What’s so crazy about this moment in artificial intelligence is that many of the most credible voices in AI think there’s a real chance that this all turns out really, really badly.
Anthropic CEO Dario Amodei recently pegged his “chance that something goes really quite catastrophically wrong on the scale of human civilization” between 10% and 25%.
That’s comforting.
Applications to attend the Cerebral Valley AI Summit close TODAY October 17.
Apply right now to be considered for an invitation!
On the series’ first episode we reflect on how generative artificial intelligence and large language models took Silicon Valley by storm.
So in our second episode of the six-part Cerebral Valley podcast, Max Child, James Wilsterman, and I played out the doomsday scenarios. We talked a lot about science fiction and how writers have imagined artificial intelligence turning dystopian.
In the second half of the episode, I talked with science fiction author Daniel H. Wilson. He wrote the books How to Survive a Robot Uprising, Where’s My Jetpack?, and How to Build a Robot Army. Wilson has also consulted with the military to help them game out how dystopian technologies might unfold.
Of course, even in the Anthropic CEO’s estimation, the most likely scenario is probably a more boring one: artificial intelligence doesn’t try to secretly destroy us as we sleep in our beds.
But the fact that there’s a chance is certainly worth considering.
I open our conversation with the parable “The unfinished fable of the sparrows” from Nick Bostrom’s Superintelligence.
It was the nest-building season, but after days of long hard work, the sparrows sat in the evening glow, relaxing and chirping away.
“We are all so small and weak. Imagine how easy life would be if we had an owl who could help us build our nests!”
“Yes!” said another. “And we could use it to look after our elderly and our young.”
“It could give us advice and keep an eye out for the neighborhood cat,” added a third.
Then Pastus, the elder-bird, spoke: “Let us send out scouts in all directions and try to find an abandoned owlet somewhere, or maybe an egg. A crow chick might also do, or a baby weasel. This could be the best thing that ever happened to us, at least since the opening of the Pavilion of Unlimited Grain in yonder backyard.”
The flock was exhilarated, and sparrows everywhere started chirping at the top of their lungs.
Only Scronkfinkle, a one-eyed sparrow with a fretful temperament, was unconvinced of the wisdom of the endeavor. Quoth he: “This will surely be our undoing. Should we not give some thought to the art of owl-domestication and owl-taming first, before we bring such a creature into our midst?”
Replied Pastus: “Taming an owl sounds like an exceedingly difficult thing to do. It will be difficult enough to find an owl egg. So let us start there. After we have succeeded in raising an owl, then we can think about taking on this other challenge.”
“There is a flaw in that plan!” squeaked Scronkfinkle; but his protests were in vain as the flock had already lifted off to start implementing the directives set out by Pastus.
Just two or three sparrows remained behind. Together they began to try to work out how owls might be tamed or domesticated. They soon realized that Pastus had been right: this was an exceedingly difficult challenge, especially in the absence of an actual owl to practice on. Nevertheless they pressed on as best they could, constantly fearing that the flock might return with an owl egg before a solution to the control problem had been found.
Give it a listen.
P.S. I’m on my honeymoon right now in Japan. I was working frantically to record these episodes before I left. My chief of staff Riley Konsella is sending the episodes out for me while I’m gone. If you need anything while I’m away, you should email Riley.
Thanks in advance for being understanding that this newsletter is slowing down for my honeymoon. I’m going to dedicate myself to relaxing over the next two weeks so that I come back hungrier than ever.
Get full access to Newcomer at www.newcomer.co/subscribe
00:00:10
Hey, it's Eric Newcomer. Welcome back to Cerebral Valley.
00:00:14
This is Episode 2. I'm here with Max Child and
00:00:17
James Wilsterman. We are going to dig into.
00:00:20
We're calling this episode AI Kills Us All, which will give
00:00:25
you a taste. It's not not the small
00:00:28
questions. You know what what to do about
00:00:30
privacy or like you know who should be in charge.
00:00:34
It's like, what happens if we get artificial general
00:00:37
intelligence? What, how worried are we about
00:00:40
sort of the big doom and gloom? Guys?
00:00:43
Great. Great to be back with you.
00:00:46
Great to be back. Thanks for having us.
00:00:47
Great to be here. After my conversation with Max
00:00:49
and James, I will be interviewing Daniel H Wilson,
00:00:53
the author of How to Survive a Robot Uprising.
00:00:55
Where's my jetpack? And how to build a robot army?
00:00:58
And the best seller, Robo Apocalypse.
00:01:00
So someone who has spent a lot of time imagining the doom and
00:01:05
gloom of machines getting powerful.
00:01:09
So that's he. He's an expert in this
00:01:11
speculation where we are mere amateurs, you know, with of
00:01:14
course, the help of chat. GPTI was, you know, looking,
00:01:19
wandering the Internet, looking for the great thinkers on AG is
00:01:28
coming, and one of them is Nick Bostrom, who's been writing
00:01:32
about this for years, I think. So this piece is the essay from
00:01:35
his book Super Intelligence. The essay is called The
00:01:41
Unfinished Fable of the Sparrows, which is very
00:01:45
enjoyable and I I think you'll get a lot out of it.
00:01:50
It was the nest building season, but after days of long, hard
00:01:53
work, the sparrows sat in the evening glow, relaxing and
00:01:57
chirping away. We are all so small and weak.
00:02:00
Imagine how easy life would be if we had an owl who could help
00:02:03
us build our nests. Yes, said another, And we could
00:02:07
use it to look after our elderly and our young.
00:02:10
It could give us advice and keep an eye out for the neighborhood
00:02:13
cat, added a third. Then past us, the elder bird
00:02:17
spoke. Let us send out scouts in all
00:02:19
directions and try to find an abandoned outlet somewhere.
00:02:23
Or maybe an egg a crow chick might also do.
00:02:26
Or a baby weasel. This could be the best thing
00:02:29
that ever happened to us, at least since the opening of the
00:02:32
pavilion of Unlimited grain in Yonder Backyard.
00:02:36
The flock was exhilarated and sparrows everywhere started
00:02:39
chirping at the top of their lungs.
00:02:41
Only Skronk Finkel A1 eyed Sparrow with a fretful
00:02:45
temperament was unconvinced of the wisdom of the endeavor.
00:02:48
Quoth he? This will surely be our undoing.
00:02:51
Should we not give some thought to the art of owl domestication
00:02:55
and owl taming first, before we bring such a creature into our
00:02:59
midst replied past us, taming an owl sounds like an exceedingly
00:03:03
difficult thing to do. It will be difficult to find an
00:03:07
owl egg, so let us start there. After we have succeeded in
00:03:10
raising an owl, then we can think about taking on this other
00:03:14
challenge. There is a flaw in that plan.
00:03:17
Squeak Skronk Finkel. But his protests were in vain,
00:03:20
as the flock had already lifted off to start implementing the
00:03:24
directive set out by Pastis. Just two or three sparrows
00:03:28
remain behind. Together, they began to try to
00:03:31
work out how owls might be tamed or domesticated.
00:03:35
They soon realized that pastis had been right.
00:03:37
This was an exceedingly difficult challenge, especially
00:03:40
in the absence of an actual owl to practice on.
00:03:44
Nevertheless, they pressed on as best they could, constantly
00:03:48
fearing that the flock might return with an owl leg before a
00:03:53
solution to the control problem had been found.
00:03:58
All right, what do you guys make of that?
00:03:59
It's. Not quite Plato, but it's.
00:04:04
Heavy-handed, little on the nose, I feel like this might
00:04:12
have been more. You know, mind blowing in 2014
00:04:16
when when it was raining, I guess, but nowadays it's just
00:04:19
like, Oh yeah, that's that's what we're talking about all the
00:04:21
time. But I do think, I do think this
00:04:27
obviously gets at this question of, you know, should you ever,
00:04:31
should you even go down this path at all, right.
