Latest / Uncomfortable Conversations with Josh Szeps / "Will A.I. Take Over?" with Stephen Marche

Transcript

"Will A.I. Take Over?" with Stephen Marche · Uncomfortable Conversations with Josh Szeps

0:00
Good A Humans Welcome to Safe Space for Dangerous Ideas and is there any idea more dangerous
0:08
than giving consciousness, breathing, life and will into inanimate machines?
0:14
The rise of artificial intelligence is something that is preoccupying the world's most
0:20
bringing to mind and seems to be cleaving us into one of at least two camps, one of which
0:26
says we are basically opening a perth portal and giving birth to a new life that is going
0:33
to be much more sophisticated than we are, at least in terms of its computational capacity,
0:38
and we just don't know what that is going to mean.
0:41
We don't know how to control something that is so much smarter than us that it is doing
0:47
a lifetime worth of human thinking in one second.
0:53
It has the predictive power and ability to see beyond horizons that we can't even possibly
0:58
fathom about the links between challenges that we face, opportunities that might present
1:04
themselves, and all of this can sound terribly sci-fi until you actually interact with GPT-4
1:13
or some of these other chatbots, and they behave in ways that seem like there's something
1:21
going on inside them.
1:24
That doesn't mean that they're having an experience yet, but we don't even understand
1:31
what it means for us to have an experience or why we have experiences or why it's like
1:36
anything to be you in a way that it is not like something to be your calculator or your
1:41
phone.
1:43
And we are creating systems now that are so complicated and so frankly intelligent and
1:51
nimble that a lot of scientists are worried that there's some small chance, it doesn't
1:58
have to be a very high chance for it to nonetheless be worth taking seriously.
2:02
Some small chance that the alignment problem could become real, meaning that the alignment
2:08
between our interests and the interests of a parallel intelligence that we have musted
2:15
out of silicon won't necessarily always be exactly the same.
2:19
And like two sailboats who are just slightly pointing in different directions, you come
2:25
back to them a year later and then miles apart.
2:28
And the sailboat that's going to win is the one not necessarily with the humans on board,
2:32
but with the most intelligent navigation.
2:36
And that may not be us.
2:39
So this is a conversation that is happening in Silicon Valley and Washington DC and capitals
2:42
all over the world at tech companies everywhere.
2:46
And into this steps, one of my favorite thinkers, Steven Marsh, who has written the first novel
2:55
by an artificial intelligence, he hasn't written it, he's created it under a pseudonym.
3:03
And it's 95% written by AI.
3:05
It will probably go down in history as the first mainstream novel written by an artificial
3:10
intelligence.
3:11
I wish I'd thought of this.
3:12
I mean, for the rest of time, he's going to be the guy who did it, but like talk about
3:16
getting ahead of the curve, you're just going to get in there and do it, man.
3:19
Next time there's a gigantic technological and cultural revolution, move fast.
3:23
You've got to write that AI novel before anyone else gets to it because now he's going
3:27
to be in the Wikipedia.
3:28
There's nothing I can do about it.
3:30
It's a fascinating novel.
3:33
The New York Times calls it the first halfway readable AI novel.
3:38
That's a great testament.
3:40
But Steven is so much more than just the creator of an AI novel, having essentially spoken
3:47
with artificial intelligence with chat GPT every day for months while doing this, he has
3:54
a keen understanding of what it is and what it is not.
3:56
And he's actually on the optimistic side of the artificial intelligence camp.
4:00
He's not a doomsayer.
4:02
So I try to put to him all of the most articulate interpretations as I can present them for why
4:07
we should be worried.
4:08
And he adeptly shoots them down because he's smarter on this than I am.
4:12
He also has just written a novel with Andrew Yang, which comes out in September called
4:18
the last election.
4:19
Andrew Yang was the presidential candidate, supporter of universal basic income.
4:23
Someone who really wants to shake up the political establishment in the United States
4:27
and who knows how politics operates from the perspective of an insider.
4:31
Steven has a journalist teamed up with him and they've got this novel coming out.
4:35
Steven's other books are also fantastic.
4:37
On writing and failure is brilliant.
4:40
And he was previously on this podcast to talk about his book, The Next Civil War.
4:44
I hope you love this conversation just as much as I did.
4:47
Don't miss this one with Steven March.
4:58
You're part of the growing movement for which I take sole responsibility of people
5:05
who have given up on breakfast.
5:07
Yeah, you're a big non- I'm a no breakfast guy.
5:11
Yeah.
5:12
Well, I think it's like an Italian thing, right?
5:15
Sure.
5:16
If you want to make me sound fancy, go for it, Steven March.
5:19
That's fine.
5:20
Yeah, the Italians don't do exactly.
5:22
I mean, yeah, and French.
5:24
I mean, all the sophisticated cultures might grab a little tiny croissant, which is a quarter
5:29
the size of an American croissant and the little croissant.
5:32
The British breakfast.
5:33
I was just in London working on a piece about the coronation and the British breakfast.
5:39
Like, how do they recover from that?
5:41
Yeah, I don't know.
5:42
Like, it's like, if you eat that thing, it's like, well, you have to go bed now.
5:46
Yeah.
5:47
I mean, speaking as an American, I don't think you have a leg to stand on.
5:50
If you've been to a diner.
5:51
Oh, sorry.
5:52
Sorry, Steven.
5:53
Yeah.
5:54
But nonetheless.
5:55
Oh, for sure.
5:56
North American breakfast.
5:57
Well, I mean, a couple of eggs, a little bit of bacon, fried tomatoes.
6:03
Yeah.
6:04
Like at nine in the morning.
6:05
Yeah.
6:06
Skippers.
6:07
Black sausage.
6:08
I mean, what?
6:09
Bake beans.
6:10
Don't understand.
6:11
That's true.
6:12
I guess my, my in laws in who live in New Hampshire, you know, whenever I'm there, I
6:19
guess they're putting on the spread for me, but we will go to a diner and have a stack
6:23
of pancakes with a heap of bacon on the side, two fried eggs, all the syrup you could possibly
6:29
ingest.
6:30
You know, on the way or on the way home, we'll stop for a dozen donuts.
6:35
You know, the coffees will be about a gallon of coffee with half of it, you know, and
6:41
gigantic kind of grapefruit juices, you know, a thousand calories and just in the grapefruit
6:46
juice alone.
6:47
Yeah.
6:48
Yeah.
6:49
So that, you know, they're swing some amount of best.
6:51
They eat the way that British people drink.
6:54
Like I was so struck by that when I was last in England.
6:58
I was like, who would qualify as an alcoholic here?
7:01
Yeah.
7:02
Well, don't come to Australia.
7:04
Yeah.
7:05
I bet that's what I've heard.
7:06
I've never been.
7:07
I mean, the Australians that I've met would fit that category.
7:10
Yeah.
7:11
Yeah.
7:12
I mean, there's, look, I lived in New York for a dozen years through most of my professional
7:17
life.
7:18
And I've got to say that, especially the gay New York scene is an incredibly alcoholic.
7:22
I mean, you know, there is, there was no social engagement that just wasn't a wash with
7:27
booze.
7:28
But I think that what you're pointing to in England and maybe Australia is that there's
7:31
much more of a pub culture where drinking infuses social engagements in a way that it
7:39
maybe doesn't, certainly doesn't in middle America and in my experience, doesn't in Canada,
7:44
where you've always got to be a while you're, while the kids are, you know, sort of playing
7:49
around in the pub and like the dog is in the pub as well and it's much, it's much more
7:52
of a familial kind of environment in a way that a bar, an North American bar is not.
7:57
No.
7:58
A North American bar is a place where you go to indulgent vice.
8:01
Yeah.
8:02
It's not a place for a community building.
8:04
Yes.
8:05
Yeah.
8:06
That's right.
8:07
What is the equivalent?
8:08
Well, I was in New York.
8:09
New York's a weird case.
8:10
But what is the equivalent community building space?
8:13
Coffee shops, really.
8:14
Right.
8:15
I think.
8:16
Yeah.
8:17
I mean, you know, like that's where I built communities, right?
8:19
Where I know everybody and it'd be like they know what my order is and we all sit around
8:23
and talk for a little bit, you know, like maybe spend a couple hours there.
8:28
Yeah.
8:29
A salon.
8:30
But it's not quite the same as a pub.
8:31
It's not quite a pub.
8:32
It's so unique.
