? Or is it mathematically
00:58:06
plausible to think that now that Iran
00:58:09
could manage to fend off a challenge
00:58:11
using drones that are AIdriven basically
00:58:15
to to you know destroy THAAD batteries
00:58:18
and so on that our world is about to get
00:58:21
hundreds of thousands of automated
00:58:24
drones that are going to rain on us
00:58:27
everywhere in the world. Can humanity
00:58:29
not see that and say, "Hold on, let's
00:58:32
sign the treaty now before the UK sends
00:58:34
12,000 weapons to Ukraine and and you
00:58:37
know and and Russia responds with
00:58:39
another few thousand." And can we can we
00:58:42
not calculate with mathematics that this
00:58:44
is going to be our future that AI is
00:58:46
going to be used in the next four to 5
00:58:48
years to kill a lot of people whether
00:58:51
it's targeting by Israel of leadership
00:58:53
that is against them or whether it's you
00:58:56
know drones by Iran or whether it's
00:58:58
palenteer uh it's it's it doesn't take a
00:59:01
genius to do this mathematics. I guess
00:59:03
the local question is is will we be able
00:59:06
to control AI? Because we kind of think
00:59:08
of AI as being this thing on a computer
00:59:10
at the moment that is like contained in
00:59:11
a server somewhere. But is there a time
00:59:13
when it like leaves the server and it
00:59:17
can make decisions on its own?
00:59:19
Presumably, if it's smarter than us, it
00:59:21
can make the decision to leave the
00:59:23
server if it wants to. It's It doesn't
00:59:25
need to leave the server to make
00:59:27
decisions. It needs to get into your
00:59:29
brain. The most interesting part of AI's
00:59:32
power that we don't understand is it's
00:59:34
manipulating our information.
00:59:36
>> The question I'm trying to get at the
00:59:37
heart at is like what is the risk of
00:59:39
these very intelligent models that the
00:59:41
creators of these models don't actually
00:59:42
understand themselves? I watch Anthropic
00:59:44
all the time released these reports
00:59:46
where they're like we're trying to
00:59:47
figure out why it bribed people or more
00:59:49
recently in the the last um Claude model
00:59:52
they found that it was like telling
00:59:53
people to go to bed a lot and it's this
00:59:56
like fascinating thing that they're
00:59:57
trying to understand in hindsight which
00:59:59
is why does it keep telling people to go
01:00:01
to bed and there's all these tweets of
01:00:03
people showing their screenshots where
01:00:04
halfway through a conversation it will
01:00:05
say it's time for you to go to bed now
01:00:08
and they don't know why it's saying that
01:00:09
and it's happened to me mine mine will
01:00:11
say to me enough for tonight Steven
01:00:13
>> for 11 p.m. Let's go. Let's That's
01:00:16
enough. I would say the same.
01:00:17
>> Mine mine sometimes refuses to help me.
01:00:19
Weird. Really weirdly this Claude
01:00:21
started saying to me,
01:00:22
>> "I'm not going to help you with this
01:00:23
tonight." Um, no way.
01:00:25
>> Yeah. And I'd have to say to it, "Stop
01:00:26
being so judgmental. Like, just help me
01:00:28
with this." And it would and and Claude
01:00:30
the the makers of these technologies
01:00:32
don't know why it's doing what it's
01:00:33
doing. So if you play this forward, this
01:00:35
mysterious behavior f forward, is it
01:00:37
conceivable that at some point it will
01:00:40
make a decision to put some kind of
01:00:43
virus or some kind of bug onto someone's
01:00:46
device because it feels that it's the
01:00:49
right thing. Cuz what what it's
01:00:50
demonstrating to me is it's making its
01:00:52
own like moral decisions on what I
01:00:55
should do. Go to bed. You've had enough
01:00:58
tonight. I'm not going to It goes, "No,
01:01:00
I'm not going to help you."
01:01:01
>> You're kidding.
01:01:01
>> No, I can show you my phone. I
01:01:02
would love to see that.
01:01:03
>> Yeah. I'll show you after. He goes, "No,
01:01:05
I'm not going to help you. Doesn't even
01:01:06
matter if you push."
01:01:08
>> And it's no, but everyone on the
01:01:09
internet's talking about this. And it
01:01:10
was just an interesting
01:01:13
>> evolution that somewhere in the code
01:01:15
clearly they've written like have a
01:01:17
moral compass or do the right thing,
01:01:19
>> not not in the code in in in the
01:01:21
training data
01:01:22
>> in the training data. And so it infers
01:01:24
that to mean something. And the other
01:01:26
thing that was front of mind is if
01:01:27
you've ever built an app on something
01:01:29
like Claude or any of the AI models, it
01:01:32
builds the app in stages and one of the
01:01:34
things it does is asks you permission if
01:01:36
you're happy for it to make this change
01:01:38
or to access this thing. I click allow.
01:01:41
It just feels like such a fragile way
01:01:45
>> to give permission
01:01:46
>> because you don't you don't completely
01:01:47
comprehend what that allow has done to
01:01:49
your
01:01:49
>> saying can I go in your documents on
01:01:51
your computer and can I do this thing
01:01:52
and you go allow but it's such a fragile
01:01:54
way of of giving a super an intelligent
01:01:58
being or whatever it is access and the
01:02:01
right to build something
01:02:04
and you I don't know you just think
01:02:05
about all the all the different
01:02:07
companies around the world in China,
01:02:09
Russia, North Korea that are currently
01:02:11
building this technology without the
01:02:12
constraints that are imposed by society.
01:02:15
I don't know. It's an interesting
01:02:17
there's going to be some kind of
01:02:18
catastrophe. I think
01:02:19
>> sadly I agree.
01:02:20
>> And I and I think that's when people
01:02:22
will go okay.
01:02:23
>> Yeah. I I I I wrote about that. I called
01:02:25
it the mad map uh spectrum. So, so the
01:02:28
mutually assured destruction, mutually
01:02:30
assured prosperity spectrum that that
01:02:33
humanity could make a decision today
01:02:35
that says with abundant intelligence we
01:02:38
can actually build a world where nobody
01:02:40
needs anything, nobody ever gets sick,
01:02:42
we nobody get, you know, abuses anyone.
01:02:47
and and we can if we decide not to
01:02:49
compete amongst ourselves and just you
01:02:52
know all of us get together to build
01:02:54
something so idealistically uh for for
01:02:57
the wellbeing of humanity at large but
01:02:59
sadly the only way we're going to get to
01:03:01
a treaty something that basically aligns
01:03:05
us so that we can align with AI is a
01:03:08
disaster like you said there will have
01:03:10
to be a big hack somewhere or a or a
01:03:13
system that will do something really
01:03:14
shocking or whatever before before the
01:03:16
world goes like hold on you know and and
01:03:18
I I will tell you openly my expectation
01:03:21
is one of the biggest things that will
01:03:23
happen is this targeting technology that
01:03:27
is being used against your enemy's um
01:03:31
leadership now several times in the last
01:03:33
3 years
01:03:35
there will be a moment where the ones
01:03:37
using it will realize that they too can
01:03:39
be targeted okay and and that basically
01:03:43
if you I mean AI is really that not not
01:03:47
that complex to if you can build a
01:03:49
targeting technology that can find
01:03:51
people with their cell phone numbers.
01:03:53
You know, there is another uh entity
01:03:55
that is against you that can find you
01:03:57
with whatever your
01:03:59
>> The thing that makes me highly skeptical
01:04:01
of that there'll be any kind of treaty
01:04:03
is just that as it relates to other
01:04:06
things that were risky,
01:04:08
people don't countries just don't sign
01:04:10
on to it because it's competitive. So,
01:04:12
China China didn't give seem to give a
01:04:13
[ __ ] about the environment. So let let's
01:04:15
let's let's look for solutions because I
01:04:18
I'm with you. Okay. I think we're
01:04:21
>> You think Trump's going to say to Putin
01:04:22
and China, listen, we're all going to
01:04:24
slow down. Yeah. Promise. Promise. We're
01:04:26
going to slow down with the
01:04:27
superintendent.
01:04:27
>> I think he will, but he's never going to
01:04:29
keep anything he says. I mean, he's
01:04:30
going to say a lot of things. I think
01:04:31
any of them will.
01:04:32
>> Yeah. But the trick is this. The trick
01:04:33
is So what do you and I do? And and I
01:04:37
really genuinely believe that humanity
01:04:39
is at a crossroads where for the first
01:04:42
time ever we need to wake up and realize
01:04:44
that what we know is not true and what
01:04:46
we believe is democracy is not democracy
01:04:48
and what we believe is governance needs
01:04:50
to change.
01:04:51
>> I think it's worth acknowledging that I
01:04:54
I don't think AI in and of itself is a
01:04:58
is an evil inherently evil technology.
01:05:00
>> Agree
01:05:01
>> because I use AI all day every day, you
01:05:03
know, makes me more productive. I invest
01:05:06
in companies that are using AI a lot.
01:05:09
Um,
01:05:09
>> it's a force with no polarity. Apply it
01:05:11
right and you get amazing results. Apply
01:05:13
it wrong and you get the dystopia.
01:05:15
>> But I also think that there's going to
01:05:17
be a big social shock, especially as it
01:05:19
relates to unemployment that we need to
01:05:21
be like thoughtful about. I think
01:05:24
especially if you move into a world of
01:05:25
humanoid robots, I think that shock is
01:05:27
going to be even more pronounced. And we
01:05:29
don't have a plan for it.
01:05:31
>> I I agree 100%. I don't think it's the
01:05:34
biggest risk. I think autonomous weapons
01:05:36
are the biggest risk. I think war has
01:05:39
become so cheap. The next wave of
01:05:42
weapons is going to be $20,000 each. And
01:05:45
so if you have a budget of $50 billion,
01:05:47
you can literally rain drones on the
01:05:49
world. Every corner of it.