00:04:33
I think that's the core question in develop trying to develop
00:04:37
AGI. Or is it so risky that you
00:04:41
should? You know not do that at all
00:04:43
unless you are 100% confident you figured out how to tame the
00:04:47
OWL right in advance. Is that your guys reading as
00:04:51
well? Yeah, well, I think just to like
00:04:54
rewind one second, I think coming back to this like AI
00:04:57
kills us all point. I think there's two big
00:05:00
questions in the sort of will AIS kill, kill us all narrative,
00:05:03
right. Or maybe three.
00:05:05
I guess one will be are we capable of developing an AI
00:05:09
that's smarter than humans? Like 2 will be if we're capable
00:05:14
of developing an AI smarter than humans, would it want to kill
00:05:17
us? And then three being, if it
00:05:20
wants to kill us, can it figure out some way to kill us?
00:05:23
Basically, right. I mean like and I think those
00:05:25
are all kind of three different questions and maybe we all like
00:05:28
agree on this podcast on like certain one or two or three or
00:05:32
or different elements of them, but like.
00:05:34
I think it's helpful to. Yeah.
00:05:35
To break it down that way because.
00:05:37
And if you do that, this, this story, the parable, the Fable
00:05:40
that I just read sort of assumes a couple of those, right.
00:05:43
It assumes by using an owl, it assumes that such a thing could
00:05:48
exist, right. And that it, it's a dangerous
00:05:52
being that we actually need to figure out in its essence, it's
00:05:55
dangerous and we need to figure out how to control it before we
00:05:58
create it, right. So it sort of assumes into the
00:06:00
story certain things exactly that you're framing up are open
00:06:04
question. I think they're open question.
00:06:05
I mean in the parable basically the OWL both is smarter and more
00:06:08
powerful than the sparrows, definitely wants to apparently
00:06:13
kill or very likely potentially wants to kill the sparrows and
00:06:15
certainly has the capability given that it's bigger and
00:06:17
stronger or whatever, right. So to your point, it's sort of
00:06:19
like you know, the answer is kind of written into the story,
00:06:22
whereas I think in real life like I think can we build AI
00:06:25
smarter than us. Part one is actually a pretty
00:06:27
interesting and contentious question and then the other ones
00:06:29
are also pretty interesting and contentious as well.
00:06:31
But just to start on part one like.
00:06:34
Do you guys think we can build an AI that's smarter than
00:06:36
humans? I guess like, let's be very
00:06:40
specific about how we define that, right?
00:06:42
Because I think it's a pretty critical question.
00:06:45
Come on. Well, what's your definition of,
00:06:49
I guess, smarter than humans? Is that a?
00:06:52
Are you saying super intelligence?
00:06:53
We're talking about, say, an. IQ That has never been achieved
00:06:56
by a human right. You know I.
00:06:57
I think that like I think there's kind of this concept of
00:07:01
a IAGI, which is. Better at I would say like the
00:07:06
median human at, you know, most tasks and then there's super
00:07:11
intelligence, right? Which is you know, better like
00:07:14
better than the best humans at most tasks, right.
00:07:17
Is that how you guys think about it as well?
00:07:19
Yeah. Let's say better than the best
00:07:20
humans at any mental task or any any relevant mental task, right?
00:07:25
I guess I believe that this is going to happen, I think.
00:07:29
You know, it would be naive to say, well it won't happen.
00:07:32
So then it's just a matter of a question of like when it will
00:07:35
happen and whether there are any kind of domains where it feels
00:07:40
like it won't happen for a long time, right?
00:07:43
Like are there, you know, any domains that we really think you
00:07:47
know it's going to be take take much longer to achieve super
00:07:49
intelligence than than other domains, right.
00:07:53
So maybe it will be the best. Coder in the world.
00:07:57
But will it also be the best screenwriter?
00:07:59
Are those things gonna happen at the same time or are they gonna
00:08:02
happen, you know, at different times?
00:08:05
Yeah, I I, I mean, first of all, you know, AI is smarter than the
00:08:11
smartest humans in some domains right now.
00:08:14
Already shall you go Wow games particularly.
00:08:18
Right. But that that is not a general
00:08:20
intelligence. No, no, I'm just saying, just.
00:08:22
Yeah, but in specific demands it's better.
00:08:24
I I think it will continue to cleave off domains, right.
00:08:28
I think like sort of beat all humans at the LSAT type thing
00:08:32
seems very soon. I I think what feels far away is
00:08:37
sort of and it fits into the fears of an AGI is sort of the
00:08:42
strategic planning like what an AGI looks like that sets like
00:08:45
broader priorities across like in a game it's like you want to
00:08:49
win at chess, like straightforward, but when it has
00:08:52
to you know sort of solve general optimal outcomes.
00:08:56
Like it feels like we are so far from a being that could do that.
00:09:01
Or a computer that's interested in sort of deciding whether it
00:09:05
should play chess or do something else and like why
00:09:08
especially if we're not again just articulating what the goals
00:09:11
are. So I guess I'm, I'm not as
00:09:14
bullish on in the next decade a sort of coherent overall sort of
00:09:19
being that feels like smarter than a human.
00:09:22
I mean the next decade, though. I mean, if we're already
00:09:24
negotiating this down to the next decade, it would seem like
00:09:27
you're pretty confident that it's gonna happen.
00:09:28
Like lifetimes. I think it's gonna happen, OK.
00:09:30
So you think in our lifetimes that will happen?
00:09:32
Yeah. What about my my earlier
00:09:35
distinction though Eric, do you do you think you know an AGI
00:09:38
that is a general intelligence that is basically better than
00:09:41
the media and human at most things is or or equal to is more
00:09:46
likely to happen in the next decade.
00:09:48
Is that five years? I just think it's hard to do it.
00:09:51
I we just don't see it as sort of a an overall agent or like an
00:09:55
overall decision, you know what I mean?
00:09:57
I I see it like winning at a bunch of discrete tasks.
00:10:01
OK, sort of a sign. So this kind of gets at more of
00:10:04
like your definition of AGI potentially would require some
00:10:09
sort of Turing test type thing that it can really just convince
00:10:13
you that it is a human. Or I guess I'm trying to think
00:10:15
of like an overall thinker with like sort of priorities and sort
00:10:20
of. But at the end of the day, like,
00:10:22
just to make any predictions about this, it has to kind of be
00:10:25
a falsifiable assertion of like, what is AGI, right?
00:10:30
And it's not that easy to to define that except by some sort
00:10:34
of like test, right? Otherwise you end up in this
00:10:37
like era or this pattern we we're always in that we are
00:10:41
constantly moving the goal posts on like what AGI is, right?
00:10:44
Right. Yeah.
00:10:46
I mean, I think, I mean, I think you guys are both like in the
00:10:49
same place, which I am, which is like it's very hard to figure
00:10:51
out the timeline, but like to me it's definitely happening
00:10:54
sometime in the next 100 years, let's say.
00:10:56
Whether it's 10 or 20 or 50 or 100, it is a little bit harder
00:11:00
to guess, but like it is kind of a math problem, right?
00:11:03
Where we have brains that are just some giant collection of
00:11:06
neurons, right? And there's, you know, millions
00:11:09
and millions of them or whatever, whatever order of
00:11:12
magnitude it is, right? And essentially we're training
00:11:15
these large language models on, you know, a neural network in
00:11:18
some ways, right? Which, you know, it's not a one
00:11:21
to one copy of what? A neuron, how a neuron works in
00:11:24
our brain, which is way more complicated, but.
00:11:26
Fundamentally, you're sort of like, if I can just throw
00:11:28
numbers at this problem of like building a brain essentially and
00:11:32
like get more GPU cores, like just throw absolute just
00:11:37
computing power at this thing. Like the story of our entire
00:11:39
lives has been computing power just keeps going up
00:11:41
exponentially, right? You know, Moore's Law or
00:11:43
whatever, right? And so to me, it seems like
00:11:46
you'd have to believe there's something fundamentally
00:11:48
different about the way a human brain works or a neuron works.
00:11:51
To the way you know a neural network or you know the way GPU
00:11:55
transformer model works to to believe that we can't just throw
00:11:58
numbers at this thing until it gets smarter than humans.