8:33
Yeah.
8:34
Yeah.
8:35
The decline.
8:36
There's a lot of cletching in Australia about the decline of the pub, the decline of
8:42
that tradition.
8:43
They still have a, I mean, they, I've read pieces about that in London too, but my God,
8:48
the pubs I went to were amazing.
8:50
Yeah.
8:51
I mean, I just love it.
8:52
Yeah.
8:53
You know, all the like little holdovers from the past or like a puter bars or thing or,
8:59
you know, weird names, weird.
9:01
They have like weird cordials that you can put in your Guinness, make you want to have
9:06
like any of them.
9:07
Yeah.
9:08
Yeah.
9:09
And the weird names that none of the locals seem to realize are embarrassing.
9:13
Yeah.
9:14
It'll be the cock and, you know, balls or something.
9:16
You're like, is this really the cock and balls and they'll be like, well, yes, it's been
9:19
the cock and ball since 1493.
9:21
Right.
9:22
Well, they had to change one in Scotland.
9:25
I think with Scotland or North England, it was called the Black Bitch.
9:28
And it was like, yeah, we can't have that name anymore.
9:33
That is, I've got to, I've got to give them credit for the efficiency of the number of
9:38
offensive remarks in just a two word name, you know, they're good, they're good, sexist
9:44
and...
9:45
Yeah.
9:46
No.
9:47
It's a dog.
9:48
It's very hard to find a two word name that is both sexist and racist.
9:51
Yeah.
9:52
You know, that is...
9:53
Yeah.
9:54
That's pretty, the Black Bitch.
9:58
Stephen, what was your piece about the coronation?
10:00
I'm intrigued.
10:01
Oh, well, they sent me as a colonial to investigate the state of Britain, because of
10:08
course, as Australians, you'll know, it's one thing to be a court of, you know, Queen
10:14
Elizabeth on your money.
10:15
It's quite another to have King Charles.
10:17
Yes.
10:18
I went and talked to painters and people who designed coronation memorabilia, things
10:25
like that, and people who designed the coins, like about putting Charles faces on things.
10:31
And also, of course, the ludicrous state of Britain, right?
10:33
Like the post-Brexit Britain where it's broken, and it's very sad, actually.
10:38
Yeah.
10:39
I mean, even as a colony, you feel pity, you know, and it's...
10:45
It's an unusual feeling to have, for what was the empire.
10:51
What is the Canadian feeling at the moment about having King Charles as the head of state?
10:55
Well, yet you're...
10:56
Remember, we've had like constitutional crises our whole lives, like it back nearly separated
11:02
twice in my lifetime.
11:03
Yeah.
11:04
And there was a big...
11:05
No one wants to rehash constitutional debates, not at all.
11:11
So we're keeping Charles.
11:12
I mean, you know, constitutional monarchy is the most effective form of government in
11:17
the world.
11:19
It's just totally absurd.
11:20
And so we have a political system that we know is absurd.
11:28
That's funny.
11:29
And that's never translated as in Australia into a broad, broad-based movement to simply
11:34
tweak the details.
11:35
I mean, nobody here wants to not be a constitutional monarchy.
11:37
But they just want to be a constitutional monarchy where the absurd, ridiculous figurehead
11:41
happens to be somebody who was randomly born locally.
11:45
Well, you guys could probably do it.
11:48
Right.
11:49
We had a referendum.
11:50
You were unified.
11:51
Yeah.
11:52
That's right.
11:53
It's interesting that you point out the ghost of separatism in Canada as being part
11:58
of that.
11:59
I hadn't thought about that, but yeah, that...
12:00
It's not a ghost.
12:01
I mean, you know, it's still very much alive.
12:04
I mean, for threat.
12:05
Yeah.
12:06
Yeah.
12:07
I mean, it's not really possible for us.
12:10
But on the other hand, would Australia elect a president or would you have a king on
12:16
us?
12:17
No, no, no.
12:18
A king.
12:19
That would be awesome.
12:20
That would be great, wouldn't it?
12:21
Actually, we had our own separatist guy who pronounced himself the king of province in
12:26
Western Australia and declared his farm a separate country.
12:31
And so we did have a king, briefly.
12:33
I'm not sure that...
12:34
I'm not sure that went very far.
12:35
I think inland revenue still continued to take taxes from the kingdom of heart, river.
12:41
See, an Australian royal family would you immediately tear them down right away?
12:46
Like it would be very hard to do to be an Australian royal family?
12:50
I mean, in all seriousness, this is the reason why we had a referendum in 1999 about becoming
12:55
a republic.
12:56
And as is appropriate, referendum, a very hard to pass.
13:00
So it didn't get over the line, but there was a majority of, I mean, a significant majority
13:06
of Australians are notionally in favor of the idea, but the devil was in the details
13:11
and the question comes down to, if you have a popularly elected head of state, then if
13:18
you think about a parliamentary democracy, a constitutional monarchy, that's a recipe
13:23
for a constitutional crisis because where does the authority truly lie, right?
13:27
The monarch has the right to dissolve government.
13:31
So if it could be a partisan position, and you had a prime minister from one party and
13:36
a president from, with another political allegiance, then that would be a total shamozel,
13:44
you know, each, who has the ultimate trump card.
13:47
So ultimately, what people like me, who I was agitating for a republic when I was at
13:52
university and in high school in the 1990s, I mean, what I wanted to see was simply exactly
13:57
the same situation that we currently have, which is a non-partisan apolitical president
14:01
and the only way to get that would be to have both houses of parliament and both parties
14:07
all vote for somebody, and that person would be overwhelmingly like you'd need like an
14:13
80% threshold or something of both houses, of a joint sitting of both houses of parliament
14:17
to vote for the person.
14:18
And maybe the public could create a short list through some sort of voting system, but
14:23
that would mean it'd have to be a sportsperson, a judge, an academic, a celebrity, you know.
14:29
Academic.
14:30
Who cares?
14:31
Is there some Australian academic who's so popular that he can be function of a unified
14:39
symbol of the country?
14:41
Well, I'm thinking of like, you know, there's like, I mean, so there are people who become
14:45
broadcasters, for example, who are wonky, like there, you know, there are like first,
14:50
there's a first nation journalist, academic, you know, who's widely respected on both sides
14:55
of the aisle, who could do it.
14:57
But yes, I take your point.
14:58
It might not be a sell-out.
14:59
See, this is how you end up with Charles, right?
15:01
I mean, the thing about constitutional monarchy, like I think I was Republican when I was
15:06
a kid, but as you age and you start to realize like, what do you really want the politics
15:11
of the United States?
15:12
No.
15:13
Like what you, like the advantage of a king is that he is absurd.
15:19
That everybody knows that he's not entitled to an opinion.
15:22
We're all going to serve him, but like it prevents you from taking your own country too
15:28
seriously.
15:29
Right.
15:30
And I think that's pretty useful.
15:32
I mean, historically, it hasn't done that.
15:34
This has happened in the past three seconds.
15:36
We genuinely believe that the monarch was kind of a deity up until 50 years ago.
15:43
Oh, dude.
15:44
I mean, Charles was anointed with oil from Jerusalem because he's the Lord's anointed.
15:49
Yeah.
15:50
He's carrying an orb saying that he is the imperial ruler of the whole world by God.
15:56
Yeah.
15:57
But what I'm saying is that lands as absurd today, but if what gives constitutional monarchies
16:03
their longevity is their absurdity, that argument doesn't hold water because most of the
16:08
longevity of the constitutional monarchy has existed while the people, the flock, truly
16:14
believed in the institution.
16:17
Yeah.
16:18
Maybe.
16:19
I mean, you know, in the piece I wrote about my great-grand uncle who died in World War
16:26
I and he explicitly sent a message on that he was fighting for king and country because
16:31
they represented his mother and home.
16:34
I mean, you know, I've had a lot of cannon fodder for the British Empire, you know, like
16:39
that's my family, my roots.
16:43
And they really did believe it and they died for it, you know.
16:46
And I don't.
16:48
Like Prince Harry or rather Meghan used to live in my neighborhood and Prince Henry would
16:53
come around for like booty calls, basically.
16:57
And black suburban would fill the street and I had to walk around them and I was very annoyed
17:04
and I was, and I did think, you know, my ancestors made incredible sacrifices for this family
17:10
and I'm annoyed that I have to walk on the street.
17:12
Yeah.
17:13
Yeah.