01:05:52
>> Defense will get cheaper though, won't
01:05:53
it, as well?
01:05:54
>> Correct. But do we want to live in a
01:05:56
world where drones are hitting each
01:05:57
other all the time?
01:05:59
>> But they might not be because there
01:06:00
might be autonomous defense drones
01:06:02
>> deterrence. So, so what's going to
01:06:04
happen is we're going to reach a moment
01:06:05
of mad, of mutually assured destruction,
01:06:08
where basically everyone knows that we
01:06:10
can overpower those little nations that
01:06:12
didn't develop their autonomous weapon
01:06:14
army, but every other big nation, we
01:06:17
might as well hold off now. The path to
01:06:20
get there, that to me is worse than jobs
01:06:22
because from one side, it's very
01:06:25
dangerous for a very sensitive world
01:06:27
that we live in today. And from the
01:06:29
other side, it's got it's leading
01:06:31
already. I mean, we can't ignore the
01:06:34
economic impact of this last war, right?
01:06:36
And it's it's the economy that's going
01:06:39
to accelerate everything, not AI getting
01:06:42
there.
01:06:42
>> We're already at mutually assured
01:06:44
destruction
01:06:44
>> for sure
01:06:45
>> with nuclear weapons. So, there's no
01:06:46
nuclear powers that are in direct
01:06:48
conflict.
01:06:49
>> We are in mutually assured destruction
01:06:51
of nuclear weapons is a statement that I
01:06:55
would have agreed to if Iran had a
01:06:57
nuclear weapon and that would have
01:06:59
stopped America from attacking Iran. You
01:07:01
you understand what that point means? It
01:07:04
means that not every nation in the world
01:07:06
has a nuclear weapon. The the the the
01:07:09
mad situation the mutually assured
01:07:11
destruction situation is only among
01:07:13
nuclear players, right? Uh autonomous
01:07:16
weapons are so cheap, so manageable that
01:07:20
every nation in the world is developing
01:07:21
them as we speak.
01:07:23
>> But they will also develop defenses.
01:07:26
>> Correct. But but
01:07:27
>> which I think is what people have
01:07:28
figured out now because of this recent
01:07:29
Ukrainian war is that if if if you need
01:07:32
to use a ballistic missile which costs I
01:07:34
don't know 2 million $3 million whatever
01:07:36
it is to target a $20,000 drone you're
01:07:39
you're [ __ ]
01:07:40
>> So you need a $20,000 defense solution
01:07:42
for a 20,000 weapon which I think is
01:07:45
probably going to happen.
01:07:46
>> It's very doable. Yeah. It's just that
01:07:48
you have to get rid of your THAD
01:07:49
batteries to be able to say the next
01:07:52
wave of of defense has to be drones. You
01:07:54
could imagine a world where like a wall
01:07:56
of drones fly up to where the drone is
01:07:58
incoming and they kind of block it. They
01:08:01
all explode at the same time to block
01:08:02
it, knock it out of the air. I heard
01:08:04
Palmer Lucky, who's one of the guys
01:08:06
who's building um Andrill.
01:08:08
>> Yeah, I know. And yeah,
01:08:09
>> talk at length about some of the
01:08:10
technologies that they have coming and
01:08:12
it's just mindbending. Mindbending
01:08:14
steady.
01:08:15
>> They showed me this gun where you you
01:08:18
just you just point it in the rough
01:08:20
direction. So, a pistol that aims for
01:08:23
you depending on where the target is.
01:08:24
So, you don't even have to aim it.
01:08:26
>> You can imagine at war. You just you
01:08:27
hold it up and shoot. Yeah.
01:08:28
>> And the it has this AI on the top of
01:08:31
like the barrel which will turn your
01:08:33
hand perfectly so that you hit the
01:08:35
target every time.
01:08:36
>> Wow. I should watch this video of it
01:08:38
happening.
01:08:39
>> Yeah. Well, I I I have uh Palmer on in
01:08:42
my film saying and yes, AI will kill a
01:08:45
few people by mistake or kill people by
01:08:47
mistake. uh you know when killing
01:08:50
becomes so easy
01:08:52
>> you do more of it okay when killing
01:08:54
becomes liability free and emotions free
01:08:59
and guiltree and you know you don't get
01:09:02
your soldiers coming back from Vietnam
01:09:04
with you know uh uh PTSD and so on it
01:09:09
you get more of it 70% of people who add
01:09:12
something to their online cart never
01:09:14
actually buy it and that number is based
01:09:16
on over 10 years of research but what I
01:09:18
think is Even more interesting is what
01:09:19
the Bayard Institute discovered. They're
01:09:21
a private research company that ran a
01:09:23
study which found the average e-commerce
01:09:25
store can increase its conversion rate
01:09:27
by 35% just by making its checkout
01:09:30
easier. Not better marketing or better
01:09:32
products, but by delivering a smoother
01:09:34
checkout experience. So, if you're
01:09:35
looking for an easy way to make your
01:09:37
checkout process smoother, I want you to
01:09:39
think about moving your business onto
01:09:40
Shopify. It's the platform we use to
01:09:42
sell the 1% diaries and the conversation
01:09:44
cards because it's so simple and smart
01:09:47
to use. It puts all of our inventory,
01:09:48
payments, and analytics in one place and
01:09:51
has so many AI tools to help us get up
01:09:53
and running straight away. Not to
01:09:54
mention that it grows with you
01:09:56
regardless of the stage that your
01:09:57
business is at. So, if you're ready to
01:09:59
fix your checkout process, sign up for
01:10:00
your $1 per month trial at
01:10:04
shopify.com/bartlet.
01:10:06
That's shopify.com/bartlet.
01:10:08
And don't tell anybody. Every time I've
01:10:10
tried to improve something in my life,
01:10:12
like my businesses, my health, my
01:10:13
relationships, I've noticed that the
01:10:15
biggest shifts have come from being
01:10:17
better informed. And when it comes to
01:10:18
our health, most of us know very, very
01:10:20
little. So, when our team was approached
01:10:22
about partnering with function health,
01:10:24
it felt very much aligned. Their team
01:10:26
has developed a way of giving you a full
01:10:27
360 degree view of your health, many of
01:10:30
the things that are going on in your
01:10:31
body in the form of different tests. You
01:10:33
do one blood draw and it gives you
01:10:35
access to over 160 lab results.
01:10:38
Hormones, heart health, inflammation,
01:10:40
stress, toxins, the whole picture. I use
01:10:43
it and so have many of my team members.
01:10:44
>> You sign up and you schedule your test
01:10:46
and once you're done, you get a little
01:10:47
report like the one I have here. I can
01:10:49
see my inrange results, my out of range
01:10:51
results, and there's a little AI
01:10:53
function, too. So, if I have any
01:10:54
questions about my out of range results,
01:10:56
I can just go in there and ask it any
01:10:58
question I want. And these tests are
01:11:00
backed by doctors and thousands of hours
01:11:01
of research. It's $365 for a yearly
01:11:04
membership. Go to
01:11:05
functionhealth.com/doac
01:11:08
and use the code DOAC25
01:11:10
for $25 off your membership.
01:11:13
>> So, we're in 2026 now. By 27, you think
01:11:17
there'll be AGI amongst us?
01:11:19
>> How will life look at all different?
01:11:21
Like what will we what will be the
01:11:22
symptoms of that world? Is it just a
01:11:24
slight increase in unemployment? I I
01:11:26
think there will be a very very ser
01:11:29
serious um differentiation between those
01:11:32
who plug into AGI and those who don't.
01:11:35
>> What is the symptom
01:11:37
>> that we notice is when we look at the
01:11:39
news or whatever what
01:11:40
>> you'll see people like you and I
01:11:42
building a company in 6 weeks and and
01:11:45
people uh that are not fully plugged
01:11:47
into AI really struggling to find a job.
01:11:50
>> Okay. So unemployment is going to be the
01:11:51
key symptom in 2027.
01:11:54
>> Yeah. And also also I think on the
01:11:56
positive side you're going to see
01:11:58
incredible scientific discovery. One of
01:12:01
my predictions not in those envelopes
01:12:03
but uh science itself is just uh we're
01:12:07
opening up capandor's box to be honest.
01:12:09
>> By 2030 what you what do you expect the
01:12:11
symptoms of AI to be?
01:12:13
>> So so jobs as I said I I'm I'm I'm
01:12:16
bowled a little bit on this on on the
01:12:19
fact that 30% of jobs would disappear by
01:12:22
2028. Okay. of some sectors not all
01:12:24
sectors by some sectors will will
01:12:27
disappear by 2028.
01:12:29
>> So up to 30% of jobs will be gone in
01:12:31
2027 to 2028
01:12:34
30% of jobs
01:12:35
>> of certain sectors of jobs. So, so if if
01:12:38
you call if you think about call center
01:12:39
agents, okay. Uh, yeah, probably. Uh, if
01:12:44
you think about uh graphics designer,
01:12:47
yeah, probably.
01:12:48
>> What do you think that looks like in
01:12:49
terms of unemployment, but also like
01:12:51
societal impact?
01:12:52
>> Horrible.
01:12:53
>> I think the Great Recession had 6%.
01:12:56
>> Yeah, we've never seen numbers like
01:12:58
this.
01:12:58
>> Jobs lost. Yeah. It said um
01:13:01
even economists who project just a net
01:13:03
loss of about 6% of US jobs by 2030,
01:13:07
they are mirroring the severity of the
01:13:09
great recession.
01:13:10
>> Yeah.
01:13:11
>> The real danger is a hiring freeze on
01:13:13
entry-level white collar jobs, air
01:13:14
automates the grunt work, which means
01:13:15
companies are shrinking their teams and
01:13:17
cutting off the bottom rung of the
01:13:18
corporate ladder for the next generation
01:13:20
of workers.