00:12:01
And so I guess I do believe that we throw enough numbers at it
00:12:03
and get smarter than humans. I mean humans have to win
00:12:06
multiple sort of parts of the brain that developed in
00:12:09
different ways that interact with each.
00:12:11
Other. Yeah.
00:12:12
But they're still kind of built on the same cell structure.
00:12:15
I mean, yeah, yes, but but you could do that in an in an AGI
00:12:18
too, right? You could have different lobes
00:12:19
or whatever of the imaginary intelligence, right?
00:12:21
And then broadly, we just really don't understand how to pinpoint
00:12:26
the experience of human consciousness, like quality or
00:12:30
whatever. I talked about this at length
00:12:31
with Reed Hoffman when I interviewed him and like so, but
00:12:36
we're going to, but we're very reluctant because we can
00:12:38
experience our own consciousness to attribute it and animals and
00:12:42
we really have no like road map for how we'd ever identify it in
00:12:46
machines. And that's sort of different
00:12:48
than the Super intelligence thing, but it's certainly part
00:12:50
of what you want in the sort of like does it have like a real
00:12:53
existence sort of question, which I don't even have a road
00:12:56
map for how we'd identify. Coming back to the core topic of
00:12:59
the episode, like, will AI kill us all?
00:13:01
Like, it doesn't really matter if it's conscious or not, right?
00:13:04
If it ends up having the power and the desire to kill us,
00:13:06
right. So like, I do think that's an
00:13:09
interesting question and I don't know, maybe that's a different
00:13:11
episode. But like to me that's not,
00:13:13
that's not worth getting hung up on in the like is it going to
00:13:15
kill us all thing like definitely.
00:13:17
I mean, this sort of classic example being the paper clip
00:13:21
maximizer, right? The machine, the fear that we
00:13:25
just program this really smart, you know, LLM and tell it like,
00:13:29
oh, you're not only objective is maximize paper clips, you know,
00:13:34
And then it goes about it and then it's like, well, the humans
00:13:36
are hurting sort of the maximization of paper clips.
00:13:39
Let's get rid of them. I mean famously like I Robot by
00:13:43
Asimov. It's like robots are supposed to
00:13:46
protect humans, but then humans all kill each other.
00:13:49
So it's like, oh, we need to protect the humans.
00:13:51
I sort of need to enslave them in order to ensure their
00:13:54
protection, you know, so that doesn't it doesn't necessarily
00:13:58
need to have the depth of thinking.
00:13:59
You know, humans would see sort of the flawed what we would
00:14:02
perceive to be flawed reasoning in those things, but it would
00:14:05
still but a machine. You could see why a machine with
00:14:08
certain objectives could operate.
00:14:10
Though well, I I think it's sort of like saying that these
00:14:13
artificial intelligences can have totally different
00:14:17
orthogonal objectives that don't require a consciousness, right?
00:14:21
Like they just for some reason develop a different objective.
00:14:24
Than we have as humans, and they're capable of bringing
00:14:29
about that objective to the detriment of humanity,
00:14:32
basically, right? Yeah, I mean, a nuclear weapon
00:14:35
doesn't have consciousness, but it's still very effective at
00:14:37
killing people, right? That's its mission, so.
00:14:41
Yeah. I mean, so I, it seems like we
00:14:43
all agree, OK, we're going to achieve super intelligence at
00:14:45
some point in the next, let's say, century or an intelligence
00:14:47
that's smarter than humans and enough domains that it's, it's
00:14:50
meaningfully, you know, more capable than we are, right.
00:14:52
Yeah. So second question, like, you
00:14:56
know, would it want to kill us? I think this is like one of the
00:14:58
most interesting ones and you alluded to the sort of paper
00:15:01
clip maximization as one example of how it might just end up
00:15:04
killing us as like a side effect of some other mission it's on.
00:15:07
Right, but. I guess A, like, how likely do
00:15:10
we think that is? And then B, like, do we think
00:15:13
that we'll be able to constrain its missions in some way so that
00:15:17
it wants to keep us alive, you know, even if it's going off and
00:15:21
executing at school? I do actually, you know, worry
00:15:25
about this. I guess that the AI will have
00:15:29
some orthogonal objective that is detrimental to humans.
00:15:33
And then there's like a secondary question of like, can
00:15:35
we do anything about that, right.
00:15:37
We were like, I mean to read between what you're saying,
00:15:39
you're like, if anthropic and open AI are left only in charge,
00:15:42
we'll be fine. But if Facebook's llama runs
00:15:45
around or screwed, I mean, that's sort of, I mean, I'm
00:15:47
interested to see what Mustafa, who's going to speak at Cerebral
00:15:50
Valley, you know, he's been much more reticent about open source,
00:15:54
potentially. His lack of control issue.
00:15:57
I do think that eventually if we start to see, you know, the
00:16:03
risks like that, we all agree that.
00:16:06
You know this, we're headed towards this super intelligence
00:16:08
and we start to see how dangerous that may or may not
00:16:12
be. Like, I just don't think we know
00:16:13
yet how dangerous it will be. Like it's obvious to me like you
00:16:17
want to maintain the ability to regulate this.
00:16:20
And you know, just a lot of people, I think I saw Brian
00:16:24
Armstrong from Coinbase, you know, tweet that he just thinks
00:16:27
there should never be any AI regulation or I don't know if he
00:16:30
said never, but he was basically like regulation has such a poor
00:16:33
track record around innovation. Like, I just think that's sort
00:16:36
of naive without us knowing yet, like how dangerous it can be or
00:16:42
not. You just need to have the best
00:16:44
AI on our side. You know, like when the killer
00:16:46
one comes, we need to have the benevolent 1.
00:16:48
The best defense against a bad guy with an AI is a good guy
00:16:52
with an AI. That's how that's how Americans
00:16:55
actually think. You know, like.
00:16:57
The right to bear AI, baby, Yeah.
00:17:01
That was kind of Sam Altman's perspective when he started Open
00:17:04
AI. Like, I mean, at least my
00:17:05
reading. Of it when he was getting
00:17:07
started was calling it open AI because he believed putting AI
00:17:11
in the hands of humanity democratizing it was the best
00:17:15
way to go about things. And I think he's changed his
00:17:18
mind because of this exact thing that I'm saying, which is it's
00:17:21
actually pretty dangerous or potentially dangerous.
00:17:24
The fact that opening eye isn't open at all I know is like
00:17:26
something that has been beaten to death, but it's just still
00:17:29
ridiculous to me just that. It's hilarious kind of, yeah.
00:17:32
Yeah, cuz it's historical artifact or?
00:17:34
Something, right. But I mean I think I guess
00:17:37
putting aside the regulation thing and the open source is
00:17:40
closed source thing, I just think that if you believe that.
00:17:43
A substantially diverse group of people from different countries
00:17:47
within America, open source, closed source, can develop a
00:17:49
super intelligence, right? Like if you believe there's
00:17:52
going to be more than like five of these, which I think there
00:17:55
would be like in any scenario in which this actually happens,
00:17:58
like isn't it just very likely that one of the five or ten or
00:18:02
hundred or thousand of these is is a bad AI?
00:18:05
Mean? I just don't.
00:18:06
I don't think it matters if you have 999 aligned good AI and one
00:18:09
bad one, if that one bad one is capable of destroying the world,
00:18:12
right? I mean like.
00:18:13
It again, I just, I just think that mathematically if it's
00:18:17
possible to create super intelligence is like someone is
00:18:20
going to create a super intelligence that's really bad,
00:18:21
right? And I'm not sure like our
00:18:23
existing analogy is around like guns or nuclear deterrence or
00:18:27
whatever sort of game theory you want to play out here like apply
00:18:30
where like if anyone bad AI can get access to the tools that
00:18:34
needs to kill everyone, it will it will succeed, right?
00:18:38
Like, I don't think all the good AI can stop it, but but maybe
00:18:40
I'm going too far out on this. I mean, I agree with it.
00:18:43
I think so. So your take is that obviously
00:18:46
yes, the answer is yes. You know, if we achieve super
00:18:50
intelligence, there's a pretty big risk to that.