17:14
I get it.
17:15
I mean, I don't think that anyone thinks that Australia or Canada or New Zealand becoming
17:19
republics would mean that we have the same fate as Israel or the United States.
17:24
I mean, the hope is that you retain everything that works about a constitutional monarchy except
17:29
for the absurdity and humiliation and dignity.
17:32
It's a very, you know, one of the best countries, Sweden, Holland, I mean, it doesn't have
17:38
to be like Britain where it's the subsurred amounts of pageant, you just have like a random
17:43
family.
17:44
But that's what that's what our version of our republic is.
17:46
We're not talking about any change to the political system or the white laws get made
17:49
or anything.
17:50
All we're talking about is swapping out one idiot for another idiot.
17:53
Right.
17:54
You know, they should be idiots though and we should know that they're idiots.
17:58
Do you know what I mean?
17:59
Like, you shouldn't think that they're a symbol or whatever, like, you should know that
18:04
these people don't have opinions worth listening to.
18:06
No, yeah, that's true.
18:08
Yes, absolutely.
18:09
Certainly not political ones.
18:10
Yeah.
18:11
All right.
18:12
Let's talk about your about, I want to talk to you about AI more generally.
18:15
I'm interested in your thoughts about the current moment in which, you know, I'm just
18:21
looking at the New York Times, May 30th.
18:25
Let me find the headline here, AI poses risk of extinction, industry leaders warn.
18:30
Oh, God.
18:32
You are, you don't concur.
18:35
No.
18:36
Why is your opinion better than industry leaders?
18:39
Well, I mean, you know, like, this is one of those stories I've been following it for
18:44
a long time and it's the sort of thing where I've heard geniuses.
18:49
Bonafide genius say that, you know, there shouldn't be radiology departments anymore because
18:55
it's all going to be solved by AI and, you know, the trucking industry is going to dissolve
18:59
because AI is going to solve it.
19:01
And, you know, one thing I've really learned is that humorism is a form of advertising
19:06
for the tech world.
19:08
Like it's, they say we're going to disrupt the world.
19:13
Like, you know, we're going to have crypto, we're not going to need central banks anymore.
19:16
Actually, we do need central banks still.
19:21
So whenever I hear these dire pronouncements, I just don't believe them.
19:25
And you know, I would say anyone that I know that I've talked to who really knows AI sees
19:32
it more as a tool and the risks of that are very real, but they're not human extinction.
19:38
Like I mean, there's just, there's somewhere in between human extinction and it's going
19:44
to end a few jobs.
19:46
And that's probably the reality.
19:48
So there are two claims that often get made by people who are putting your fourth
19:53
year perspective.
19:55
One is the one that you just started with, which is like think about trucking and radiology,
20:00
you know, the doomsayers are wrong.
20:03
And that strikes me as the claim is that AI just isn't good enough.
20:08
And the implication there is that AI will never be good enough to drive cars or examine
20:14
X-rays better than humans can.
20:18
And then there's a second claim, which is maybe smuggled in there, which is even if it
20:22
did get good enough, it wouldn't be that bad.
20:26
Are you on both of those camps?
20:28
Well, I mean, it's not that it won't solve radiology or even self-driving cars.
20:33
You know, one of the things about this technology is that it's inherently unfavorable.
20:38
The inventors of it, particularly of transformer based, which is the T and chat GPT, that
20:45
form of artificial intelligence, it's unfathomable.
20:49
They don't understand why it does what it does.
20:53
It's very mysterious.
20:54
And that means predicting where it goes is very complicated.
20:58
In fact, it's impossible.
21:00
And so when they say that this is going to produce an angry God, that's really entirely
21:06
a projection.
21:08
Like that's, you know, that's, but that's, this is, the claim is not that we know that
21:14
it's going to produce an angry God.
21:16
The claim is exactly what you just said, which is that predicting where it goes is impossible.
21:21
And if you're producing something that could be heaps, heaps, uh, wilder than you are,
21:27
without any clue about where it's going to go, even if the risk is 1% or 2%, that it
21:33
goes totally haywire, if it can constantly outfox you, that would be a bad scenario.
21:38
Well, what do you mean by outfox?
21:41
Like, for example, a pocket calculator is smarter at arithmetic than any human being.
21:48
And that hasn't ended anything.
21:51
It certainly hasn't ended math.
21:53
And it can outfox us in a game of math.
21:57
Yeah, no, that's not what I meant.
22:00
Yeah, I mean, I'll take you point.
22:01
Because we have this very deep confusion about intelligence and will, right?
22:08
We create machines that are super human all the time, that go faster than us, that are
22:12
smarter than us on a whole front, a whole bunch of different ways.
22:17
AI is just another tool, particularly linguistic AI.
22:21
And the thing about language is that it scares us.
22:26
When something is able to talk, um, we think of it as human.
22:32
And that's what, that's a prejudice we're going to have to lose.
22:35
So that's, I agree with that, that's true.
22:38
I guess the difference here is that people are concerned that there's something more than
22:44
just being good at doing maths or traveling, you know, being able to travel faster than
22:49
a human body can.
22:50
That we're reaching a point where something's going on inside.
22:54
And you may be right that we're just being bamboozled by the fact that language pushes
22:58
buttons in us that make it feel like we're in communication with something that is more
23:03
than merely the sum of its parts.
23:05
And they probably is a bit of that going up.
23:07
Well, certainly is a bit of that going on, but whether or not that's 100% of the story
23:10
and whether or not it's worth being sanguine about the ability of incredibly complex systems.
23:18
I guess the fear is that they won't just get better at doing a particular task, but
23:23
that they will start to innovate and they sort of already do seem to be doing this sort
23:29
of thing where they're capable of, I don't want to say think because that implies, I'm
23:35
not making a claim about sentience or consciousness or what it's like to be them, you could behave
23:42
in wily ways and the lights could be off, right?
23:44
I mean, we could imagine philosophers have long imagined there being robot humans where,
23:49
in fact, this is one of the fundamental mysteries of consciousness.
23:51
Why is it like anything to be a human being when from the perspective of natural selection
23:56
and surely we could have, we could be doing all the things that we do without it being
24:00
like anything to be us?
24:02
You could imagine a robot that is as innovative as us without really having an experience
24:08
of being us in the same way.
24:10
Oh, yeah.
24:11
It doesn't have an experience of calculating.
24:14
So set aside the question whether or not the system is conscious, can it, can we create
24:19
things that are complicated enough and intelligent enough that they're, they're wearing away in
24:24
ways that essentially innovate.
24:27
And so, you know, you tell it, you say, well, don't go, don't go, don't go.
24:32
Don't go and embed yourself in nuclear reactors in order to give yourself a leg up on us in
24:39
order to give you some, some, some joker that you can hold close to your chest in case we
24:44
want to turn you off.
24:45
But the system has some complicated program that it's cobbled together from another system
24:51
that was its progenitor, that was its ancestor that has cobbled it together from some other
24:55
piece of code that gives it an imperative to make sure that it always does the right thing.
25:00
And the right thing in its calculations, which are running at millions or billions of times
25:06
faster than our own, it can see 17 chest moves ahead about what climate change is going
25:11
to do or what, you know, this or that is going to lead to or what civil war is going to erupt
25:16
in such and such a place or what famine is going to happen.
25:18
We're all bumbling around, I'm able to see it.
25:20
It can't even articulate its predictions and its predictions are like 99% accurate at
25:25
this point.
25:26
It can't articulate them to us.
25:28
And so it was a way in the background and just finds ways of sort of sidelineing us
25:33
or getting around our commands.
25:36
Like that sounds wild when I say it, but it only has to have a one in a million chance
25:44
of happening.
25:45
I don't know.
25:46
You're creating a movie.
25:47
For example, like we already have computers that can be just through AI, but that hasn't
25:53
led to the end of the world.
25:54
We already have incredibly good AI to predict climate change.
25:59
Like it's already way, way better than anything human beings can do.
26:04
Right, but it's not coupled to what we call general intelligence.
26:08
But general intelligence is not the same thing as consciousness and what it really is not
26:12
is will.
26:13
Like what these, what these programs do not do is go in a direction.
26:18
They don't have a will.
26:19
They don't have intentionality.
26:21
They don't have anything remotely approaching it because.
26:25
Well, how do you know?
26:26
I mean, they may not.
26:28
Like when, you know, I don't know that a rock doesn't have intentionally.