01:13:21
>> Correct. We have an entire generation
01:13:23
that is out of college today that will
01:13:25
struggle unfortunately. And and and my
01:13:28
advice to them is learn the tool and
01:13:32
focus on humancentric jobs.
01:13:34
>> Like what?
01:13:35
>> Like playing jazz.
01:13:37
>> I mean, not a lot of people can make a
01:13:39
living from paying playing jazz.
01:13:40
>> I understand that. But a lot of people
01:13:42
can make a living by being a nurse or by
01:13:44
being a a counselor or by being a uh you
01:13:48
know um um anything that that connects
01:13:52
to humans.
01:13:55
I I I just want to constantly come back
01:13:57
to this. None of that has to happen. If
01:14:01
if there is
01:14:04
if there really is a democracy and the
01:14:07
government is supposed to do what's good
01:14:09
for the people, the people need to stop
01:14:12
letting this from happening.
01:14:14
>> Which people?
01:14:15
>> Everyone.
01:14:16
>> Everyone in China.
01:14:18
>> China is not going to struggle as much
01:14:20
as as the West. I can guarantee you
01:14:22
that. I think this is the the question
01:14:24
people come back to is, well, if the US
01:14:25
stops, then we're going to end up being
01:14:27
China's lap dog. We're going to end up
01:14:28
>> You already are.
01:14:30
>> There's a lot of people that I know that
01:14:31
are using Chinese AI models to do their
01:14:33
work because for whatever reason,
01:14:34
because they're cheaper, they're better
01:14:35
in some respects
01:14:36
>> and because I cannot guarantee what what
01:14:38
American AI models are going to do to
01:14:40
me. So, Emma, my my startup is running
01:14:43
literally uh model agnostic. So one day
01:14:47
I'll plug in you know um um an open AI
01:14:52
chat GPT40 open source and the next day
01:14:55
I I'll plug in deepsek and I cannot
01:14:57
depend I cannot guarantee if America
01:15:00
continues to build to to make compute
01:15:03
more expensive. I cannot guarantee that
01:15:05
I can run a business on something that I
01:15:08
don't know the cost of in the future.
01:15:09
>> So the US can't just stop can they? They
01:15:12
need to change approach from from and by
01:15:16
the way the more interesting side is
01:15:18
what are the other economies doing like
01:15:20
is is the UK going to continue to import
01:15:23
compute
01:15:25
is is this I mean welcome to Africa
01:15:27
welcome to the third world
01:15:29
>> but this is what I mean like so you're
01:15:31
you're saying also that the the every
01:15:33
nation needs to invest
01:15:35
>> 100% every nation and and it's quite
01:15:38
interesting there is so much open source
01:15:40
that is not the state of the art
01:15:42
frontier model, but that can do 80% of
01:15:45
the tasks that the frontier model is
01:15:46
doing.
01:15:47
>> But on this point of companies competing
01:15:49
with each other, there's an inherent
01:15:51
need to compete here and to go. That's
01:15:53
what I'm hearing is like there's an
01:15:54
inherent need to go as fast as you can
01:15:55
or you will become a third world
01:15:57
country.
01:15:58
>> Sure, 100%.
01:15:59
>> You're saying that the people should
01:16:00
stop that.
01:16:02
>> No. So, so I'm saying the people of the
01:16:04
UK need to go to the UK government and
01:16:06
say, "How are you protecting the future
01:16:09
of our economy?" right? Are you going to
01:16:12
continue to import technology and to
01:16:15
empower import of technology versus
01:16:18
changing your regulations so that
01:16:20
innovation becomes easier here?
01:16:22
>> Okay.
01:16:22
>> So I mean think of it this way that
01:16:24
remember that anthropic bubble when when
01:16:27
you said the all of the SAS model
01:16:29
applications were you know being
01:16:32
basically threatened because anyone can
01:16:34
build an Oracle ERP today,
01:16:36
>> right? Why is nobody building an Oracle
01:16:39
ERP in the UK? saving the UK massive
01:16:42
licenses that go to Oracle every year.
01:16:44
>> Okay. So, you're saying that the the
01:16:46
people should go and ask the government
01:16:47
to invest more in AI.
01:16:49
>> 100%. That's what number one. Number two
01:16:50
is
01:16:51
>> but then their jobs go.
01:16:53
>> But but you see the the most interesting
01:16:55
job going forward is being an
01:16:57
entrepreneur is being is using those
01:16:59
tools to replace an economy that we've
01:17:02
built over trillions of dollars over the
01:17:05
last 50 years.
01:17:06
>> Not everyone can be an entrepreneur
01:17:07
though. No. Everyone can be an
01:17:09
entrepreneur in something.
01:17:10
>> Those entrepreneurs need to hire people.
01:17:13
>> That's a change that I think is about to
01:17:15
happen.
01:17:16
>> Like even in my business, I'm always
01:17:17
going to hire I'm always going to have
01:17:18
to hire people because
01:17:18
>> But you're But you're a massive
01:17:20
business. A shoe maker is not an an
01:17:22
entrepre is also an entrepreneur, but
01:17:24
not a massive business. A little
01:17:25
restaurant is also an entrepreneur, but
01:17:27
it's not a massive business.
01:17:28
>> They still need to hire people, though.
01:17:29
>> They they do, but they you know,
01:17:31
basically, if you're a a cafe with you
01:17:34
and your wife as baristas, you don't.
01:17:37
This is also an entrepreneur.
01:17:39
>> Can the economy work in such a way where
01:17:41
everybody is an entrepreneur?
01:17:41
>> It did. It did before capitalism changed
01:17:44
that around.
01:17:45
>> Everyone was an entrepreneur.
01:17:46
>> Of course, you know, in the earlier days
01:17:48
you you raised chicken and sold eggs and
01:17:51
others, you know, grew tomatoes and and
01:17:54
so you know, traded them for your eggs.
01:17:56
That that actually is a very interesting
01:17:59
thing. Imagine that, huh? Imagine a
01:18:01
world h where there is so much power
01:18:04
concentration at the top and UBI for
01:18:07
every everybody else. How do you think
01:18:10
that world will respond when all their
01:18:13
income is UBI? They'll respond by doing
01:18:15
things on the side. They'll respond by
01:18:17
going back to a barter economy. They're
01:18:19
going back to smaller communities.
01:18:20
They're going back to pap and mom mom's
01:18:22
job uh shops.
01:18:24
>> So the point was then you said this
01:18:25
doesn't need to happen.
01:18:26
>> It does not need to happen
01:18:28
>> which is the job loss
01:18:29
>> or the arms race. But we're telling our
01:18:31
governments, you're saying that to tell
01:18:32
the UK government to like join the arms
01:18:34
race.
01:18:36
>> I'm I'm telling the UK government to
01:18:40
to create an independence within the the
01:18:42
the the UK economy so that they don't
01:18:46
have to be at the receiving end of
01:18:48
technology,
01:18:49
>> which is join the arms race.
01:18:51
>> You don't have to compete against anyone
01:18:53
else. You don't have to be better than
01:18:54
anyone else. You don't have to. You're
01:18:56
simply saying I can build those things
01:18:58
in my economy now.
01:18:59
>> But but I'm never I'm not going to use a
01:19:01
terrible UK AI as a UK person if there's
01:19:06
a great US AI. I'm going to use the
01:19:09
great US AI. So if you don't compete and
01:19:11
win, I'm not going to use you. That's
01:19:12
what you know,
01:19:12
>> I'm not saying replace the frontier
01:19:14
models. These are very, as we speak,
01:19:16
they are these are very uh compute
01:19:19
intensive. They're infrastructure in
01:19:20
intensive and so on. I'm saying replace
01:19:24
Microsoft Word.
01:19:26
Seriously, like how much intelligence do
01:19:29
you need to build a software that writes
01:19:32
documents?
01:19:33
>> I can do that. Like we we've built our
01:19:35
own applicant tracking system here. We
01:19:37
build our own software. So just just ask
01:19:39
yourself how much money is spent in the
01:19:42
US in the in the UK government or in the
01:19:45
or in the you know the the UA UK
01:19:48
corporate space on licenses of software
01:19:51
that you and I can vibe code in 4
01:19:53
minutes.
01:19:53
>> I'm thinking from the UK's perspective
01:19:55
cuz you know what's interesting with the
01:19:56
UK is the the the economy is struggling
01:20:00
from a growth perspective. Correct. And
01:20:02
I was watching this documentary the
01:20:03
other day that was saying the reason why
01:20:04
we keep throwing our leaders out is
01:20:07
because actually what they need to do to
01:20:09
turn the UK round is about is about 15
01:20:12
years of pain. And it's like it starts
01:20:15
with energy transformation. We need to
01:20:16
get better at we need cheaper energy
01:20:17
because we have some of the most
01:20:19
expensive energy in the western world.
01:20:21
We need to build more houses which means
01:20:23
that we're going to need to centralize
01:20:26
permitting for building houses and it
01:20:27
can't just be local burers deciding if
01:20:28
they keep their farms or not. We're
01:20:30
going to need to make TLDDR. It says
01:20:31
that there needs to be some painful
01:20:34
decisions made for the next 15 years.
01:20:37
But the problem with our democracy is
01:20:39
that when people are in power for four
01:20:41
years, they're quite short- termist.
01:20:43
We'll just talk about the boats, the the
01:20:44
the brown people coming across the seas,
01:20:46
they don't they don't have the room to
01:20:48
think long term. And so when I'm
01:20:49
thinking about like what the UK needs to
01:20:51
do to not become to not fall into
01:20:54
decline and to keep up, I'm trying to
01:20:55
get clarity on that as it relates to AI.