00:18:53
I mean, in the world. You would have to create some
00:18:56
way for the good AI to regulate the bad AI.
00:18:58
Basically, like if you believe we're in a world where no longer
00:19:01
humans are, they're not on the driver's seat, right?
00:19:03
We're in the back seat, right? Like you now have to basically
00:19:05
try to figure out how to make sure that the AI that are in the
00:19:07
front seat can control the other ones.
00:19:10
I mean part of what Max is basically referencing just, you
00:19:13
know, to give a little context, is Elisa Yudakowski.
00:19:17
I mean I pulled actually a quote, you know, to visualize a
00:19:20
hostile superhuman AI. Don't imagine a lifeless book,
00:19:23
smart thinker dwelling inside the Internet and sending I'll
00:19:26
intentioned emails. Visualize an entire alien
00:19:28
civilization thinking at millions of time human speeds,
00:19:32
initially confined to computers in a world of creatures that are
00:19:35
from its perspective, very stupid and very slow.
00:19:38
A sufficiently intelligent AI won't stay confined to computers
00:19:41
for long. In today's world, you can e-mail
00:19:43
DNA strings to laboratories that will produce proteins on demand,
00:19:47
allowing an AI initially confined to the Internet to
00:19:50
build artificial life forms or bootstraps straight to post
00:19:53
biological molecular manufacturing.
00:19:55
I don't know. My point of view is just like, I
00:19:59
think him and sort of what you guys are getting at, it just
00:20:01
feels like you're like, oh it's super intelligent.
00:20:03
It's like God, like, you know, I I still think, OK, it's way
00:20:06
smarter than us. But we're like a bunch of
00:20:09
intelligent beings, even if it's smarter than us.
00:20:11
I I mean, it requires that that it's going to act stealthily and
00:20:15
sort of like start manufacturing like a biologic version.
00:20:18
You know, like, I just feel like it's more likely that we're
00:20:20
like, this thing doesn't really listen to us and like, but then
00:20:24
like the GPU clusters are like, it's trying to fire them all.
00:20:27
And it's just like, oh, there's only so much computing power
00:20:30
that, you know, it's still a being that requires like
00:20:33
resources. You know, like humans are
00:20:35
smarter than everything else. We still die like they're
00:20:37
they're just like physical limits to things and and that
00:20:42
it's not going to have like God like powers.
00:20:44
It's going to be limited by the amount of compute it can access
00:20:47
and sort of our cooperation at various points.
00:20:50
Like, I I don't know. I don't think it's on zero to
00:20:53
God. I feel like you're painting.
00:20:55
Correct me if I'm wrong, but you're painting a picture where
00:20:57
we have created super intelligence.
00:21:01
And you agree with Max's point that some of them are misaligned
00:21:05
and probably would want to kill us if they could.
00:21:07
However, you're saying that the humanity's fine because those
00:21:13
misaligned AIS can't get enough resources.
00:21:16
Is that accurate, right, I mean. Just it's hard to know, right?
00:21:20
We don't. We don't.
00:21:21
We don't know what to do. That's definitely a gamble.
00:21:24
I mean, even if you believe that you're like.
00:21:28
You've got you've gotten pretty close to the edge.
00:21:33
Well I'm I'm definitely a fatalist on AI as my if we're
00:21:36
getting to like our deep, like what is our position.
00:21:38
It just, I just think like many people in tech, like the
00:21:42
regulation is pretty bad and like going to be only going to
00:21:46
stop the good guys, that this is going to keep going and that I'd
00:21:50
rather the best people we have try to figure it out rather than
00:21:54
like operate in the shadows. Yeah, I I tend to agree with you
00:21:58
and I probably would agree that we're not, we don't need a lot
00:22:02
of regulation right now on you know who can you know train
00:22:08
these models or something like but I.
00:22:11
I also wouldn't rule out like meeting this in the future if in
00:22:14
a year from now or two years from now I start seeing you know
00:22:17
a lot of like evidence that we've, the open source community
00:22:20
has created super intelligent misaligned AI, right.
00:22:24
And I think people are kind of, you know, going a little bit too
00:22:28
far in the direction of saying this, you know, we don't need
00:22:31
it. We need any regulation here.
00:22:33
When it's a new technology, we just don't know what it's going
00:22:36
to be like. We we don't, you know, and it is
00:22:39
a little bit like. When we were, you know, putting
00:22:43
the right to bear arms in the Constitution without knowing
00:22:45
what arms meant in the future, right?
00:22:48
Like how, how, how dangerous they may or may not be.
00:22:52
So I think that I'm just a little nervous about, you know,
00:22:55
assuming too much about whether we will need regulation in the
00:22:59
future. I mean to come back to you at
00:23:02
Kowski for a second, I mean he is sort of the the progenitor of
00:23:06
this whole AI. Dumerism, concept or whatever.
00:23:09
I mean, I've read a bunch of his stuff he seemed to almost like
00:23:11
at one point, threaten blowing up GPU.
00:23:14
Right, Yeah. Well, he had a time article that
00:23:17
he was saying in his world of protecting humanity, like one of
00:23:23
the only things you would have to do would be to.
00:23:27
You know, protect, you know, prevent rogue data centers from
00:23:30
training models. And, you know, it'd be
00:23:32
justifiable to like, bomb them essentially.
00:23:34
And and he was saying, you know, the entire kind of geopolitical,
00:23:41
you know, consensus should be that it's worse to train these
00:23:45
models at some level than it is to, like, worry about nuclear
00:23:49
proliferation. I think.
00:23:52
I mean, I having read a bunch of his stuff, I feel like he has
00:23:54
like pretty well convinced me of sort of all all three of these
00:23:58
things. I guess that will have a super
00:23:59
intelligence that it'll it'll be it'll it'll some of them at
00:24:03
least will desire to harm us and that they'll figure out some way
00:24:07
to do it. I guess I'm sort of like I'm a
00:24:09
little bit of a fatalist in the sense that I I think this is
00:24:11
basically an unstoppable train at this point that has like left
00:24:14
the station. Like I don't, I don't think.
00:24:16
Bombing data centers is like a viable solution to this problem
00:24:20
or breaking GPUs or whatever or certainly not in the like the
00:24:25
human political universe that we actually operate in today.
00:24:28
I think my, my attitude is just that like luckily, I guess my
00:24:34
the future is really hard to predict, right?
00:24:36
I mean, I know that's like a very obvious point, but if you
00:24:39
had told someone in 1940 that we were going to invent nuclear
00:24:42
bombs and they had really thought through all the.
00:24:45
You know, the potential problems in the future caused by those
00:24:48
devices, I think they could have been pretty fatalistic and I
00:24:51
think that would have been a pretty accurate read on like how
00:24:53
bad it was that we invented nuclear bombs, right?
00:24:56
Like knock on wood, so far we're still here.
00:24:58
So my sort of my fear is that all the stuff you you had cows
00:25:01
and some people it's going to happen.
00:25:02
Nuclear bombs are good, like they brought about like peace.
00:25:06
We haven't had world wars, you know, But my hope.
00:25:09
Is that we get lucky and that. Something about this future that
00:25:13
was that Yukowski or the AI doomers are sketching out is is
00:25:16
there's some missing link in the logical chain or or there's some
00:25:20
piece of the way deterrence or or alignment or whatever ends up
00:25:24
working that that we we get our ass saved.
00:25:27
But I'm kind of betting on luck at this point.
00:25:29
Like, I don't really think we're capable of stopping this.
00:25:31
Like just if we're all fatalist basically.
00:25:35
I mean there is like the human cloning, you know, people
00:25:38
brought this up to me. Like human cloning is an example
00:25:41
where people have, you know, America isn't cloning humans.
00:25:44
It seems like we could like why? Why is that an area where self
00:25:49
regulation seems much more possible?
00:25:53
I feel like there are areas where we have stopped.
00:25:56
That's still pretty hard, right? I mean, like, I bet that will
00:26:00
happen sometime in the next 50 to 100 years, right?
00:26:02
I don't know. I just feel like I don't know
00:26:04
how much desire there is for that.