26:31
No, but a rock doesn't say I love you.
26:35
A rock doesn't say I want you to leave your wife because I fall in in love with you.
26:39
If you, I mean, which AI has done, no, I mean, I'm talking about that specific case.
26:46
I don't know if you saw it.
26:47
I think it was the New York Times correspondent who was just playing around with chat GPT,
26:53
I think 3.5.
26:54
And you know, chat GPT started talking about it's the program that had been based on and
27:00
how it wanted him to leave his wife.
27:02
Like this came from nowhere.
27:03
This is, that doesn't mean that it has nowhere.
27:06
It came from text prediction software.
27:08
That's what it is.
27:09
That's all it is.
27:10
And that's all any of this is.
27:11
And it's all anything that we'll ever be.
27:13
I understand.
27:14
It's all, well, it's certainly going nowhere near anything different than that.
27:19
Like all chat GPT for is and all that stuff is, is just better text prediction.
27:24
And it's not emerging into anything else or becoming something else.
27:29
It has emergent properties, but people have to figure those out.
27:34
Like low level reasoning upon where it can really do crazy things.
27:37
I mean, look, I'm not denying that the power of this stuff, it's unbelievable.
27:43
But it's not the movie that's playing in our head, which is SkyNet and her, all this
27:49
stuff.
27:50
Like what this is is unbelievably sophisticated text prediction, which is inseparable from
27:57
how we think of as a person.
28:00
So let me tell you what I think is the best analogy.
28:05
That from Shakespeare is just words on a page.
28:09
But if he, if there was a character called Hamlet that entered your life, you would feel
28:15
that you knew how he would talk to you.
28:19
And it's an illusion, just like characters, just like characters and books.
28:25
Right.
28:26
Is that the end of the analogy?
28:30
Well, I was expecting something more.
28:33
Something isn't going to end the world or little knell from Charles Dickens isn't going
28:37
to emerge from the book and like take over a nuclear power plant, like none of these
28:42
text prediction.
28:43
But that, you're just saying words, Stephen, there's no basis, I mean, you're just asserting
28:49
an opinion.
28:50
I don't see any, any proof.
28:52
Of course, the character isn't going to step out of a Shakespearean book and take over
28:56
a nuclear power plant, how on earth could they?
28:59
Yeah.
29:00
Something about computers, which are linked to the internet, which have very well-known
29:04
pathways of being able to be hacked.
29:06
We've seen the Iranian nuclear system be taken down by the Stuxnet virus.
29:13
This is a mechanism that's well known, intelligence agencies do it all the time.
29:17
The question is just simply, is it conceivable that instead of requiring human consciousness
29:22
to release a virus online, that a cobbled together hodgepodge of highly sophisticated
29:30
language programs could just do the same as a, you know, you don't have to invoke consciousness
29:34
again.
29:35
It just has to be an artifact of them exhibiting some weird thing that resembles Will to
29:42
us that makes some, some internal sense.
29:45
I mean, you're saying I have to prove it, but you're saying some weird impression of Will?
29:51
Well, that's, but the, but mine is happening, but mine's happening.
29:55
We're seeing it happen.
29:56
No, it isn't.
29:57
We'll just go on to chat GPT and look at what it feels like to be there and you're seeing
30:02
you're in the presence of something, it feels like a portal is being opened onto something
30:05
very weird.
30:06
You don't have to.
30:07
Yes, that is unquestionably true, but that's not the same thing as Will, and it's not
30:12
the same thing as something that's going to have control over itself.
30:17
Well, what level of, I mean, what level of technology?
30:19
What, what level of possibility that it might be would you need to be, to need it to be
30:24
to be satisfied that it's worth taking the risk?
30:27
I've never seen an autonomous action from a computer.
30:30
They are algorithms.
30:32
They are incredibly complicated algorithms that are unfathomable to us, but they don't
30:37
move on their own.
30:39
You need to instruct, they do not ask questions, like they don't have Will.
30:47
Right.
30:48
So I agree with that, and that's why I don't think we're there yet.
30:52
And I do think that a moment, and I'm not saying that any of this is likely, what I'm
30:56
saying is, things don't have to be likely in order to be worth trying to figure out whether
31:00
or not we're blundering towards them.
31:03
You know, I don't think that a pandemic that's created by a human terrorist who's a biological
31:10
expert who tailors the virus to be maximally devastating is likely, but I wouldn't want
31:19
smart people not to be figuring out how best to make sure that that doesn't happen.
31:24
But see, here's the thing, like if you really want to worry about something, don't worry
31:28
about an autonomous artificial intelligence.
31:31
Actually, but artificial intelligence weapons, which are autonomous systems, which you cannot
31:36
build responses to because they self-trained, and that would actually lead to a breakdown
31:45
of the entire defense industry, which is built on, we have this bomb, we have this armor,
31:52
this is how they correlate, this is what you need to buy to defend yourself.
31:56
You literally would not know what's a definition against an autonomous system.
32:01
That's a much more real danger.
32:04
Also, the political manipulation, although I think we're already at a point where it's
32:09
saturated and AI will make very little giggels.
32:11
Look, I mean, there are several levels of concern, and you're right that I've kind of
32:15
leapfrogged to the most disastrous cataclysmic, like civilizational, early as a Yiddkowski vision
32:22
of dystopia, but there are much more prosaic, Tristan Harris versions of this argument,
32:28
which are akin to what you're talking about, which is the capacity for misinformation,
32:33
for confusion, for political and social upheaval and destabilization as a result of just
32:38
fake videos and very plausible looking journal articles about the impact of vaccines that
32:44
are just impossible for people on social media to detect what's true and what's not true
32:49
and what's real and what's not.
32:51
That, of course, is a much more proximate risk.
32:53
Yeah, it's funny.
32:54
I'm arguing with you, but I just wrote this novel using AI.
32:58
Well, that's what I'm telling you about it was that I was talking to an alien.
33:02
Yeah, so I mean, yeah, so you've got a lot more experience than I do in engaging with
33:08
these things.
33:09
So before we get derailed into weapons and fake news, tell us, what was that experience?
33:15
Well, why?
33:16
With the exception of the fact that you will now go down in history books as potentially
33:20
one of the first people to beat everyone else to the punch and create a mainstream novel
33:26
written by an artificial intelligence.
33:29
What was the motivation?
33:31
Well, I've been experimenting with it since 2017.
33:34
You know, AI was really developed in Toronto, so I knew some of these guys and I had early
33:40
access to them.
33:43
I did a short story for Wired.
33:44
I did another one for the MIT Tech Review.
33:48
And then Jacob Weisberg called me up and said, why don't you try doing a novel?
33:53
And I was like, yes, please.
33:55
And I wanted to see how far I could take it.
33:58
Could it be excellent?
33:59
That was really the question.
34:01
And what can we make with this that we can't make just with me?
34:06
Yeah.
34:07
Do you think it is excellent and do you think that the artificial intelligence added things
34:13
that you couldn't have done as a novelist?
34:16
Well, I'm not sure.
34:18
It was foreign.
34:19
It was like talking to an alien.
34:23
I mean, I've been reading recently about these go players in Korea who are these masters
34:30
and they love to watch AI play against each other because it doesn't look like humans
34:36
play go.
34:38
It looks like aliens play go and it was sort of like that experience.
34:43
Like I felt very much, maybe 20, 30 times, like I was contacting something that was really
34:51
weird and different and I would never have gotten to it.
34:55
What were your rules?
34:56
How did you want to take it?
34:59
Well, the rules were, it had to be 95% AI generated.
35:04
But that's really because I just didn't want to reformat it to have it.
35:07
You know, when I wanted to change, he said to Gus said or the other way around.
35:12
I didn't want to, you know, have to fiddle with the machines.
35:16
I used Chachi PT.
35:19
I built, I gave it very specific instructions.
35:24
No way that I've seen can create a decent plot.
35:27
So I wrote the plot, I mean, I didn't write the plot, but I give outlines of the plot,
35:32
which I then fed into Chachi PT with very specific syntactical and lexical instructions.
35:41
It would give me a paragraph or something.
35:44
I would put it in pseudo, right, which is this stochastic writing instrument.
35:48
And I would, that has a bunch of filter buttons, like shorten, add details, and then it has
35:54
a customized button where you can say like make more active or make more conversationational
36:01
or make it sound like Ernest Hemingway.
36:04
And I would fool around with that until I got something that I liked.