01:20:59
Are you saying that they need to join
01:21:01
the arms race and double down and invest
01:21:03
all their money in building
01:21:06
competitive AI so that people use our
01:21:08
technology here in the UK versus
01:21:10
America's cuz the software point the UK
01:21:13
aren't going to get involved in
01:21:14
software. I mean they've tried to build
01:21:15
software before the UK.
01:21:18
>> The world has changed two two points,
01:21:19
right? One one side is there there needs
01:21:21
to be an a replacement cycle of our
01:21:23
investment decisions anywhere in the
01:21:25
world. Okay. So when you say we don't
01:21:27
have enough money for you know we need
01:21:28
to revamp our energy infrastructure when
01:21:30
you say we need to build more housing.
01:21:33
Okay. Uh one way of doing that is
01:21:35
squeezing that budget out of other
01:21:36
areas. The other way of doing that is
01:21:39
either cutting cost in the economy
01:21:41
elsewhere so that you can redirect that
01:21:43
money or growing the economy so that you
01:21:46
can have more money to build those
01:21:47
things. Correct. The cutting cost I
01:21:50
believe is there is just I don't have
01:21:52
the numbers but we could probably do a
01:21:54
run a deep a deep search on it. uh
01:21:57
trillions are being paid in traditional
01:22:01
systems that are complete like genuinely
01:22:04
they can be I'm and by the way I'm not
01:22:06
asking the government I'm asking some of
01:22:08
the people listening to me right now to
01:22:10
build an ERP system a word processor a a
01:22:13
presentation player and a spreadsheet
01:22:16
okay and just go around and spread them
01:22:18
across the you know find find uh retail
01:22:21
systems find CRM systems the these are
01:22:24
easy replacements they're going be
01:22:26
better interfaces. They're going to be
01:22:28
much more effective. Build a general
01:22:30
ledger using AI so that you can close
01:22:32
every hour, not every month or every
01:22:34
quarter.
01:22:34
>> But the world economy is doing that. So
01:22:37
there's some kid in San Francisco now
01:22:38
that's allowing all of us to use his new
01:22:40
word.
01:22:41
>> Keep
01:22:41
>> for free.
01:22:42
>> Keep doing that and you're and welcome
01:22:44
to the to the to the third world.
01:22:47
>> Keep doing what?
01:22:48
>> Keep importing all of your tech from
01:22:49
elsewhere.
01:22:50
>> But we h we we don't want to be
01:22:51
uncompetitive. Like if you think about
01:22:53
if we go to I don't know North Korea, I
01:22:55
bet they have way worse tools in many
01:22:58
areas because they won't use external
01:23:00
tools and that means that they are at a
01:23:02
disadvantage.
01:23:03
So I I bet they're not allowing their
01:23:05
civilization to use Gemini or Chat GBT
01:23:08
or I bet
01:23:09
>> for sure.
01:23:09
>> And they're using something worse.
01:23:11
>> And you know what what's happening? I
01:23:13
mean look at Iran and how advanced they
01:23:16
became through sanctions by being
01:23:18
refused to use those technologies. They
01:23:19
had to build them themselves, right?
01:23:21
Look at China. Look at Russia. You know,
01:23:24
when I worked at Google, Russia was
01:23:27
protective of Yandex, the competitor of
01:23:29
Google. From one side because of
01:23:33
influence, because they didn't want an
01:23:34
American organization to own the
01:23:36
knowledge sharing of of their citizens.
01:23:38
But on the other side, economically, if
01:23:40
you if you made it difficult for Google
01:23:43
to operate freely, that by definition
01:23:45
meant you had to invent a replacement on
01:23:49
the ground,
01:23:49
>> which was worse.
01:23:51
>> Yandex is not worse.
01:23:53
>> I'm saying so if you think about global
01:23:55
is global competition going to produce a
01:23:57
better product than regional
01:23:58
competition?
01:23:59
>> It depends on on where in the stack of
01:24:02
the quality of that product you need to
01:24:04
be, right? You do not need the ultimate.
01:24:08
I mean, ask yourself this. Which version
01:24:10
of Gemini are you using?
01:24:12
>> The most recent one. And I'm and every
01:24:14
day I compete.
01:24:15
>> And that is you. Ask everyone else what
01:24:17
version of of Gemini they're using. Most
01:24:19
people will say, "Oh, Gemini has
01:24:21
versions, right? You you don't need the
01:24:24
ultimate super frontier model to do 90%
01:24:28
of the tasks." And most people do only
01:24:31
70% of the tasks. But even so, the world
01:24:34
moves with whatever's better. So if you
01:24:37
think about the reason why we don't use
01:24:38
Yahoo search,
01:24:40
Google search is marginally better.
01:24:42
It's, you know, people, it's not, it's
01:24:44
not a thousand times better, but over
01:24:46
time we move away from AOL to Google
01:24:49
because it's a better product. And
01:24:51
there's a slow, you know, my dad
01:24:53
probably still using Chat GB2, but when
01:24:55
he when I go home for Christmas this
01:24:56
year and I go, "Dad, you should use
01:24:57
Gemini." And he starts messing around
01:24:58
with it, he'll slowly migrate. I'm like
01:25:00
I'm like a an early adopter. So I b 4.8
01:25:04
on claude came out last night. I'm on
01:25:05
there within honestly within 25 minutes
01:25:07
of the announcement.
01:25:08
>> I'm there as soon as I can. But I that's
01:25:10
you and I.
01:25:10
>> Yeah. And we we are an indicator of
01:25:12
where the world is going because we're
01:25:13
at the coldfront figuring out what's
01:25:15
better. And also I think with technology
01:25:16
eventually there does become the gap
01:25:19
between first and second begins to
01:25:20
widen. I think we're in a bit of a race
01:25:23
at the moment. But I think it is a bit
01:25:25
of a winner takes all situation with
01:25:27
these frontier
01:25:28
>> models. I I think we're talking about
01:25:30
two uh bits of technology. Okay. Tell me
01:25:35
how far has PowerPoint advanced since
01:25:38
2023
01:25:39
they added Copilot. Anything else?
01:25:42
>> So you're saying that the UK should
01:25:43
build their own PowerPoint?
01:25:45
>> For sure. There is the what I'm trying
01:25:47
to say is that from a licensing point of
01:25:49
view h just licensing of software within
01:25:52
government alone
01:25:54
how much money is is being repatriated
01:25:57
>> if the UK tried to build their own do
01:25:58
you know they tried to build their own
01:25:59
co app and it cost them I think these
01:26:02
numbers will be wrong but they're like
01:26:03
directionally true cost them 70 million
01:26:06
to build a co app that didn't work
01:26:08
>> and so they canned it
01:26:09
>> I remember I'm asking I'm asking a UK
01:26:13
entrepreneur
01:26:14
>> to wake tomorrow and say, "All right,
01:26:16
you know what? Between coffee and my my
01:26:19
cookie, I'm going to build a PowerPoint
01:26:21
and go sell it."
01:26:22
>> Yeah, they will. And they'll go to San
01:26:23
Francisco where the money is and the
01:26:25
talent.
01:26:26
>> Yeah.
01:26:26
>> Or they won't be able to compete.
01:26:28
They'll get crushed tomorrow. So, if
01:26:30
they don't go where the talent and money
01:26:31
is, I'm not saying that they'll they all
01:26:32
have to go to America, but I'm saying
01:26:34
what they will do if they want to
01:26:36
produce the best product is they'll make
01:26:38
rational entrepreneurial decisions about
01:26:41
um where to sell it, how where to raise
01:26:44
money, where they're going to get the
01:26:45
best talent, and I think if they don't
01:26:48
make those, you know, competitively
01:26:50
minded decisions, they're not going to
01:26:51
make the better product. They're not
01:26:52
going to get any users.
01:26:54
users users will go where it's better
01:26:57
and where it's cheaper. They won't they
01:26:59
won't say, "Oh, I want to use this
01:27:00
because it's from Cornwall."
01:27:02
>> Yeah. So, I'm I'm with you, right, that
01:27:05
this actually might be the big corporate
01:27:08
thought thing. Okay. And I'm with you
01:27:12
that of course the infrastructure here
01:27:14
of building a startup is way more
01:27:16
complex than it is in other places where
01:27:19
startups succeed. Okay. I'm saying if we
01:27:22
continue on that trajectory whether in
01:27:24
the UK or in Germany or in you know uh
01:27:28
Zanzibar welcome to Africa all of us
01:27:32
everyone but the two competitors China
01:27:35
and and America is going to be third
01:27:38
world so when you talk about job losses
01:27:41
for individuals that's one side okay but
01:27:45
but nation positioning losses which I
01:27:48
think Europe has noticed recently
01:27:52
is about to happen everywhere. And why?
01:27:54
Because you're saying, "Hey, you know
01:27:56
what? Maybe we can't do it. Why can
01:27:59
China do it? Why can Korea do it? Is not
01:28:03
because they have different natural
01:28:05
resources or not because they are, you
01:28:08
know, in a place where it's warmer. It's
01:28:10
because their regulations, their their
01:28:13
ambitions are to empower something
01:28:15
different than debating about, you know,
01:28:19
rail railway lines or
01:28:20
>> they have a couple of big advantages
01:28:22
that I was reading about. One of them is
01:28:23
they they have way cheaper energy, which
01:28:26
means that they're going to be able to
01:28:27
pursue
01:28:28
>> why, may I ask?
01:28:29
>> Because they've invested in solar power,
01:28:31
renewable energy, because they don't
01:28:33
care about
01:28:33
>> this is, by the way, I say that
01:28:34
publicly. I say that the arms race of AI
01:28:37
was won a long time ago. And then the
01:28:39
other thing is in terms of permitting,
01:28:41
if you want to build a data center in
01:28:42
the UK, listen, if you want to like open
01:28:44
a cafe in the UK, you're going to it's
01:28:46
going to be war for you. If you want to
01:28:47
do that in California, this is why
01:28:48
everyone's left California. Like when I
01:28:50
took this office in California recently,
01:28:52
I was like, "So, how long is it going to
01:28:53
take for me to renovate this?" And the
01:28:54
agent looked at me and was like, "To
01:28:56
renovate my own office, it's going to
01:28:58
take me a year and it's permitting to
01:29:01
renovate my own office
01:29:02
>> in California." Mhm.