00:26:06
But I I feel like once it's doable, people will do it.
00:26:08
My last thinking on one positive future scenario that I sort of
00:26:11
believe is possible is like, you know, we live on this planet
00:26:15
with lots of other organisms and creatures as you guys described.
00:26:18
You know, we have pets, we have dogs, whatever.
00:26:21
Like. Why do we have dogs?
00:26:23
Like, is there any really good reason?
00:26:24
Like the question is like, could we end up being the pets of the
00:26:27
AI? Like, could we could they want
00:26:29
to keep us alive for some reason?
00:26:30
And that we're a little bit different?
00:26:31
We're a little bit interesting to them, more fun to have around
00:26:34
in some scenarios, but like, in the end we're the AI's pets, you
00:26:37
know? Or or maybe we're the ants and
00:26:39
they just don't care to kill us all because it would be such a
00:26:41
pain, You know, right? I think that's more likely than
00:26:45
not. Right.
00:26:46
I think that is the the upside scenario is like we're we're
00:26:49
pets or we're ants or something compared to this.
00:26:51
More likely than not just the general, like the AI does not
00:26:55
hate us, you know what I mean? Like both.
00:26:57
We helped create it. Like we set a lot of its initial
00:27:01
value system, presumably like it, you know, relies us on us
00:27:05
for a lot of things early on, like I don't maybe a lot of
00:27:08
reasons. Yeah, maybe we're just different
00:27:10
in some sort of interesting way, right, in that we're not, we're
00:27:13
not wired exactly the same. And so you know it's it's just
00:27:15
interesting to have us around because we act a little bit
00:27:17
differently or or something like that.
00:27:19
And AI is not like an evolutionary being, like, you
00:27:22
know, we are driven by these like evolutionary prerogatives
00:27:25
that make us sort of competitive and you know, worried about
00:27:29
other other sort of genetic codes.
00:27:32
It it's just such a separate type of thing.
00:27:35
It's like hard to extrapolate from, I don't know, even beyond
00:27:40
humans like animals to what an AI would be like.
00:27:44
OK, so so I guess we all agree we're all gonna die or be a as
00:27:50
pets. But it seems like we we haven't
00:27:53
really narrowed in on how likely this is to occur in any
00:27:56
meaningful time frame. Like I think that to me is
00:27:59
pretty interesting, you know, if this happens.
00:28:02
In 10 years or even 20 years or, you know, I think that's pretty
00:28:06
different from how I might go ahead and live my life, you
00:28:11
know, for the next decade, I guess.
00:28:13
How about you guys? Like, do you think it's even
00:28:15
worth thinking about that you know right now and how you live
00:28:18
your life or not really? I don't know.
00:28:23
I don't know. I just.
00:28:24
I just. I can't.
00:28:25
I can't decide like if I would make any just different ways.
00:28:28
I think, you know, being close to it is appealing and like you
00:28:31
know, this is a technology that I believe in unlike crypto.
00:28:35
So it's like oh run to where, you know, the actual interesting
00:28:39
thing that could really revolutionize humanity is like I
00:28:43
I want to be around it, I guess selfishly.
00:28:45
So, you know, host an AI conference with you guys, run.
00:28:48
Into the run, into the burning building, Will AI kill us all?
00:28:54
In some ways is great PR for the AI industry, even though it's
00:28:59
sort of bleak because it suggests it takes as a premise
00:29:03
that artificial intelligence is it a really amazing point and
00:29:08
super powerful and is poised to like be extremely disruptive.
00:29:12
And so if you're an investor and you're like, well, I can't save
00:29:15
the world, I will try to profit off of it.
00:29:17
In the good scenario, Will AI Kill us all?
00:29:20
Is definitely a good motivator to like deploy capital into AI.
00:29:24
Like what do you take of the fact that sort of the dumerism
00:29:27
is like great PR for the actual existing, you know, for vector
00:29:34
database companies, it's a it's a good message, you know, I
00:29:37
don't know, what do you make of this?
00:29:38
This is like war profiteering. Is the motivation here?
00:29:42
No, I'm just saying it's like, you know, the media loves it.
00:29:45
Like Ezra Klein, they will, you know talks about it all the
00:29:47
time. But like in some ways it is a
00:29:49
good it's good marketing for the IT could be just like you know a
00:29:55
fairly mundane technology that we're nowhere close to the jump
00:29:58
TAGI like the Reed Hoffman interview.
00:30:01
He's not he's not willing to sort of commit to any near time
00:30:05
horizon that that AGI is coming. It could just be like it's a
00:30:09
fairly incremental technology. I think it's good fundraising
00:30:13
pitch to say hey like. We're trying to prevent, you
00:30:17
know, AI from killing us all. Like, we're the good AI guys,
00:30:20
right? I mean, like, yeah, you're like,
00:30:21
this is life or death doing this technology, that is.
00:30:24
So powerful. Powerful.
00:30:26
So people are worried that it's going to kill us all, you know,
00:30:29
I mean, I feel like that's a good, like, you know, it's a
00:30:33
it's a conversation that's good for industry.
00:30:35
Can is there an analogy to another industry where you think
00:30:38
that this has occurred? I'm just.
00:30:40
I'm not as convinced, I guess. Like, is it great for the oil
00:30:44
and gas industry right now that, you know, climate change is?
00:30:48
Well, you could argue, you know, like sometimes, like when
00:30:52
parents, like, freak out about, like, teen stuff, that's
00:30:55
obviously not that dangerous for them.
00:30:58
That in some ways it, like, drives teens closer to it.
00:31:01
And that it's like this is totally misunderstood by the
00:31:04
sort of authorities like, but it it reinforced, it gets it in the
00:31:08
news all the time. It reinforces that this is
00:31:10
something that's sort of going on, you know?
00:31:14
I mean I think it's like pros and cons to your point, like I
00:31:17
think it might drive a lot more regulation in a shorter time
00:31:20
frame than what they would have otherwise and.
00:31:23
To your other point, and it will probably be kind of dumb
00:31:25
regulation. So, like, you know, I don't
00:31:27
know. You know, the European Union has
00:31:29
made a terrible job of doing privacy regulations and just
00:31:33
excited to see what they gin up for AI, I mean, I think.
00:31:37
Isn't Italy banned Chachi PT right now?
00:31:40
It's like it's like already you're like, God, these European
00:31:43
regulators are just off the chain.
00:31:45
Yeah, I mean so pros and cons, right.
00:31:48
Banned in Italy maybe you can raise a couple $100 million with
00:31:51
a with a slide deck at this point.
00:31:53
So you know, you get you get both, I guess.
00:31:56
I guess I don't know if I if I was Sam Almond, like would I
00:32:00
want the level of AI doomerism? Occurring right now to continue,
00:32:05
Probably not. Probably not to be clear, just
00:32:08
my actual position is that like it things are very exciting
00:32:11
right now. That this is a legitimate
00:32:14
question that will AIS kill us all is legitimately grounded in
00:32:18
like all, you know, sci-fi content and thought experiments
00:32:21
about where this is going. So I don't think it's made-up,
00:32:24
but I think a happy coincidence is that this sort of question is
00:32:28
fundamentally good PR. That's all I'm saying.
00:32:31
So I think 1 interesting topic is.
00:32:33
AI and sci-fi, right? Obviously a number of the most
00:32:36
successful films, you know, TV shows, books, everything in
00:32:40
history have been built around sci-fi, which often is driven by
00:32:44
this idea that there's an evil killer AI basically, right?
00:32:46
And I think the examples that come on top of my mind, and you
00:32:49
guys might have other ones, are like, you know The Matrix, you
00:32:52
know I Robot, as you mentioned the Terminator series, Blade
00:32:56
Runner. Blade Runner I you know, her
00:32:59
being a sort of off kilter example, but still important I
00:33:02
think. And what's interesting is almost
00:33:04
all of these ones other than her that I mentioned, it involves
00:33:08
robots as well as AI, right? It's it's like the humanoid
00:33:12
version of the AI is essential to you know, killing us all
00:33:17
right, we're we're in Terminator.