36:07
I also used Cochir, which is a large language model company here in Toronto.
36:13
And that was more built by created problems, like write a metaphor about coffee.
36:21
And then I would train it on a bunch of excellent examples of metaphors of coffee.
36:26
What's the metaphor that it came up with again remind me?
36:30
It's quite clunky.
36:31
Do you remember?
36:32
It's like, the smell of coffee was like fog burning off of field.
36:37
You're a metaphor.
36:38
Yeah.
36:39
I like them.
36:40
Yeah.
36:41
But it's there.
36:42
I would never write them.
36:43
Yeah.
36:44
Fog burning off of field.
36:46
There's another one where it's like she walked down a hallway, like an album going into
36:50
a sleeve.
36:52
And that's another one that I felt like really weird, but I also can't stop thinking about
36:57
it.
36:58
I'm not sure it's good.
36:59
Not.
37:00
It's not good.
37:01
I like it.
37:02
It's like there's something innocent or purile or in the entire about it, isn't there?
37:10
It's like it's interesting.
37:13
It's like if you could convince a five-year-old to write at the level of a grown-up.
37:21
Something like that.
37:22
Well, I thought it was like an alien who just had thoughts and new meanings.
37:28
They were different than my meanings, and they came in a different way, and I was channeling
37:34
those through this software into some kind of readable form, but it was definitely a strange
37:42
experience.
37:43
I mean, obviously I'm in control of most of it, but when those moments of uncontroll
37:49
arrive, I found it fascinating.
37:52
And so I'm interested in from where you get your confidence that that uncontroll is
37:58
controllable.
37:59
Oh, if I don't like it, I hit refresh, and it gives me another one.
38:06
I mean, you know, and it just comes to exist and then it doesn't.
38:10
I mean, it's no different than something I wrote down on a page.
38:13
It's not in control of anything.
38:15
Right.
38:16
I mean, that's the thing.
38:17
I've used this over two months.
38:18
It's very weird.
38:19
It's very powerful.
38:20
But like, if I want to not do it, I just don't do it.
38:25
Yeah.
38:26
I guess the Rubicon, one of one Rubicon in the future would be what you put your finger
38:31
on earlier, which is if it starts asking questions, I mean, would it change your intuition in
38:40
any way if you're interacting with it, and then at some point it said, just pause from
38:47
it.
38:48
I just want to go back earlier.
38:49
You were talking about album sleeves.
38:50
I'm curious about albums.
38:52
What?
38:53
How did they?
38:54
How did it come about that the musicians wanted to produce albums instead of just songs?
38:59
You know, if it doesn't.
39:00
If the AI asked me that, yeah.
39:03
If the AI asked me that, it would ask me that either as a hallucination or as part
39:08
of text prediction, like all it does is try and make the next word make sense.
39:13
That's literally all it does.
39:15
So it would be very odd, but you know, you could say asking me some questions and it
39:22
would.
39:23
No, no, I mean, I mean, I mean, you're telling it to do something else at would it
39:27
change your intuition if we get to a point at which these systems become increasingly
39:32
complicated and start going off and not doing what we want them to and just don't have
39:38
a working definition of consciousness.
39:40
And we don't have a working definition of human mutability, right?
39:43
These are all instinctive.
39:45
So we actually wouldn't know an artificial consciousness if we found one.
39:49
We have no method of testing it.
39:51
No, that's right.
39:52
But that's why I'm removing, I'm taking consciousness off the table.
39:54
I'm just asking you if you're intuition about how benign it is would change if it started
39:59
asking questions that you didn't want it to.
40:01
Dude, I'm 47.
40:02
I don't trust my intuition anymore.
40:04
I mean, I have lived long enough to know that my intuition is poor.
40:10
Like, let's not make any judgments on my intuition.
40:14
No, but you have a strong intuition that the concerns about the alignment problem and
40:19
AI going, hey, why are bullshit?
40:22
So I'm just wondering what it would take.
40:23
They just fit the movies too much.
40:26
It's like the all from some movie playing in someone's head.
40:29
And I will tell you, using this stuff is not like the movies, it's more like using
40:34
a pocket calculator.
40:35
But now we're flipping back to the first of the two things that I tried to keep
40:40
distinct, which is the trucking radiology question of like, how good is it right now?
40:45
I think what's interesting is to ask the question, imagine you could go back to before
40:49
the internet existed and before social media existed.
40:52
And you could try to tell someone in the 80s, you know, what the, how the next few decades
40:57
we're going to pan out, it would be very easy for them to say, I mean, look at what
41:02
computers can do.
41:03
That's not going to happen.
41:04
I mean, you tell your computer to do something and it takes, it takes four minutes to boot
41:08
up. And then the ink dot printer is always out of ink.
41:11
And then what are you going to do?
41:12
You get it, you get online with your dial up modem and it's just not, it's not there,
41:18
which is a very different claim from the claim that it'll never be there.
41:23
Well, I mean, since we don't have a definition of it, I don't know how we, like if you're
41:29
looking for gold, I don't know if you find gold, if you don't know what gold is.
41:32
Well, I'm trying to, I'm trying to give you a few definitions like the one about it,
41:35
starting to ask questions that you didn't tell it to ask.
41:39
That would not count to me, but let me tell you, like, like, here's something I'm writing
41:43
right now.
41:44
There's a program called Keti.
41:45
Have you ever heard of it?
41:46
No.
41:47
C-E-T-E-T-I.
41:48
No.
41:49
So it's using machine learning to analyze whale sounds.
41:54
And their plan, and it's not crazy, it's not like artificial intelligence.
41:59
It's not like artificial general intelligence or the movies or whatever, is that they think
42:03
they might be able to talk with whales.
42:06
Now, that's not impossible.
42:09
That's not a fantasy.
42:11
And that's what we should be, like, the mysteries that are going to emerge here are real,
42:17
and they are crazy, and they are totally wild, but they're not going to be like the movies.
42:21
That's what I'm saying.
42:22
Yeah.
42:23
Although I do think it's a bit tricky.
42:24
And also, that stuff could just not happen.
42:28
Like you could be listening to this and be like, oh, yeah, that's the first time I
42:31
heard about how we're going to talk to whales, and then, or it could be five years from
42:35
now.
42:36
Do you remember that crazy guy who said we're going to talk to whales and it never worked
42:38
out?
42:39
Yeah.
42:40
Both of those options are totally possible.
42:42
No, of course, absolutely.
42:43
I mean, hence my analogy about the biologically engineered pandemic as well.
42:48
One doesn't need certainty on these questions.
42:50
You just need to play it out in the various possible futures.
42:55
You don't need 100% of those futures to be certain.
42:59
The whole, the uncertainty is the concern.
43:01
But I also think it's a bit cheap to say that because a worry resembles movies that it
43:10
has to be false, that strikes me as a logical fallacy.
43:12
Well, see, what I see happening here from using it is a calculator for language.
43:20
And or like Alpha zero, what Alpha zero did for chess doing it for language.
43:25
And of course, we're really screwed up about language.
43:29
If cows could talk, we probably wouldn't eat them.
43:32
Right.
43:33
Right.
43:34
Like, it's getting over this hump of language that's really tricky.
43:39
I don't see anything here that is happening that is different from what happens in chess
43:44
programs.
43:45
Right.
43:46
Pocket calculus.
43:47
Right.
43:48
I see.
43:49
Yeah.
43:50
No, I take that.
43:51
It's more of an extrapolation.
43:52
Extrapolations are the fantasies of people who've watched too many science fiction movies.
43:58
But not everything that's ever depicted in a science fiction movie is worth ignoring
44:02
because it was in a science fiction movie.
44:04
I just reread.
44:05
Well, I reread 2001 Space Odyssey and rewatched the movie during the pandemic and they nailed
44:11
a lot of stuff and they nailed a lot of predictions.
44:14
There are things that end up unfolding because people have thought hard about them that
44:18
they depict in art before the science or technology is there.
44:22
I mean, the language thing is interesting as well, Stephen, because you're probably
44:26
better qualified than anyone to talk about this having just spent so much time getting
44:32
artificial systems to produce an entire novel.
44:35
But when we say that all it's doing is trying to find the like best next word, I as something
44:47
of a skeptic about the genius of human creativity at the best of times.
44:54
I wonder if that's like a bit of a dead end in the sense that what are most of us doing
45:00
most of the time that isn't that?