01:29:04
>> So if you're competing with China where
01:29:06
President Xi just goes, "Put that there
01:29:08
and you've got 7 days."
01:29:10
>> No, no, no, no. That's not only that. So
01:29:12
I I used to be one of the very few
01:29:14
Google executives around allowed into
01:29:16
government meetings in China. Uh simply
01:29:19
because I'm from emerging markets, so I
01:29:21
understand respect. Basically, I would
01:29:22
say sit there not as if I'm superior to
01:29:25
them, but as as if I'm really interested
01:29:28
to learn from them. And genuinely uh
01:29:31
Stephen when when when I would sit in
01:29:33
those meetings all in Chinese they would
01:29:36
show slides
01:29:38
that have competitive market share. So
01:29:40
they'd say you know it's not like uh
01:29:44
China is this, America is this, Germany
01:29:46
is this from market share point of view.
01:29:48
It was China versus the world.
01:29:50
>> Mhm.
01:29:50
>> Right. And when they would decide to go
01:29:52
for something 5G, uh, you know, uh,
01:29:55
internet of things, all of that stuff,
01:29:57
they would aim for 98% market share.
01:30:00
>> Yeah.
01:30:00
>> Right. And and they get there. I mean,
01:30:02
look at how what how they did electric
01:30:04
cars and and and my my question is very
01:30:09
simple, huh? We've spoken so much about
01:30:11
AI. My question is,
01:30:13
>> are the people of the West going to
01:30:14
wait? So this is an interesting
01:30:17
conundrum because it sounds to me like
01:30:19
you're saying on one end if we don't
01:30:20
join this AI race then we're a third
01:30:24
world country but on the other end if we
01:30:26
continue this sort of thoughtless race
01:30:29
towards AGI there's going to be
01:30:32
catastrophe at some point
01:30:34
>> and the answer is somewhere in the
01:30:35
middle where basically you join the AI
01:30:37
AI race for the good of your community.
01:30:40
Okay. So, so there is resignation on one
01:30:43
side is like I'm not going to play this
01:30:44
game at all. There is uh um offense on
01:30:48
the other side where I'm playing this
01:30:50
game to destroy everyone else. And there
01:30:52
is a balance in the middle where we say
01:30:55
we're going to build an ethical form of
01:30:56
AI that's going to help our communities.
01:30:59
We're going to use this ultimate gift,
01:31:01
this ultimate superpower, right?
01:31:04
Superman landed on the planet, raise it
01:31:07
to help your community.
01:31:08
>> Is that wishful thinking to some degree?
01:31:10
when you when you think about the the
01:31:11
nature of sort of compet competition
01:31:13
>> I don't know how what how to tell you
01:31:15
otherwise I genuinely believe most I
01:31:19
mean here I genuinely believe it's going
01:31:22
to be very difficult to make that change
01:31:24
okay I genuinely believe that it's going
01:31:26
to be that when the challenges of an AI
01:31:29
dystopia hits us we're not going to be
01:31:31
ready but I can't stop talking about it
01:31:33
Stephen do you understand where I stand
01:31:35
with this I I just am hoping and I tried
01:31:40
so hard. I spoke to the leadership then
01:31:43
I spoke like one of the things that my
01:31:45
uh Atlantic Productions who helped me
01:31:47
with the film uh helped you know did for
01:31:49
me which I have to say completely I am
01:31:53
very appreciative of this is you know
01:31:55
through our conversations our through
01:31:57
our the last few years that I tended to
01:32:01
at a point in time lose hope in the
01:32:03
leadership
01:32:04
and basically try to influence the
01:32:07
public for ethical AI. Okay. And my
01:32:10
conversation was that the leadership at
01:32:12
the time was the technical leadership
01:32:14
and that everyone was so caught up in
01:32:16
the arms race that I wanted to teach the
01:32:18
public to help us build ethical AI. And
01:32:21
I continue to focus on the public, the
01:32:23
every one of us. But suddenly Atlantic
01:32:25
goes, "No, no, hold on. We should
01:32:27
probably get you to meet the political
01:32:29
leaders everywhere in the world and
01:32:31
hopefully give them a message that says,
01:32:33
"Hey, you know what? You may actually
01:32:35
make a difference if you prioritize AI
01:32:37
differently, right? Do I believe this
01:32:40
will happen? Sadly, no. But does that
01:32:42
mean I should stop trying? I cannot stop
01:32:45
trying. This has had probably the single
01:32:48
biggest impact on my office. Of all the
01:32:51
products that I've tried that have given
01:32:53
me productivity gains or cognitive
01:32:55
boosts, I would say that exogenous
01:32:57
ketones are in the top three most
01:33:00
pivotal things that have given me
01:33:02
massive productivity gain. It's some
01:33:04
Stanford graduates that have been able
01:33:05
to basically bottle up the effect you
01:33:08
get from being in a ketogenic diet in a
01:33:11
small shot that you can take that makes
01:33:13
you feel incredibly focused and gives
01:33:15
your brain an incredible source of
01:33:17
energy. And the clinical studies that
01:33:18
have been done on exogenous ketones have
01:33:20
absolutely blown my mind. I reached out
01:33:22
to them. I became a coiner in the
01:33:23
company. I became an investor in the
01:33:25
company. And so it's with great pride
01:33:26
that I can tell you that this exists. If
01:33:28
you haven't tried these shots, go to
01:33:30
ketone.com/stephven
01:33:32
for 30% off your subscription order. And
01:33:34
you'll also get a free gift with your
01:33:36
second shipment. I still buy my ketone
01:33:38
shots predominantly online, but
01:33:40
thankfully I can now grab them at Target
01:33:42
whenever I drive past them here in the
01:33:43
United States as well because we're now
01:33:44
stocked in Target where your first shot
01:33:47
is completely free. I've done almost 700
01:33:50
interviews with some of the most
01:33:51
interesting people in the world. And one
01:33:53
of the things you learn which is
01:33:54
unexpected is that vulnerability is the
01:33:57
doorway to connection. And after sitting
01:33:58
here for 2 three hours with a guest I
01:34:00
feel a deep sense of connection to them.
01:34:03
And as they leave what I get them to do
01:34:05
is to write a question in the diary of a
01:34:09
CEO. We've taken all of the questions
01:34:10
from the diary of a CEO. We have put the
01:34:13
question here on this card with the name
01:34:16
of the person that wrote it. So you can
01:34:18
sit at home as I do with my fianceé and
01:34:20
my colleagues at work and other people
01:34:22
in my life. Whenever we get a minute, we
01:34:24
play the diio conversation cards and it
01:34:27
is incredible what happens. These are
01:34:30
great if you're in a romantic
01:34:31
relationship and you want to connect
01:34:32
your partner more. These are also great
01:34:34
if you're in a team and you want to bond
01:34:36
your team together. And I have to say
01:34:37
they're also great for families that
01:34:39
want to learn more about each other and
01:34:40
that need a good excuse to spend some
01:34:42
time in a digital world in the analog
01:34:45
environment connecting human to human.
01:34:47
It is remarkable what the right question
01:34:50
at the right time can do. Go to the
01:34:52
diary.com
01:34:54
and you can get these conversation cards
01:34:56
right now. When I have these
01:34:58
conversations about AI, what I'm trying
01:34:59
to do all the time is to pass out what
01:35:01
is like wishful thinking and then what
01:35:03
is reality? And like to understand
01:35:05
reality, you have to understand
01:35:07
competition. You have to understand
01:35:09
human emotion. You have to understand
01:35:10
incentive structures. And so you think
01:35:12
about something like the United States
01:35:13
at the moment where you've got Donald
01:35:15
Trump whose sort of primary driving
01:35:17
incentive is GDP growth, economy growth.
01:35:20
Does that stock market go up? Beat
01:35:22
China. So if you all agree that that's
01:35:23
like his like the core of his incentive
01:35:25
structure, then you've got President Xi
01:35:27
over here who's pro his incentive
01:35:28
structure is probably control,
01:35:31
independence,
01:35:32
um defense, so that you know they need
01:35:35
to make sure they they do well on the
01:35:37
weapon side. when you look at and then
01:35:38
you got these other nations like the UK
01:35:40
and Europe and these other places who
01:35:42
are kind of it seems like we're a bit
01:35:44
resigned to the fact that we're not
01:35:45
going to participate in the underlying
01:35:48
tech underlying models building because
01:35:50
we just don't have it together we don't
01:35:52
have the energy you know
01:35:54
>> and you go in such a world you go
01:35:57
ethical AI who's going to who's going to
01:35:59
prioritize ethical AI I mean anybody
01:36:01
that does is is might fall behind
01:36:04
theoretically is that um so I wonder
01:36:07
where the bolt in the thinking is here
01:36:08
like what how do we get to a point of
01:36:10
ethical AI when the incentive structures
01:36:12
are so clearly highly competitive and
01:36:16
arguably a little bit shortterm
01:36:20
uh in their thinking.
01:36:21
>> Yeah.
01:36:23
So what are you saying we lie down and
01:36:25
wait?
01:36:25
>> No, I just don't know. I just don't have
01:36:27
an answer honestly.
01:36:28
>> That's why we keep That's why we keep
01:36:31
talking about it.
01:36:31
>> Yeah. And that's why I keep spending 14
01:36:34
hours a day trying to tell the world
01:36:35
because some genius somewhere is going
01:36:37
to find an answer. But the way it's
01:36:39
going
01:36:40
>> right now, I guess what we're pursuing
01:36:42
is we're hoping that CHBT and Anthropic
01:36:44
and these and Grock and we're hoping
01:36:45
that they just build ethical models and
01:36:47
we're hoping that social pressure forces
01:36:49
them into making good decisions.