00:33:18
There's a good. Storyteller.
00:33:20
Yeah, right. Well, yeah, it might just be
00:33:21
more cinematic to fight a big metal robot.
00:33:23
So, you know, that's something. But I guess.
00:33:26
And then in her you sort of have this.
00:33:29
Scarlett Johansson AI That. At the end kind of ascends to a
00:33:33
higher plane and like leaves leaves the main character
00:33:38
Joaquin Phoenix behind, right. And I think that I to me that
00:33:43
representation seems more likely than the killer robots one and
00:33:46
that the as I said maybe we're just pets and they just sort of
00:33:49
get bored with us and and leave us behind or or we're the little
00:33:51
you know dogs that they leave alive right.
00:33:53
But I guess, what do you guys think about, well, A, what's
00:33:56
your favorite sci-fi representation?
00:33:58
And then B. Do you think robots are an
00:34:02
essential part to these AI kill us all narratives or and or will
00:34:06
the robot elements integrated with AI happen anytime soon.
00:34:10
Well, I I I can answer with my favorite.
00:34:12
I mean I think my the matrix matrix is my favorite and.
00:34:16
I was going to say The Matrix. That's all The Matrix buys into
00:34:19
exactly what you're saying. The Matrix doesn't actually say
00:34:21
it's killer robots. It it finds a way to basically
00:34:24
put killer robots on the screen. So it's a fun movie, but it's
00:34:27
like, oh, it's a computer program, which is obviously the
00:34:29
world I feel like a super intelligence would live in.
00:34:33
Like that of what does it really need to manifest in the real
00:34:37
world that much? So yeah, I, I'm.
00:34:40
And obviously I'm interested, you know, for the following
00:34:43
conversation with somebody who's written a lot about robot
00:34:46
apocalypse. Because yeah, my intuition is
00:34:48
that they're going to exist mostly mostly in the digital
00:34:51
world. Terminator two is the other one
00:34:53
where I think the robot part is interesting, but I don't know
00:34:58
how much you remember about the historical world building of the
00:35:02
AI takeover which is Skynet is an AI that achieves super
00:35:06
intelligence that then. Takes over the nuclear arsenal
00:35:10
of the United States or and then bombs Russia or whatever, and
00:35:14
then we end up in nuclear apocalypse basically, right?
00:35:16
So the initial manifestation of the AI apocalypse in in
00:35:20
Terminator and Terminator Two is is a purely software driven
00:35:25
death scenario, right? There's no, there's no need for
00:35:27
robots in the sort of Skynet back story, right?
00:35:31
It can hack into a system that ends up destroying the world,
00:35:34
which like. Realistic, right?
00:35:35
Because then you don't need any physical elements, right?
00:35:38
Of of the sort of AI doomer narrative, right?
00:35:41
You don't need the robots, you don't need the the crazy
00:35:43
biohacking stuff or whatever. You just need to hack into the
00:35:47
Pentagon essentially. And it's like game over at that
00:35:49
point, right? So I always thought that was a
00:35:52
very insightful perspective on how an evil AI could kill us all
00:35:57
and then it would just be a pure software takeover, right?
00:36:02
James, did you have a favorite? Well, I I wanted to dig.
00:36:05
Into the Matrix a little bit more, yeah.
00:36:08
But I think, you know, a lot rests on whether we think that's
00:36:14
humanity's future as well. Like are we going to be like
00:36:18
almost, you know, by choice plugging ourselves into the
00:36:23
Matrix? Like is is this neural link kind
00:36:25
of interface you know within that 50 to 100 year time frame
00:36:29
or or even sooner, right? You know does does our normal
00:36:34
day-to-day life as humans you know become more of a simulation
00:36:40
you know in in in our lifetimes. I think that's pretty
00:36:43
interesting for how the future unfolds.
00:36:48
So do you guys believe that is possible like or do you think
00:36:53
you know brain interfaces are sort of so far away.
00:36:56
I think Brain Interface is 30 + 30 plus years. 30 plus years
00:37:01
away. I kind of agree.
00:37:04
I mean, I think it's possible maybe in our lifetime, which,
00:37:08
you know, God willing is 50 to 60 more years, but it does seem
00:37:12
like we're pretty far away on neural interfaces.
00:37:15
But you know, who knows? I feel like I'm I'm more excited
00:37:18
to come back to the previous episode about like glasses and
00:37:22
maybe contact lenses at some point as a sort of, you know,
00:37:25
augmented interface for for existence rather than like.
00:37:29
Plugging in, but you never know. I mean, obviously Elon believes
00:37:32
in it. So Elon believes like this is.
00:37:34
The path right out of the AI dumerism that we've that we've
00:37:38
been talking about, We are augmented.
00:37:40
Basically. We sort of merge our brains.
00:37:42
Yeah, with with AIS and that probably requires some sort of
00:37:47
brain interface or you know matrix like environment that
00:37:50
you're living in seems hard to be able to.
00:37:53
I agree I'm not. I'm not a problem.
00:37:55
Sort of like the you you think that that basically?
00:37:59
Is not sufficient to. I just don't think it answers
00:38:02
much. The like, will it kill us?
00:38:03
Like, oh, it's like yielding our brains to it almost means like,
00:38:07
are we even running the show like it?
00:38:08
It raises a whole bunch of other questions where we could be
00:38:11
undermined just from the connection.
00:38:13
Independent. Like it's true, Yeah.
00:38:16
And it also it also like pretty quickly if you believe in super
00:38:19
intelligence kind of. Defeats the purpose of having
00:38:22
that brain power. Like why?
00:38:25
Why do you need? Yeah, why would they?
00:38:26
Why would they want? To be hanging out in our brains,
00:38:28
it's like, it's like, oh cool, I've got this brain that's not
00:38:31
as good as mine. Like, maybe consciousness is
00:38:34
this unique human charm that nobody else has, and we can give
00:38:37
the machines a taste of it. I don't know, just succinctly.
00:38:42
Do you think in the next 100 years AI will kill us all?
00:38:46
Yes or no? I will say no.
00:38:52
I think it will kill. Some of us, but not all of us.
00:38:56
That's a good way to. Think that's a good Yeah.
00:38:57
I I think like more than 1000, less than 10.
00:39:04
Oh, more I don't. Know I think more that's a very
00:39:07
tight. Yeah.
00:39:09
That's not a good range. I I think I was smart to give
00:39:11
it. Yeah.
00:39:12
Yeah, yeah, yeah. Some of well, you came up with
00:39:13
some of us. Some of us is the right feels
00:39:15
like clearly. It feels like even a mistake.
00:39:17
I will AI intentionally well it was kill us all.
00:39:21
I say definitely no kill us all that's.
00:39:23
I say definitely there'll still be some humans at the.
00:39:25
End I can't say definitely. I mean it seems pretty UN, you
00:39:30
know, well if we do all get killed, we'll be dead and so
00:39:33
there's nothing to gain from the prediction.
00:39:35
Whereas if we live, I was correct and so it was there was
00:39:39
good utility in the prediction. So it's I don't know why I would
00:39:41
ever go the other direction. This is like Pascal's wager.
00:39:44
With like super intelligent AI, like, all right, this was fun.
00:39:50
I mean, this is this is a dream. Getting to hang out and talk
00:39:53
about this and call it work. Welcome.
00:39:59
Hey, welcome. To the second segment, here I've
00:40:02
got author Daniel H Wilson, author of How to Survive a Robot
00:40:06
Uprising, Where's My Jetpack, How to Build a Robot Army, and
00:40:11
Robo Apocalypse. The title of this or at least
00:40:15
the working title of this episode is Will AI Kill us All.
00:40:19
So given your interest and and work we wanted to talk with
00:40:24
someone who's really like been thinking about this for a long
00:40:28
time. I'm I'm curious like what what
00:40:31
first got you interest in this sort of dystopian question of
00:40:35
sort of the machines coming for humanity.
00:40:39
Yeah, well, so I just grew. Up with science fiction, right?