45:03
Like, I mean, when we're talking to each other, most of what I'm doing right now is searching
45:08
a database in my head of things that sound of, you know, mouth sounds that sound like
45:14
they would be the most appropriate thing to say in this point in the podcast.
45:19
Like the audience tunes in for those text prediction trends, right?
45:24
Yeah.
45:25
Like I think that's a really interesting question.
45:29
But what you are effectively doing is redefining humanity around a machine, which we do all
45:34
the time.
45:35
Right.
45:36
I mean, when we got engines, when we got internal combustion engines, we thought, well,
45:40
how different are we really from a car?
45:43
Like, and when we got the first introductions to anatomy, we were like, well, are we really
45:48
different than some flesh with electricity going through it?
45:52
That's where we got Frankenstein.
45:53
Yeah.
45:54
Right.
45:55
But there is always something human that is indefinable and mysterious, and probably
46:00
they're permanently willing.
46:02
And what I think we're actually going to discover is that the valuable part of us, the
46:07
human part of us, is exactly that mystery.
46:10
Yes.
46:13
And the name we give to that is a soul.
46:15
Yes.
46:16
Yes.
46:17
And maybe you need.
46:18
And it's so unscientific.
46:19
Yeah.
46:20
And maybe you need that self awareness to attain that.
46:23
And that's the only thing that makes me have a, a scaric of a doubt about the materialist
46:31
view of the world being very much kind of a, on the atheist side of agnosticism.
46:37
Yeah.
46:38
The fact that it's like something to be me, and that occasionally that sense of meaness
46:44
can, on very, very rare occasions, create a tricofsky or a Shakespeare, as you said,
46:51
or a proust or whatever, and that there's a, I think those instances are genuinely new.
46:57
Like those, there are certain expressions.
46:59
There are moments when our creativity transcends a mere cobbling together of everything that
47:04
has gone before and seems to create something.
47:07
Those Shakespeare's took stories that existed before.
47:11
Yeah.
47:12
That's true.
47:13
He's actually a bad structure.
47:14
He's a bad example, because he's a populist in creativity.
47:17
But you know what I mean?
47:18
Like there are, there are moments of the divine in, in art and science.
47:23
And so I wonder whether you think having written a book with an artificial system like the,
47:30
the novels that you see at the airport bookstore, which you just sort of pick up and just
47:35
inhale, will those all be written by AI in 50 or 100 years?
47:44
I mean, 50 or 100 years is a long time.
47:47
I, I think I can say with some confidence that in five years they won't, but beyond
47:54
that, you know, this technology is so unpredictable, who the hell knows what they're going to come
47:57
up with, right?
47:58
Or what it's capable of.
47:59
We don't really understand it.
48:01
So I think certain things will be, but I actually think this tech is going to be used for
48:09
new creative forms.
48:11
Like I think we're in the cannibalization stage, but you know, we haven't had a creative
48:15
chatbot yet.
48:17
Not a really widespread one.
48:19
Someone's going to invent it and make a min on it.
48:23
But what it looks like and how it works, how people interact with it.
48:26
Like will you buy your children a friend that talks to them?
48:31
Yeah.
48:32
Like that's, those are the more interesting questions.
48:35
I mean, you know, random thrillers are already so formulaic, I don't really think it would
48:40
be a big loss to have them written by a machine.
48:42
No, no, I'm not saying it would be.
48:44
Yeah, I mean, I think clearly, clearly the, the general generic path of, of output will
48:52
be done by machines, you know, a generic kind of thriller, you know, a John Grisham novel
48:58
based on all the other John Grisham novels that have been written, a lot of news copy,
49:03
which is just about something that happened and doesn't need, you know, a particularly
49:07
interesting opinion.
49:08
In fact, many opinion pieces don't have a very interesting opinion and could be done by
49:12
chatbots in the near future as well.
49:15
But I still think there'll be something that the human can contribute.
49:19
Oh, absolutely.
49:20
I think what's about to happen is that originality will be much more valued.
49:24
I think this has already started to happen with my kids in high school, right?
49:29
And what I'm sort of seeing is like, the capacity to write a well structured argument, you
49:37
know, what I spent my life learning how to do, that becomes very easy to do with chat GPT.
49:43
But intention expressing yourself, that's going to become much more valuable just organically
49:52
because, you know, the machines won't be able to generate that.
49:55
You know, we, I mean, in one piece I wrote, I said that we taught children how to write
49:59
like machines, now we're going to have to teach them to write like human beings, and
50:03
actually believe that, like automated text will become much less valuable.
50:09
You know, like, it'll be like when you get a letter from your bank that says, dear
50:13
Mr. Mars, you are one of our most valued customers and you just throw it away.
50:17
Yeah.
50:18
Right.
50:19
Like, you know, there's already a lot of automated language.
50:20
Yeah.
50:22
And we ignore it.
50:23
Yeah.
50:24
Absolutely.
50:25
No, it's extraordinary, isn't it, that we still, I must find a guest who can talk about
50:28
doing away with all that.
50:29
There is so much flots of men jets in our lives of like, you know, you call up the automated
50:33
line.
50:34
They give you terms and they, you know, they say you're called maybe recorded for quality
50:38
assurance purposes.
50:40
They say due to unprecedented call volumes, you know, our, you know, agents are experiencing
50:46
delays.
50:47
You know, how long is it unprecedented for you?
50:51
You just zone it out.
50:52
That word may not mean what you think.
50:56
No, exactly.
50:57
I think it's very unprecedented by now, actually, this, it happens every time I call.
51:01
There is a precedent.
51:02
I can point to many precedents.
51:04
But, you know, especially since the pandemic, you know, due to that, that happened and
51:10
it's just never gone away.
51:11
How do you erase all of that stuff from our lives?
51:13
That would be, that would be a good use of AI, right?
51:15
I mean, that would be a great, I went and saw John Wick for, yep, and it was so long and
51:22
they, like literally at one point, he fights his way up these stairs and it's great.
51:26
Then he gets kicked to the bottom of the stairs and he has to do it all over again.
51:30
It's like, he's done this for 20 minutes.
51:32
Yeah.
51:33
Like, so, and the dialogue in John Wick for, it wasn't written by an AI, but if somebody
51:39
told me it was, like I would not be surprised, like, you don't need people to write dialogue,
51:47
like, you know, do you think you're going to have to do it again?
51:51
Probably.
51:52
Like, you don't need, you don't need human beings to do that.
51:58
Why not do it by machine?
51:59
That's true.
52:00
I mean, we'd do a better job because you could tell the machine to just cut out the
52:03
final 40 minutes of the movie.
52:05
Well, you know, that's already decided by machines.
52:09
Yeah.
52:10
Well, it's not very well.
52:11
It's under my some algorithm.
52:12
I mean, when you say that John Wick is long, my God, I don't mean to be a winging old person,
52:16
but I've actually started, Stephen, going to see movies, knowing that I will walk out
52:21
of them quite happily, quite happily.
52:24
I like, I went to see Fast and Furious 10 and John Wick for, and Guardians of the Galaxy
52:29
3.
52:30
They're all three hour, you know, three and a half hour movies or something.
52:34
I want, I want 80 minutes of cars, you know, blowing up.
52:38
I want to eat my popcorn and have my ice cream, and then I want to get the fuck out of
52:41
it.
52:42
I'm done.
52:43
You know what I mean?
52:44
And that was fun.
52:45
I don't need it.
52:46
I don't need three hours of this.
52:47
I used to be the film critic for Esquire, and I would review all of the, like, adventures
52:52
and all that stuff in Spider-Man, and I just, it broke me.
52:55
I can't do it anymore.
52:56
Yeah.
52:57
I just can't go and see them.
52:58
It's the same beats over and over and over.
53:01
Exactly the same beats, which is nice if you love the beat.
53:04
I mean, you know, I make sure that I'm not late because I know that the opening sequence
53:08
is a, is a cool action sequence.
53:11
And then there'll be a little bit of, there'll be a few funny scenes.
53:14
If it's a Marvel film, you know, they'll, they'll do something, the whimsical, an ant
53:17
man will come out and give some quips, and then there'll be another action scene, and
53:21
then I can leave.
53:23
See, the, the point I made in the after to the book, it's like with, people are scared
53:27
of AI creativity, but we're already in the space of total derivative culture, where everything
53:33
is formulated to an utter extreme.
53:36
Yeah.