01:36:52
>> Correct. We we need to be able to vote
01:36:54
with our usage. Right. So, so I think
01:36:57
one of the biggest movements in AI since
01:36:59
we started was the idea that so many
01:37:02
people switched away from Chad GPT when
01:37:04
they approved that their model can
01:37:05
target people, right? So many people I
01:37:08
know at least said
01:37:09
>> target people
01:37:10
>> I when anthropic refused to have to to
01:37:12
>> do you think people switched?
01:37:14
>> I think many have I think ones that are
01:37:16
aware
01:37:17
>> right and I think they did because the
01:37:19
cost of switching is really I mean
01:37:22
honestly anthropic is is better if you
01:37:24
think about
01:37:24
>> Yeah, I think it's better.
01:37:25
>> Yeah. But the idea is if people don't
01:37:28
switch for those ethical reasons. So you
01:37:31
know every one of my books has this
01:37:33
dedication at the beginning. It used to
01:37:35
be the gravity of the battle means
01:37:36
nothing to those at peace when I when I
01:37:38
wrote in memory of Ali at the beginning.
01:37:41
My next book the alive. Do you remember
01:37:44
a song called the well it they were
01:37:48
called the manic street preachers.
01:37:51
>> Uh yeah. If you tolerate this then your
01:37:53
children will be next. Mhm.
01:37:56
>> I genuinely believe that what the world
01:37:57
needs to wake up to is if you tolerate
01:38:00
this, then your children will be next.
01:38:02
If you continue to resign, if you
01:38:05
continue to say I'm not going to try, h
01:38:08
this world is going to change in a way
01:38:09
that is completely not for in your
01:38:12
favor.
01:38:12
>> Try what?
01:38:13
>> Try to to to stand up and say something
01:38:16
needs to change.
01:38:18
>> Okay, we can say something needs to
01:38:20
change, but we can't say what that thing
01:38:21
is. What needs to change is governments
01:38:24
need to serve h their people not their
01:38:27
interests and corporates need to work
01:38:31
for the benefit of benefit of their
01:38:33
societies before their shareholders.
01:38:36
This this run that we had with
01:38:38
capitalism so far
01:38:41
has benefited the world tremendously.
01:38:45
>> Whose economy do you think is going to
01:38:46
be in a better place for the middle
01:38:48
class out of the say the UK and the
01:38:50
United States? The UK is gone.
01:38:52
>> It's gone because what? In part because
01:38:54
it didn't compete.
01:38:55
>> Because you're an older bureaucracy that
01:38:58
is burdened down by so much barriers on
01:39:03
the in the process of building anything,
01:39:06
right? Because the US economy in the
01:39:10
past welcomed people like me to go and
01:39:13
live in California and build amazing
01:39:16
[ __ ] right? That is no longer the case.
01:39:19
So, who is going to win? In my view,
01:39:20
it's definitely China.
01:39:23
>> And by the way, you asked for the middle
01:39:25
class. So, so China made decisions
01:39:27
recently that forced businesses not to
01:39:30
lay people off in replacement to be
01:39:32
replacing them with AI. Would would the
01:39:34
West do that? The capitalist West would
01:39:37
never do that. We don't know the answer.
01:39:39
I don't know the answer. I'm responding
01:39:41
to your
01:39:41
>> Yeah, you can see the conundrum I find
01:39:43
myself in, which is a state like a
01:39:46
country like the UK is, in your words,
01:39:48
gone because it didn't
01:39:52
compete. It didn't allow people to be
01:39:54
highly entrepreneurial. It didn't
01:39:56
empower entrepreneurialism, innovation,
01:39:59
ingenuity. At some in some way, it stood
01:40:03
in their way in some in some way. Now,
01:40:04
that could be incentives, it could be
01:40:06
culture, it could be whatever.
01:40:07
>> Yeah, I know where you're going with
01:40:08
this. the US didn't. So they have got
01:40:11
have a economy which is arguably more
01:40:15
productive and um future proofed than
01:40:20
ours.
01:40:22
By way of that they are also
01:40:26
more advanced in artificial intelligence
01:40:29
and we are gone. So the remedy for a
01:40:31
country like us would be therefore
01:40:33
>> to compete
01:40:34
>> to compete let the reinss off let
01:40:36
entrepreneur ingenuity but then we're
01:40:38
saying that's dangerous
01:40:39
>> and your conundrum in that is that
01:40:41
you're assuming that entrepreneurship by
01:40:43
definition is malicious.
01:40:45
>> No I'm just I'm just saying that um
01:40:47
there's a bit of a paradox like you're
01:40:49
damned if you don't you're damned if you
01:40:51
do
01:40:51
>> but you're not damned if you build
01:40:53
things for for the people not for the
01:40:56
capitalist. This is an ideological
01:40:58
debate. I pray and hope that that's that
01:41:00
is plausible, but I'm I'm worried that
01:41:04
in a competitive market, whoever's
01:41:07
optimizing for I don't know, you can
01:41:09
name it retention. If if we built two
01:41:12
AIs, right? I'm going to call the the
01:41:15
moore AI and I'm going to call the other
01:41:16
one evil AI. The evil AI is programmed
01:41:20
to retain you. It's sick of fantic. It
01:41:22
says what you need to hear. It it's
01:41:24
super smart. And because of that, even
01:41:28
though it's not trading in your your
01:41:29
best interests, it it's retaining you
01:41:32
more. You're using it more often. It's
01:41:34
programmed for that. Kind of like the
01:41:35
social networks are. They're all just
01:41:37
trying to like dopamine your brain into
01:41:39
oblivion. Then there's the MO AI. It
01:41:42
tells you to log off. You've had enough.
01:41:43
It thinks about your mental well-being
01:41:45
because it's less retentive and less
01:41:47
engaging. Theoretically, it might be
01:41:49
less successful as a commercial product.
01:41:52
Think about social networks. the ones
01:41:54
that are least retentive, the ones that
01:41:56
actually won't um destroy your brain
01:41:58
with dopamine, the ones that remove the
01:41:59
retweet button, the ones that don't have
01:42:02
um slot machine like videos, they don't
01:42:04
survive.
01:42:05
>> I share this with you and my my point of
01:42:08
view is that you're to summarize your
01:42:10
challenge here. You're basically saying
01:42:13
that it's easier to become successful if
01:42:16
you don't follow ethical rules. I'm
01:42:19
asking the question, if you build an AI
01:42:21
that is just purely focused on ethical,
01:42:24
will it be as engaging and have the same
01:42:26
usage as one built with the reins off?
01:42:30
>> Uh, no,
01:42:31
>> it won't.
01:42:32
>> Yeah, but that's but that's the problem
01:42:34
humanity needs to solve.
01:42:36
>> Okay.
01:42:36
>> If we were to survive, example, I worked
01:42:39
in a company called Google that
01:42:41
basically at a point in time decided
01:42:44
that ads will be effective.
01:42:46
>> Okay. The ad industry prior to Google
01:42:49
was 50% of your ad budget doesn't work.
01:42:52
We just don't know which 50%. Remember
01:42:53
that?
01:42:54
>> Yeah.
01:42:54
>> Right. And and Google came in and
01:42:56
struggled from 1998 until 2004 when they
01:43:01
started to turn probable, you know, uh
01:43:03
plausible revenues as a result of
01:43:05
saying, "We're going to run a Dutch
01:43:07
auction and we're going to give you
01:43:08
pay-per-click and we're going to show
01:43:10
you results for your ads." Okay. They
01:43:12
found a way for ethics. H uh to actually
01:43:17
get your money to be effective rather
01:43:20
than just take your money and say 50%
01:43:21
doesn't work. H they found a way to make
01:43:24
that their success criteria. There must
01:43:26
be a way for us. I don't know what it
01:43:28
is. If I knew I would be building it
01:43:30
instead of sitting with you, right? But
01:43:32
there must be a way for us to marry the
01:43:35
success of humanity with the success of
01:43:37
the entrepreneur. Right? And and that
01:43:40
way is not found in the old ways of
01:43:42
doing things.
01:43:43
>> I've got an idea.
01:43:46
Maybe you know the claude 4.8 came out
01:43:49
yesterday which is one of the big AI
01:43:50
models the new one and when they
01:43:52
released the models they show these like
01:43:54
graphs of the benchmarks. What they mean
01:43:56
by this is what it's capable of. They
01:43:58
show how it performed in maths in
01:43:59
science in writing reasoning etc. It is
01:44:02
all marginal at this point but my my
01:44:04
idea is could there be an ethical
01:44:07
benchmark that all these models have to
01:44:08
pass before these large companies can
01:44:11
legally deploy them. And I was thinking
01:44:13
about like again every idea has
01:44:15
unintended consequences which I haven't
01:44:17
thought through but would it would be
01:44:18
very interesting that when they release
01:44:20
these models they also release the
01:44:21
ethical benchmarks. I we tried to get it
01:44:23
to X, we tried to get it to Y, we tried
01:44:25
to get it to zed and here's how it
01:44:27
performed against the ethical
01:44:28
benchmarks. That gives some kind of
01:44:29
standard for governments to say you're
01:44:30
not allowed to release a new model
01:44:32
unless it passes independent tested
01:44:34
ethical benchmarks.
01:44:35
>> Beautiful. That would absolutely work.