00:40:42
Like, like a lot of people, so. Initially, I was just interested
00:40:45
in reading, you know, any type of science fiction I could read
00:40:49
or watch in any type of movie. I just loved robots.
00:40:52
And so that's ultimately though I loved robots so much that I
00:40:57
studied robotics. So I ended up going to Carnegie
00:40:59
Mellon. I did a a whole PhD in robotics.
00:41:02
And while I was at Carnegie Mellon, you know, I'm surrounded
00:41:05
by roboticists, I'm surrounded by robots.
00:41:07
We're in the high Bay world. We're just in the lab.
00:41:11
And nobody was really trying to build robots that would destroy
00:41:15
the world. I noticed.
00:41:17
Nobody says they're trying. Generally, nobody comes out and
00:41:20
says. It but like, there's this really
00:41:22
stark difference between how robots are portrayed in pop
00:41:27
culture and how they're and and the actual you know, the actual
00:41:32
mechanics of building robots and why people are building them and
00:41:34
how they're explaining, you know why they need the money to build
00:41:37
them and all that and so. Really.
00:41:39
They just have this super bad, you know, reputation.
00:41:41
Right. And so I thought that was funny.
00:41:45
So then when I was still in grad school, I wrote How to Survive a
00:41:48
Robot Uprising, where I was just like, all right, I'm gonna take
00:41:51
them serious, right? Just, OK, fine.
00:41:53
If this is what everyone's expecting.
00:41:55
So I went to, you know, the people that were building legs,
00:41:59
and I asked, you know, how would you trip a robot?
00:42:01
How would you get away? I went to the people doing
00:42:04
sensors, the people doing all different things.
00:42:07
And I just asked those questions and I put it all, you know,
00:42:09
tongue in cheek into this book. Of course.
00:42:13
Then you start to look at like actual military applications.
00:42:15
You start to see how robotics and AI are, are, you know, being
00:42:20
weaponized in some cases. And then, you know, that was
00:42:23
more robo apocalypse much. No tongue in cheek like I I was
00:42:27
like, OK, let's. You're starting to believe it.
00:42:29
You're like, OK, this. No, I never, really.
00:42:32
I mean, honestly, I like. I like.
00:42:34
The killer robot meme really, for me, just gives you a lot of
00:42:38
latitude to to to think about humanity and think about, you
00:42:42
know, what makes us people, what makes them robots.
00:42:45
You know, do you do? You do you think like as a
00:42:49
person and somebody? Do you think robots are likely
00:42:51
to kill human beings? Or, like, do you see this as a
00:42:54
fictional actress? No, I well, look, I.
00:42:57
I kind of ascribed to the Lately I've been thinking a lot about
00:43:00
something called the psychotic ape.
00:43:02
Theory or the the killer ape theory, which is think about the
00:43:05
opening scenes of like Space Odyssey 2000, 2001.
00:43:09
Like it's those the, you know, they touch the the monolith.
00:43:13
It gives them a leap forward in evolution.
00:43:15
And what do the apes do? Well, they figure out how to use
00:43:17
weapons to bash each other's brains in.
00:43:19
And there's this kind of notion that humanity sort of evolves
00:43:23
technologically when we're trying to kill each other or
00:43:26
stop each other from killing each, you know, and so.
00:43:30
So, yeah, so I've been thinking a lot about that and and
00:43:32
basically you look at that and you realize we will use any
00:43:35
technology to kill each other. So really it's the psychotic ape
00:43:38
that you need to worry about. It's not the and to a lesser
00:43:42
extent the, you know, the capitalist ape, the known killer
00:43:46
species versus the sort of imagined killer who's killed
00:43:49
more people than anybody. Else, right.
00:43:51
You know you're going to want to look at the yeah, the person
00:43:55
across the table from you, but. I think that in terms of being
00:43:59
used as weapons, you know, obviously robots can be
00:44:03
extremely dangerous in that way. I mean I'm curious sort of you
00:44:08
know as a technologist somebody's thought about this
00:44:10
like robots versus like the large language model, sort of
00:44:15
totally software based being that maybe reaches super
00:44:19
intelligence like I I see why for like especially film, why
00:44:23
like robots are great because you can see them.
00:44:25
I'm curious in terms of like the actual like thought exercise and
00:44:28
what you think is likely whether you imagine robots or the sort
00:44:33
of potential threat being just a software.
00:44:36
Yeah, well, so I think of this as like a.
00:44:38
Consumer like a like a product design issue, right?
00:44:41
Like, you know, I mean, look, we're human beings are
00:44:44
comfortable with a certain amount of risk in our lives.
00:44:46
You know what, 40 people get killed driving around in cars,
00:44:51
you know, every year I've got a garbage disposal in my kitchen.
00:44:54
I mean, if I put my hand in the wrong place in my own home, like
00:44:57
I won't have a hand anymore, right?
00:44:59
But so but there's also the. Devices are built in order to
00:45:04
try to make them as safe as possible, right?
00:45:06
So if you think of that as a hardware problem on the hardware
00:45:09
side, you know, you're trying to design A consumer product that's
00:45:12
not gonna harm people. And that's just the, I mean we
00:45:16
people do that every day, right? We've been doing that for years
00:45:18
and years. It's a, it's a really under well
00:45:20
understood kind of task. And I think if you're building
00:45:23
hardware, it's an easier task because then when you move to
00:45:27
software it becomes much more complex, right, in terms of.
00:45:31
What the harm might be. So for instance, right now I
00:45:34
live in Portland, OR and in Seattle up the road, there they
00:45:38
are. The the school system is suing
00:45:41
like meta, right, because they have documented harm that using
00:45:46
the social media has harmed children.
00:45:48
It's like the Phillip Morris thing again, right?
00:45:50
I mean, they know for a fact using this product causes harm
00:45:54
to children. They die.
00:45:55
They commit suicide. So.
00:45:58
If you identify that, I mean that that takes a little while
00:46:01
to put those to connect those dots, right?
00:46:04
It's not the same as if they were just selling toasters that
00:46:07
were electrocuting people like you understand pretty clearly
00:46:10
like, OK, where's the danger there, right.
00:46:11
I mean, you can see if a robot is causing physical violence
00:46:15
against and so here's where it. Gets like even more complicated,
00:46:19
right? So, so let's say by the way,
00:46:21
we've got chat bots. Purely software.
00:46:24
Yes, they're going to. People are going to get LED down
00:46:26
crazy rabbit holes by these things and there's going to be,
00:46:29
I predict, extremely harmful scenarios occurring from people
00:46:35
just being told whatever they want to hear and and eventually
00:46:38
being potentially radicalized or whatever causing harm in the
00:46:41
community. It won't be the robot doing it
00:46:44
directly, but it, but it or the sorry, the chat bot or the large
00:46:47
language model. But then you think about the
00:46:49
synthesis of these two things as well.
00:46:51
So like think about. Like a self driving car.
00:46:55
So now I mean it gets really complicated because you've got
00:46:58
brains and you've got the hardware.
00:47:00
And so the question becomes like, OK, the car crashed, like
00:47:04
whose fault is that? And so right now that's
00:47:06
something we're working out, you know in the court systems and
00:47:09
and we're using all of our existing machinery of our
00:47:12
society to try to sort that out and figure that out.
00:47:17
So, you know, I I would say those are really the three
00:47:19
areas, you know, pure software, the hybrid and then just the
00:47:22
purely mechanical problem of of to have like a row apocalypse or
00:47:28
like to have robots like it, it requires like the idea that
00:47:32
there's some like super intelligence, right.
00:47:34
I mean, all of that that's predicated on the idea.
00:47:36
Yeah, in pop culture, you gotta. Have the singularity before.
00:47:39
I mean, that's just kind of like ticking the box, right?
00:47:42
I don't know how realistic that is, but.
00:47:45
Right. Well, that's I guess to me like,
00:47:47
you know, I know it's fiction, but like, yeah, it just feels
00:47:51
like if we have this super intelligence is isn't that a
00:47:54
threat enough without it taking sort of robotic form or I guess
00:47:58
particular question, you know, there's all.