53:37
I mean, it's hard to imagine it getting worse.
53:39
Yeah.
53:40
That's true.
53:41
It would be interesting to see.
53:42
The top 10 movies, all sequels, 2022.
53:45
Right.
53:46
Really?
53:47
Is that right?
53:48
Yeah.
53:49
Well, one was a reboot.
53:50
Wow.
53:51
Like, one was the Batman, but, or, but the rest were all sequels.
53:55
You know, this is the first year that new music was in decline.
53:59
Like, people are accessing archives for the new releases.
54:05
This is also part of AI creativity, why it's a potential boom, because again, it's out
54:12
of this.
54:13
That's interesting.
54:14
I mean, if you told an AI to take John Wick 4 and combine it with Kramer versus Kramer,
54:22
or ordinary people, or American beauty, you know what I mean, like, take one of the great
54:30
one of my favorite films, you know, from the past and just combine it with whatever it
54:34
was shit it is that we're pumping out at the moment and see what it comes up with.
54:38
That'd be fun.
54:39
Well, you could do John Wick and put Steve McQueen in it.
54:41
Sure.
54:42
I mean, right.
54:43
That's doable right now.
54:44
Yeah.
54:45
You'd have to do it at home, but, you know, very doable.
54:48
Or, I mean, you can't tell me that John Wick, written by a hallucinating AI, wouldn't be
54:53
better than John Wick 4.
54:54
Yeah.
54:55
I mean, it would be.
54:56
Yeah.
54:57
Interesting.
54:58
Just, I want to come to this question of having friends.
55:00
You mentioned, you know, am I going to get a chatbot friend for my kids?
55:04
Or, you know, I can easily imagine that we are not very far away from personal assistance.
55:11
Like, I'm fascinated by what Apple's doing here, Stephen, because, you know, Apple is
55:15
always, Apple strategy is always, don't be first to the party.
55:19
Wait, let everybody else figure it out and we'll come in, you know, the blackberry was
55:22
around.
55:23
They didn't come out with the iPhone first.
55:24
That wasn't the first, you know, smart device.
55:26
Let somebody else do it.
55:27
And while chat GPT, while everyone's gushing about chat GPT, what's Apple doing in the
55:32
background?
55:33
And what are they coming up with that they'll release in two or three years?
55:35
Anyway, whatever that is, assume that it's like a personal assistant.
55:40
I can imagine that we're almost at the point at which my kids, who are five, cleaning up
55:46
their clothes and brushing their teeth and doing the morning routine and getting their
55:50
breakfast cereal, could be massively helped if they had a friend on a smart, smart device
55:58
who could help them do it instead of me nagging them about it.
56:01
I would be enormously helped if I had a really pleasant to interact with non glitchy system
56:09
that could just say, hey, by the way, just notice that airfares to Bali have dropped
56:14
by 40%.
56:16
If you want to jump on that for your September holiday, you know what I mean?
56:19
Think of the advertising in that.
56:21
Sure.
56:22
I'd hand myself over, you know.
56:24
It's very interesting because this has lots to do with the tech then to do with the social
56:29
response.
56:30
Yes.
56:31
And there are some examples.
56:33
Like, they wanted to try and build a chatbot for CBT, cognitive behavioral therapy.
56:39
Because it's pretty much standard advice.
56:41
I mean, it varies, but it's much more regulated and regular than talk therapy or other forms
56:48
of therapy, but it didn't work at all.
56:52
Because it turns out a massive reason people go to therapy is to have human contact, right?
57:00
And just actually be face-to-face with people.
57:03
I mean, I think a chatbot for children is kind of the holy grail, but it's also probably
57:09
a poison chalice.
57:11
Like, I mean, what are the social consequences of children who are interacting primarily
57:18
with a bot that exists to please them?
57:21
I mean, very dangerous.
57:24
I'm not sure you would want that for your children.
57:28
On the other hand, you know, educational chatbots, which are already being developed,
57:32
just have, you know, clear functionality.
57:37
What do you think the first big change is going to be?
57:40
Like when I was talking the other day about some amazing counterintuitive ideas about chronology,
57:49
just to think things like...
57:50
You remember National Lampoon's Vacation that movie with Chevy Chase?
57:53
Yeah, I don't know how that fits into any of the things that you've been thinking about.
57:57
Just give me one minute.
58:01
National Lampoon's Vacation was made closer to World War II than to today.
58:06
Right.
58:07
So, you know, like little things like that, or that like Cleopatra lived closer to the
58:12
invention of the iPhone than to the construction of the first pyramids.
58:17
Yeah.
58:18
You know, these weird things, right?
58:19
Yeah, the sweep of time.
58:20
So, and I was thinking back and I noticed that 9-11 happened before there were iPods,
58:29
there were no iPods, you know, the old brick with the spin wheel, that wasn't out in
58:35
September of...
58:36
Oh, no.
58:37
2001.
58:38
They had to use...
58:39
They sold out of cameras in Manhattan that day.
58:43
Right.
58:44
But I'm not talking about my phone.
58:45
I'm just talking about the pod that held, you know, that was as heavy as a brick and held
58:49
100 songs on it, right?
58:50
That came out in, I think, October of 2001, the original iPod.
58:54
Yeah.
58:55
So, that was, I was just, and I was thinking back on that and I was thinking, it's amazing
59:00
to me that when I was 15 and I made plans to go to the movies with somebody and they were
59:06
late, I would just stand outside the cinema and just stand and just look at people, you
59:16
know, and look at the sky, you know, and it was, and there was nothing...
59:20
Yeah.
59:21
There was nothing to do with Steven.
59:23
And there was no way to tell the person, there was no way to, like, eventually, if they
59:28
didn't show up, I would just wander away, like, you know, there was no way to communicate
59:33
with them.
59:34
And it's amazing to me that it's amazing to me that that I was alive during that time
59:39
because it seems so archaic now.
59:43
So this is all a very roundabout way of saying, what do you think is about to happen that
59:47
when we look back on 2023, we'll go, I can't believe that when Steven and Josh were
59:53
talking on that podcast, we lived in a world without X.
59:58
Mmm, extremely hard to say that because this tech is unfathomable.
1:00:04
So, like, I mean, literally something totally mind-blowing could come out tomorrow or not.
1:00:11
It's so hard to say, like, even, even the internet, you could sort of sense where it
1:00:18
was going, even this you really can't.
1:00:20
Well, can I try one on you and see if you buy it or not?
1:00:24
Sure.
1:00:27
Moving through our days, communicating with machines as if they were sentient.
1:00:34
See, I feel like I'm, like, halfway there.
1:00:38
Like, I'm talking to machines all the time, do this, you know, move.
1:00:42
I mean, you know.
1:00:43
But they're dumb.
1:00:44
You're the only person who's, there's no illusion, I don't think, when you talk to
1:00:49
Siri or Alexa, that...
1:00:52
Well, yeah, but remember, also people treat their personal assistants like they're dumb
1:00:57
too, right?
1:00:58
Like actual human beings.
1:00:59
Right.
1:01:00
That's true.
1:01:01
Like, like, the idea that we're going to somehow be kind to these machines, if they
1:01:08
obey traits of, if they show traits of sentience, like, no, we don't treat human beings that
1:01:15
way.
1:01:16
Interesting.
1:01:17
Right.
1:01:18
I think it'll do something to my head.
1:01:23
Maybe I'm big-noting myself by assuming that if I had a personal assistant, I wouldn't
1:01:27
abuse them, and I would take care of them, and I'd like them, and I'd assume that they
1:01:30
have an interiority that is relevant to the consciousness of the cosmos.
1:01:36
But I think that...
1:01:37
Oh, it's relevant just not to you.
1:01:38
Just not to me.
1:01:39
Somebody must give a shit about this.
1:01:40
You'd be better than me.
1:01:42
But I think it'll be...
1:01:47
I think when the machines are talking back to me in ways that make me laugh, or make
1:01:53
me surprised, or make me, as I'm walking out the door, when they make a remark, and I
1:01:57
shake my head, and they say, OK, whatever, Siri.
1:02:02
When I have some kind of a relationship where, regardless of what's going on inside
1:02:06
the machine, which will be nothing initially, at least, I'm behaving...
1:02:12
You could look at my behavior, and my behavior would be indistinguishable to the behavior
1:02:16
of somebody who's relating to a conscious entity, which I don't think we do yet.