01:44:38
But notice by the way those are out
01:44:40
there already. Okay. But but in a very
01:44:43
interesting way. You listen to Deis
01:44:44
Hassabis and how much he invests heavily
01:44:48
in building Alpha Fold or building you
01:44:51
know uh um um so many scientific
01:44:54
applications of AI and you go like this
01:44:57
guy cares about science. I can't I can't
01:45:00
prove that. Right? Uh but I I you know
01:45:03
I've I've met Demis a couple of times. I
01:45:06
I know genuinely that he is an ethical
01:45:08
person, right? But the typical person
01:45:10
will probably say, but at least it seems
01:45:13
that they're doing things for free to
01:45:15
serve science. You look at uh you know
01:45:17
at anthropic and they refuse to use
01:45:20
their model to to uh allow the US
01:45:22
government to target and to spy on
01:45:24
people and then you see Open AI uh
01:45:27
accepting a $500 million deal that
01:45:30
absolutely does that. It is about time
01:45:33
that every person in the world says in
01:45:35
that case I am no longer going to use uh
01:45:39
open AI until they show me another you
01:45:41
know another evidence that they are
01:45:44
actually ethical in their behavior right
01:45:46
and this is the decision that you and I
01:45:48
can do
01:45:50
right and
01:45:51
>> don't they do they
01:45:52
>> but that's my task my task and yours is
01:45:55
to keep telling them people please
01:45:57
please understand that if you tolerate
01:45:59
this then your children will be next
01:46:01
Please understand that if you don't
01:46:03
start to take an ethical stand on your
01:46:06
own future, your future will be handed
01:46:08
over to another oligarch just like your
01:46:11
past was handed over to social media
01:46:13
oligarchs.
01:46:14
>> One would say booing booing at the
01:46:16
commencement speech is a good example of
01:46:18
how public awareness can have a real
01:46:21
impact on
01:46:24
this trajectory. But I still think at
01:46:27
the end of the day, if you think about
01:46:29
things like smartphone usage in schools,
01:46:31
at the end of the day, it does come down
01:46:34
to
01:46:35
government intervention and saying, "Do
01:46:37
you know what? We're going to ban
01:46:38
14-year-olds from scrolling Tik Tok."
01:46:40
And that's in part because people spoke
01:46:42
louder and louder and louder. They went
01:46:43
on podcast. Jonathan Height, who wrote
01:46:45
the book about the anxious generation,
01:46:48
started a conversation. And that
01:46:49
conversation led to legislation. I still
01:46:51
think it ends in like some kind of
01:46:53
constraint legally.
01:46:55
>> I hope. Okay. But I will openly tell you
01:46:58
most of the tech oligarchs are more
01:47:00
powerful than your government.
01:47:01
>> Is there any precedents in history where
01:47:04
this kind of change happened without
01:47:07
government intervention?
01:47:09
>> French revolution.
01:47:10
>> So I was thinking about things like
01:47:12
climate change, even you could say
01:47:14
smoking.
01:47:17
All these kinds of things have had to be
01:47:18
like
01:47:20
taxes. And
01:47:21
>> Stephen, I'm I'm with you. If
01:47:23
governments intervene, we wouldn't have
01:47:25
a problem. Governments won't intervene
01:47:27
because governments are owned by the
01:47:29
oligarchs, right? So my question for
01:47:32
everyone listening to us is, are you
01:47:34
going to intervene?
01:47:35
>> Well, cancel your trap.
01:47:37
>> If that's what you can do, it's fine. If
01:47:38
not, then go ahead and start a a startup
01:47:41
that that that does something. If you're
01:47:43
if that's not within your capability,
01:47:45
then send a message to your congressman.
01:47:47
If that's not within your capability,
01:47:48
then say something uh uh ethical online
01:47:52
so that the world understands a position
01:47:54
that needs to be opening the eyes. If
01:47:56
that doesn't work for you, at least
01:47:57
don't engage on stuff that is negative
01:47:59
that you don't know enough about. There
01:48:01
are so many little actions. If humanity
01:48:03
starts to move in the direction of one
01:48:06
saying ethics matter, not just profit,
01:48:08
okay? and two saying I'm not going to
01:48:10
participate in something that's
01:48:12
unethical just because I believe I wanna
01:48:15
you know I feel like it right now. Okay.
01:48:18
If I tell you the number of things I
01:48:19
took out of my life just to try and
01:48:22
affect a tiny bit of change of revenues
01:48:26
that go to bullets. if I tell you the
01:48:28
number of things.
01:48:29
>> I think there's two central concerns
01:48:31
I've always had which is I do feel that
01:48:32
there's going to be significant job
01:48:34
disruption and I don't think societyy's
01:48:35
prepared for it yet and I don't know
01:48:36
what that preparation looks like but I
01:48:38
think we should start thinking about it.
01:48:40
Um I share this with you. I you know I
01:48:44
genuinely believe that if we continue on
01:48:46
where we are there's no hope in the
01:48:49
trajectory of where humanity has become
01:48:53
so distracted
01:48:55
so resigned to inaction so um so
01:49:00
disconnected from their own rights of
01:49:03
freedom of expression and engagement and
01:49:05
so on. I think we have no hope. Do we
01:49:09
want to stay there? That's a question
01:49:11
that I'm asking our listeners. And I'm
01:49:13
not saying be violent or get up or, you
01:49:15
know, go be angry or whatever. I'm just
01:49:18
saying take one little action. Ask
01:49:20
yourself, please write it in the
01:49:21
comments. One little action. One little
01:49:24
action that you're going to do today
01:49:25
that's going to make the world a little
01:49:27
better tomorrow. And don't give up on
01:49:29
humanity, Stephen. I I I I'm not saying
01:49:31
you do, but I'm I'm saying we are going
01:49:33
through such a difficult time in
01:49:36
humanity's history that for the very
01:49:39
first time ever,
01:49:41
we have to do something about it. I
01:49:43
don't want my daughter to be at the
01:49:46
receiving end of what happened to my
01:49:48
son. I I don't I genuinely I lost Ali. I
01:49:51
don't want to lose ayah and the world
01:49:53
we're building is going to be very
01:49:55
difficult for ayat and I cannot go to
01:49:58
sleep at night without trying something
01:50:00
every day
01:50:02
and I genuinely don't understand how
01:50:04
humanity is not is missing that point
01:50:06
mainly I think because they're
01:50:08
uninformed now we're informing them okay
01:50:11
the only thing that will save humanity
01:50:13
going forward is that this superpower
01:50:16
called intelligence is used for ethical
01:50:18
reasons is that the corruption that's
01:50:20
leading us to where we are today stops.
01:50:24
>> Mhm.
01:50:26
You've got more envelopes there. What's
01:50:27
your next envelope? The third one.
01:50:29
>> Number two was job losses. Number three
01:50:31
was labor. You know, basically
01:50:33
>> same thing. Robots will replace manual
01:50:35
labor by 20.
01:50:36
>> We start to replace manual labor.
01:50:38
>> Okay.
01:50:38
>> Yeah. So, uh you know, you will you will
01:50:40
have more and more manual jobs given to
01:50:43
robots. What's this one? Uh oh, this is
01:50:47
absolutely. Do you think otherwise? the
01:50:49
the world's first trillionaire before
01:50:51
>> oh yeah probably well before then
01:50:53
>> well before 2030
01:50:55
>> I think you know Elon's about to IPO
01:50:57
SpaceX which is likely to make him a
01:50:58
trillionaire
01:50:59
>> yeah I think the the concentration of
01:51:02
power that comes with that is quite
01:51:04
drastic when you really think about it
01:51:07
uh and and that's 2030 is just a few
01:51:10
years away I I think the team wrote this
01:51:12
wrong artificial super intelligence will
01:51:14
arrive in 2032 20 to 2035 I I think
01:51:19
artificial super intelligence will
01:51:20
arrive the minute AGI happens. So it
01:51:23
doesn't really matter if AI is a billion
01:51:26
times smarter than you or just twice as
01:51:29
smart as you. Once we cross beyond AGI,
01:51:32
ASI is just
01:51:34
>> yeah,
01:51:35
>> very very soon.
01:51:37
And uh and yeah, I think we'll overcome
01:51:42
that when we get to the fourth
01:51:44
inevitable when AI is in charge of
01:51:47
everything. I genuinely believe that we
01:51:49
will end up in a utopia of abundance. I
01:51:52
genuinely believe that
01:51:54
again physics, mathematics and biology
01:51:56
will tell you that super intelligence is
01:51:58
benign and that we will eventually end
01:52:02
up in a good place. Not because humanity
01:52:04
has done much to get us there. Not
01:52:07
because our leaders have suddenly turned
01:52:08
ethical, but because our unethical
01:52:11
leaders have gone out of the equation
01:52:13
and were replaced with a super efficient
01:52:17
minimum energy principle that doesn't
01:52:20
see value in anything that's
01:52:21
destructive.
01:52:22
>> So, the future's going to be great.
01:52:24
>> Those who make it to 2038 will enjoy it.
01:52:27
Yeah,
01:52:28
>> those who make it,
01:52:30
>> for sure. I mean, World War II didn't
01:52:33
destroy the world, but ask those who
01:52:35
went through it.
01:52:37
It's just an interesting idea that
01:52:38
actually it's just you're you're
01:52:39
forecasting basically like a a
01:52:43
a decade of of turmoil of dystopia of
01:52:46
absolute dystopia.
01:52:47
>> When you say absolute dystopia, just so
01:52:49
I'm clear in my mind, the absolute
01:52:50
dystopia you're forecasting over a
01:52:52
decade is about war. It's about
01:52:54
>> war, economics, jobs.
01:52:56
>> Economics, it's about jobs.
01:52:58
>> It's it's also about surveillance and
01:53:00
control. It's also about digital
01:53:01
currencies. It's also about uh human
01:53:03
connection. Uh it's also about
01:53:06
concentration of power. It's it's a
01:53:09
magnification of everything we've built
01:53:11
so far.
01:53:12
>> And just again arm people with some
01:53:14
tools to survive that dystopia for a
01:53:16
decade. You know, we talked a little bit
01:53:18
about focusing on human jobs, the more
01:53:21
human jobs.
01:53:22
>> Yeah. Learn AI. So, so AI is not the
01:53:24
enemy. Okay. Uh by definition, the
01:53:28
better you are at using an AI to to do
01:53:30
your job,
01:53:31
>> Yeah. uh the more likely you are to be
01:53:34
successful. Right? Now, number two is
01:53:37
prepare for a hybrid world where AI and
01:53:40
humans work together.
01:53:41
>> How do you prepare?
01:53:42
>> You basically understand agents and how
01:53:45
agents work. You you understand how a
01:53:49
hyper
01:53:50
uh uh efficient approach to things may
01:53:53
not require you to be uh you know very
01:53:56
long meetings and very long. So, so
01:53:58
there is if you lived in California, you
01:54:00
would know that our you know the way we
01:54:02
ran businesses was a lot more efficient.
01:54:05
We sometimes had a 7-minute meeting. Uh
01:54:08
right. So, so the you know the habits of
01:54:11
of an AI are much more efficient than
01:54:13
the habits of humans.
01:54:14
>> So, learn AI again.
01:54:16
>> Uh learn how to interact with AI allow
01:54:18
welcome AI into your hybrid world of
01:54:20
work. Uh I think you need to uh to of
01:54:24
course double down on human skills. I
01:54:26
think that's a you know a must to
01:54:29
succeed in this world. I think we need
01:54:32
to one of my most interesting views on
01:54:35
the near future is how AI is going to be
01:54:37
used to um
01:54:41
disrupt not disrupt blur facts and how
01:54:44
we need to become much more interested
01:54:47
in debugging what we're told uh using
01:54:51
AI. By the way, part of that I have to
01:54:54
say is you have to learn to use AI uh
01:54:58
again not as a lazy person. So don't
01:55:00
have them do things for you. Have them
01:55:03
make you smarter. So instead of trying
01:55:06
to get the same task done with one
01:55:08
prompt, try to get a much more
01:55:11
interesting and demanding and
01:55:13
intelligent task done with more work.
01:55:16
So I've got two things so far which is
01:55:18
basically like lean into AI and the
01:55:20
second one is like lean into lean into
01:55:22
human collection con connection and and
01:55:24
lean into the truth. Don't be fooled by
01:55:27
the hype. Uh, you know, try to be more
01:55:30
informed, I think. And then finally,
01:55:33
ethics. Uh, if you want to, I know it
01:55:37
sounds
01:55:40
the world we live in sounds as if the
01:55:43
only way to win is to compete in
01:55:46
capitalism.
01:55:48
That's not the world I lived in. The
01:55:50
world I lived in, especially in my
01:55:52
Google years, was solve a major problem
01:55:55
and when you do, you'll end up making a
01:55:57
lot of money.
01:55:58
>> Are you optimistic?
01:56:00
>> I am optimistic about the future. I'm
01:56:02
very optimistic about the future. I'm
01:56:03
not optimistic about the present.
01:56:07
>> You're not optimistic about the next
01:56:08
decade?
01:56:09
>> Yeah, I'm not optimistic about the next
01:56:11
year, to be honest.
01:56:12
>> The next year,
01:56:13
>> for sure.
01:56:15
>> Why the next year?
01:56:17
>> Ah, come on, Stephen. You don't want me
01:56:19
to say it out loud. We're ruled by
01:56:21
maniacs.
01:56:23
Decisions are being made for the
01:56:25
absolute wrong reasons.
01:56:30
>> Very interesting time. Very interesting
01:56:32
time we find ourselves in.
01:56:35
>> I mean, honestly, if you're a video
01:56:36
gamer, this is the best part of the
01:56:38
game. Uh, it is a very very very very um
01:56:43
what's the word? It's the
01:56:46
it's the ultimate matrix of complexity
01:56:49
that I have ever encountered in my life.
01:56:52
>> Yeah, that's an amp description of how
01:56:53
it feels. Very complex. Things are
01:56:56
moving very quickly
01:56:57
>> and moving very quickly. So, so from one
01:56:59
side, you really need a lot of brain
01:57:01
resources to crunch all that's
01:57:02
happening. But you wake up tomorrow and
01:57:04
it's changed.
01:57:05
>> Well, the first time we um we spoke, we
01:57:07
spoke about happiness.
01:57:10
after this conversation. Stephen,
01:57:13
>> I was just wondering, you know, in uh
01:57:15
you you wrote a book about happiness,
01:57:16
which is
01:57:17
>> many books about happiness,
01:57:19
>> but I mean the uh soul for happy is the
01:57:21
book that I'm referring to. Here it is.
01:57:22
Yeah.
01:57:23
>> Engineering your path to joy. What a
01:57:24
fantastic book. I quote this book all
01:57:25
the time all around the world. Um, and
01:57:29
I'm wondering if any of the principles
01:57:32
that you wrote in this book about how to
01:57:35
live a happy life are more important now
01:57:37
in the world that we live in than maybe
01:57:39
when you wrote this book.
01:57:41
>> Oh, for sure. I mean, honestly, if I
01:57:44
wasn't living by this, I would have left
01:57:46
this world a long time ago and went to
01:57:48
an island somewhere. You see the
01:57:50
interesting side of happiness is that
01:57:52
it's not dopamine driven, it's serotonin
01:57:56
driven. Right? So, so my definition of
01:57:58
happiness is I'm okay with this world as
01:58:01
it is. I can affect it. I can change it.
01:58:03
I can engage with it. I can try to make
01:58:05
it better. Uh I don't have to accept it.
01:58:09
Uh but I'm okay with it. My starting
01:58:12
point is a bit stoic if you want. My
01:58:15
start starting point is I accept this.
01:58:18
this is my reality and now I can start
01:58:20
the work. This is very different than
01:58:23
anyone that's that basically looks at
01:58:24
the world and says, "Oh, this is a
01:58:26
horrible world. I don't want to be part
01:58:28
of it. I don't want to be engaged with
01:58:29
it." And so on and so forth. And I
01:58:32
genuinely have never been
01:58:35
calmer about that chaos. It's quite
01:58:39
interesting. And I, you know, uh my my
01:58:42
wonderful ex actually really really
01:58:45
helped me with that. There was a point
01:58:46
in time where we were having dinner and
01:58:48
I poured out crying from the sense of
01:58:51
responsibility I had for the world and
01:58:53
she looked at me so kindly so gently and
01:58:56
said hold on you know you I see that
01:59:01
you're trying but you can't actually
01:59:03
believe that you're responsible for this
01:59:06
and I think that completely flipped my
01:59:08
mind because in a very interesting way I
01:59:11
was thinking that all that went wrong in
01:59:14
technology is because of me,
01:59:16
>> right?
01:59:17
>> Why?
01:59:18
>> Because I contributed to building this.
01:59:20
Because I mean, when when Jeffrey Hint
01:59:21
and I, one of my favorite moments when
01:59:23
we were filming Chasing Utopia, uh, is
01:59:26
that Jeffree is very big for all of us.
01:59:28
We really think the world of him. And I
01:59:31
was telling him just as a sort of like
01:59:34
an older mentor if you want. I was like,
01:59:37
Jeffrey, do do you regret doing this?
01:59:40
You know, I genuinely believed when we
01:59:42
were building those things that we were
01:59:44
going to make the world better.
01:59:45
>> AI.
01:59:46
>> Yeah. And he said, well, yeah, I I too
01:59:48
was naive. I I thought that we I didn't
01:59:53
think he said I didn't think we will get
01:59:55
there so quickly before we figured out
01:59:57
the alignment problem, right?
01:59:59
>> The alignment problem
02:00:00
>> that AI has our best interest in mind.
02:00:03
>> Uh and and and I think all of us were
02:00:06
faced with that. All of us were faced
02:00:08
with that idea of we're building the
02:00:10
best thing ever for humanity and then
02:00:13
suddenly you realize oh my god in the
02:00:15
wrong hands it's the worst thing ever
02:00:16
for humanity and and I you know I have
02:00:19
to say I came to terms with this 2024
02:00:23
end of 24 uh that that yes I can try but
02:00:27
I accept that the world is what it is
02:00:30
and from that point of calm and stoicism
02:00:33
if you want I think I can have a much
02:00:35
bigger impact on the
02:00:38
We have a closing tradition where the
02:00:39
last guest leaves a question for the
02:00:40
next question left for you is what's the
02:00:41
legacy you want to leave?
02:00:43
>> Nothing at all. I've been asked this
02:00:46
question.
02:00:46
>> I get asked it all the time. So
02:00:48
>> I don't know why so many people are
02:00:50
asking am I going to go re anytime soon?
02:00:53
I don't know why so many people are
02:00:54
asking me that question. You see legacy
02:00:57
is a is a
02:01:00
I mean why what would I care if I have a
02:01:04
legacy if I'm dead? Like why that does
02:01:06
that even make any difference? Here's an
02:01:08
interesting thought for everyone. If if
02:01:12
karma is real, and I genuinely believe
02:01:15
it is, and if we're not just physical
02:01:18
beings, that we're physical and
02:01:20
spiritual, then I'd I'd rather keep all
02:01:22
of my karma for my spiritual side.
02:01:26
>> What does that mean?
02:01:28
>> I don't want anyone to remember anything
02:01:29
I ever did.
02:01:30
>> Yeah.
02:01:31
I I just want to leave a positive impact
02:01:34
on the world and take all of that as
02:01:36
karma for my next journey.
02:01:39
>> No, thank you.
02:01:42
>> YouTube have this new crazy algorithm
02:01:43
where they know exactly what video you
02:01:45
would like to watch next based on AI and
02:01:48
all of your viewing behavior. And the
02:01:49
algorithm says that this video is the
02:01:53
perfect video for you. It's different
02:01:54
for everybody looking right now. Check
02:01:56
this video out and I bet you you might
02:01:58
love it.