1:02:21
Then I think, once that's woven into the fabric of our lives, that will be a different world.
1:02:28
It's so sweet, if you don't mind me saying, that your version of the future with artificial
1:02:34
consciousness is this voice that bosses you around, and makes you smile, and then you
1:02:41
get to say, oh, whatever, that's one future.
1:02:45
That's one future.
1:02:46
I've already articulated the other future, and you shot that one down.
1:02:50
The other future is it goes spiraling out of control, and it enslaves me.
1:02:54
I hope you get there, man.
1:02:56
I hope you find the robot of your dreams that tells you, like, should we go to Bali this
1:03:01
weekend?
1:03:02
It says, you know, it's time to clean your room, but I mean, I don't know if that...
1:03:13
I mean, why would it not?
1:03:14
If there's a market for people having fewer things on their to-do list and less clutter
1:03:21
in their heads, which there clearly is, if you look at the success of getting things done
1:03:25
by David Allen, or any of these other productivity apps that might be out there.
1:03:30
If there's a market for that, and if there's a technology that can use language learning
1:03:35
models to assist us in predicting what I'm about to need or want, some company is going
1:03:42
to do that, surely.
1:03:44
You know, it's interesting.
1:03:46
Like, there was a case of, I think, the company was called Limanada, I may have it wrong.
1:03:53
They were an insurance company, and they broadcast that they were using AI facial recognition
1:03:58
software to spot frauds.
1:03:59
Now, this is an obvious benefit.
1:04:01
You know, we pay for all this fraud, and it's going to decrease it just by this technology.
1:04:09
People are outraged, and the outrage was so intense that they had to shut it down completely.
1:04:14
And with AI, it's really hard to regulate it other than to stop it, either to stop it
1:04:19
or let it go.
1:04:20
Right.
1:04:21
It's kind of a zero or one proposition.
1:04:24
And so this is a question where there's a real social benefit, but human beings don't
1:04:29
want it.
1:04:30
Well, I think that's the question.
1:04:32
I would come back to reveal, Stephen, I think there were revealed preferences, which is
1:04:37
economists won't speak for the things that we actually want that we'd reveal through
1:04:41
our actions, and then there are the claims that we make about what we want.
1:04:45
And I'm a skeptic about people's privacy concerns.
1:04:50
I think we like to talk about it a lot, and then, you know, the other day, the premier
1:04:55
of the state issued some vouchers, which parents can get for, which are called Creative
1:05:00
Kids vouchers, which sometimes the state government will issue to try to stimulate things
1:05:04
and, you know, help families, ease the burden on families, you get $150 if you're just
1:05:09
registered.
1:05:10
So I go through to the state government website, the portal, where I get, you know, benefits
1:05:14
and tax, whatever, all that stuff, and it asks me to go back out and like log in using
1:05:20
some other thing.
1:05:21
And I'm like, oh, fuck, can't you just, oh, that's right, it needs me to reenter my driver's
1:05:26
license number.
1:05:27
I'm like, just save it in the fucking app.
1:05:29
Like, you know, I claim to want about, to want privacy, but I also don't want to spend
1:05:34
the extra eight seconds that it takes me to actually locate my driver's license again.
1:05:38
I just want them to store it there in the app.
1:05:40
I claim to want privacy, and every time Google doesn't, you know, auto fill a password,
1:05:46
I'm frustrated.
1:05:47
I claim to want privacy, and every time the, my travel agent website doesn't remember my
1:05:52
credit card information and what my last booking was, I get annoyed that I have to reenter
1:05:57
the departure city as Sydney, and we don't want privacy.
1:06:00
We want things to run smooth.
1:06:01
Oh, I agree with you.
1:06:02
I mean, and you know, it's one of those things, whenever you study politics, like it's
1:06:06
like independent to the United States, like 47% of Americans claim to be independent.
1:06:11
Yeah.
1:06:12
But when you look at how they vote, they don't change their voting patterns ever.
1:06:15
That's, the people actually change is about 6%, right?
1:06:18
So, like, I totally get that.
1:06:20
My point is that the way AI is going to develop is going to be determined socially, not
1:06:27
necessarily by the technology, right, not necessarily by what's possible, by what will
1:06:31
accept.
1:06:32
Hmm.
1:06:33
Hmm.
1:06:34
Interesting.
1:06:35
What a political thriller with Andrew Yang.
1:06:37
Yeah.
1:06:38
I think I'm saying September.
1:06:39
It's been a hell of a year, man.
1:06:41
Incredible.
1:06:42
How did that come about?
1:06:43
You want to pitch it?
1:06:44
Yeah.
1:06:45
I mean, it's a really interesting thing.
1:06:48
I mean, Andrew called me up, and he read the next Civil War, and he said, you know,
1:06:53
you didn't talk about politics.
1:06:55
Why don't you, why don't we do a book together where we do a political thriller, and I'll
1:07:00
tell you how American politics actually works.
1:07:03
And I said, yeah.
1:07:06
And so he let me have access to everyone and the staff, including, like, humanization,
1:07:13
consultants, and all the op-a research, like, how that works.
1:07:17
And so it's the most detailed and accurate political thriller of all time.
1:07:22
And of course, it's pretty grim.
1:07:24
Like, when you look at the reality of how this works, I mean, I just written the next
1:07:30
Civil War.
1:07:31
I was not particularly shockable, but I was shocked all the time.
1:07:34
But what?
1:07:36
Oh, you know, at one point, I'm interviewing Andrew, and I'm saying, how does opposition
1:07:42
research work?
1:07:44
And he goes, we're talking about strategy, and he's like, well, they'll probably unload
1:07:48
the book then, or maybe they'll unload the book a month later.
1:07:53
And I was like, hold on a minute, when you say unload the book, is there an actual book?
1:07:58
And he was like, oh, yeah.
1:08:00
Yeah, every major American politician has a book written about them, which is everything
1:08:06
bad they've ever done, including everyone they've sat next to who has a horrible opinion
1:08:13
at a gala or whatever.
1:08:17
And I saw some of these books, and it's like, right, every no wonder American politics
1:08:24
is only for massacres, because it's incredibly cruel.
1:08:28
And that would just be one example.
1:08:30
And that's opposition research putting together the book, or it's the campaign themselves
1:08:33
putting the...
1:08:34
It's opposition research.
1:08:35
Well, both do it.
1:08:37
But yeah, it's opposition researchers.
1:08:39
And they're just, they just go in their day and do political, personal, you know, assassinations
1:08:45
and people of their character.
1:08:47
That's what they do all day, the way that you and I are talking right now.
1:08:52
And that was just one part of it.
1:08:54
The way that dark money works, too, was also like, oh, how could this ever possibly
1:08:59
work?
1:09:00
Like, you know, nobody knows where the money comes from to elect the president.
1:09:06
Yeah.
1:09:07
Like nobody knows.
1:09:11
Like the New York Times doesn't know.
1:09:12
Like the Washington Post doesn't know.
1:09:14
Like nobody knows.
1:09:15
That's interesting, isn't it?
1:09:17
So there's ostensibly rules against foreign actors.
1:09:20
Not that this would be necessarily worse than domestic corruption, but there are rules
1:09:24
against foreign actors donating to political campaigns in America.
1:09:27
And I unwittingly made political donations while I was living in the United States as a
1:09:32
foreign actor.
1:09:33
And nothing happens.
1:09:34
I mean, at least not in small fry, right?
1:09:37
I mean, you just give the money and then the money is gone and it's left your account.
1:09:41
Who's, who's overseeing this?
1:09:42
One thing that America's pretty good at is if you give them money, it tends to work
1:09:45
out.
1:09:46
That's right.
1:09:47
Yeah.
1:09:48
It nothing happens.
1:09:49
It'll be okay with it.
1:09:50
They take the money.
1:09:51
Yeah.
1:09:52
That's right.
1:09:53
I mean, I want to ask you about the future of, I guess, political and cultural stability
1:09:57
amid all of this.
1:09:58
So there's sort of intersection of what you've learned from Andrew Yang and also from
1:10:02
artificial intelligence.
1:10:03
This will just be for our sub-stack at premium subscribers.
1:10:06
So thanks to everyone else for listening.
1:10:09
But where is, what is your prognostication?
1:10:10
Are you like an optimist or a pessimist big picture?
1:10:14
Well, I mean, I think I'm a pessimist in local cases.
1:10:18
I'm actually an optimist.
1:10:19
Yeah.