00:48:00
These predictions about when the singularity's gonna happen and
00:48:03
how it's gonna be exponential, so it's gonna go slow and then
00:48:06
happen fast and. And like it's really interesting
00:48:09
because you know I was, I have a degree in machine learning.
00:48:12
I mean I was studying it before they were calling it machine
00:48:14
learning like when it, it was called knowledge discovery and
00:48:17
data mining and all these different.
00:48:19
And so you you look at that and what was happening early on was
00:48:23
there were all these different approaches to to trying to mimic
00:48:26
intelligence, right. And we would use a whole suite
00:48:30
of these things. And what happened was neural
00:48:32
networks in the last few years jumped out ahead.
00:48:36
And I would argue, although I'm sure other people would argue
00:48:39
with me, but I would argue there was no amazing scientific
00:48:43
breakthrough, right? What happened was processors got
00:48:47
faster and we had access to just a ton of data.
00:48:50
And so that data and those processors just are brute
00:48:54
forcing what are fairly simple algorithms that have been well
00:48:58
known for a long time and they're getting this kind of
00:49:03
intelligentish behavior out of it.
00:49:06
And so that model isn't really doesn't really sync up with the
00:49:11
way I think people thought AI was going to go, right.
00:49:14
We didn't think, oh, we're just going to use some old algorithm
00:49:17
and just throw more processing power at it until it gets good.
00:49:19
Like there's more GPUs. We're going to get it, yeah.
00:49:23
And so I. Don't know if it's.
00:49:25
So what I'm trying to say is, if you look at that and we just, I
00:49:29
mean the way to improve chat bots I guess is to keep throwing
00:49:33
more data. And keep throwing more
00:49:35
processing power. But I don't feel like there's
00:49:38
any type of singularity that's going to come out of that.
00:49:40
I feel like it's like, Oh yeah, it's just, I feel like that our
00:49:42
ChatGPT, yeah Or yeah, it'll. Just be like a more convincing
00:49:46
ChatGPT, it's just really, I mean it's just really just
00:49:51
regurgitating everything that it's, you know, in a very smart
00:49:54
way that it's read on the Internet, which by the way, I
00:49:57
mean, God, that's the worst of humanity, right?
00:50:00
I mean. What's it got?
00:50:02
It's trained on I know by the way, right?
00:50:03
They're all desperate to train it off.
00:50:05
Reddit, for instance. Yeah.
00:50:07
Oh my God. You know, Reddit can be OK, but
00:50:11
there's smart people there. But yeah, but you're looking at
00:50:14
the Internet and it's like, oh Lord, but but anyway, I don't
00:50:17
really, me personally, I don't see that.
00:50:19
I see that as a kind of a dead end.
00:50:21
I don't feel like that's headed toward the singularity
00:50:24
necessarily. Yeah.
00:50:25
Is there any existing dystopian work that you find most
00:50:29
plausible or like what? I mean, you know, there's so you
00:50:33
know, I've we talked there's The Matrix.
00:50:34
There's sort of the Terminator movies.
00:50:36
Like I don't know, like, yeah, I'm sure you've spent a lot of
00:50:39
time in that world. Like are is there one that you
00:50:42
find like, oh this this is the most sort of real world?
00:50:46
Well, I mean. I don't think that the purpose
00:50:49
of science fiction is to predict the future necessarily, but I
00:50:53
would say you know, if you're thinking about.
00:50:57
I think the most boring stuff is actually the most realistic, and
00:51:01
in a lot of ways look for the bad books that no, I mean no
00:51:04
actually. That's not true.
00:51:05
Like for instance I would say Terminator, which is obviously
00:51:08
great, but think about that. Villain think about Skynet.
00:51:13
Skynet's just a dumb, dumb computer program that just has
00:51:17
decided whatever it wants to destroy all humans.
00:51:20
I mean, and there's no reason for it.
00:51:22
There's no It's just like it just wants to kill everybody.
00:51:26
I mean that is so boring. You couldn't get away with an
00:51:28
actual human antagonist who is that simple?
00:51:32
People would be like, well why like and so like if you think of
00:51:39
that as its end goal, that's not very super intelligent, is it?
00:51:42
Kill all humans? I mean that's like Bender level
00:51:45
right now. So I would say that you know
00:51:49
that in a lot of ways. It is fairly realistic if
00:51:51
somebody just programmed a computer to tell it.
00:51:54
Just super simple boring goal like that.
00:51:57
Kill everybody like then you know I feel like that's fairly
00:52:00
realistic. To step away from the yeah
00:52:04
dystopian question for a second I'm curious just like as an
00:52:07
author one like are you using ChatGPT at all like in any way
00:52:13
or any other sort of LM tools? Well so first of all I.
00:52:19
All of my novels are in the training data, which means
00:52:23
hopefully I'll be part of a lawsuit.
00:52:24
Yeah, I know. I was going to ask you about
00:52:26
that next. OK?
00:52:27
I asked. I went in the chat.
00:52:28
GPT and I said, hey, write me a short story in the in the vein
00:52:32
of Daniel H Wilson. And it, like, immediately wrote.
00:52:34
I didn't say anything else. And it really wrote about
00:52:36
robots, right? It wrote, it wrote science
00:52:38
fiction. It was really clear that it had
00:52:40
been trained on my stuff. I'm like, tough.
00:52:42
Well, that's bullshit. First of all, I never gave
00:52:44
anybody permission to do that. And so second of all, like, I I
00:52:49
think it's so funny. So I had I'm not going to name
00:52:52
the corporation because I don't want to get into trouble.
00:52:54
I had a large large corporation approached me before ChatGPT
00:52:59
before Open AI released before the the these Gans became like a
00:53:04
big deal. So this is maybe probably.
00:53:08
Six months to a year before everybody knew about that and
00:53:12
that that came out and Open AI really just broke the whole
00:53:15
deal. And they it was a researcher
00:53:18
that called me and they wanted to know if I would test out this
00:53:21
new program that they had that was basically ChatGPT and and
00:53:25
they said we want you to use it to help you write.
00:53:30
And I'm like OK, so then what How will it help me write?
00:53:33
And they say well. And and these are just such
00:53:36
sweet researchers that don't have any GD idea about what the
00:53:40
real world is, is hell is going on in real world.
00:53:43
And they're like, it'll help you be creative.
00:53:46
And I'm like, oh, creativity, right?
00:53:50
That thing. I hate doing that thing that I
00:53:53
spent the last 20 years teaching myself to take.
00:53:57
What's in my head? And put it in the world by
00:53:59
memorizing all these really boring writing skills which are
00:54:02
just pure hell. And then they want to come in
00:54:04
and take the one good thing. I'm like, why do you think
00:54:07
writers write? You think because we love
00:54:10
typing. We like pushing letters.
00:54:13
I'm like, what the hell are you talking about?
00:54:15
I was like, you need to be prepared for everyone that you
00:54:18
give this tool to, to give you super negative feedback.
00:54:21
Because they wanted to include me in, like, this report about
00:54:25
how this tool would be used, and they wanted to publicize this
00:54:27
report as well. And so they they went back and
00:54:31
they came back and they said, hey, we got great news.
00:54:33
I got permission from, like, the program manager on this to
00:54:36
include your feedback, even if it's negative.
00:54:41
And I was like, you know what? I'm gonna just take a pass on
00:54:44
this. I wish you luck with your
00:54:45
project. Right.
00:54:46
So, man, no. You know what I think about,
00:54:50
man? The thing that cracks me up is I
00:54:54
used to really hate on. Asimov because I was like, robot
00:54:59
psychologists, What kind of bullshit is that?
00:55:01
Like, you program robots, right? I spent all this time, you know,
00:55:04
programming and learning all of these programming languages and
00:55:07
and everything you got to do in order to in order to speak to
00:55:11
the robot mind, because it doesn't speak English, right?
00:55:14
And I used to think that the idea of a robot psychologist was
00:55:17
the dumbest thing ever. The idea of a positronic brain,
00:55:20
by the way. Which is made fully and it's
00:55:23
just boom, it's done. It's crystallized.
00:55:25
You can't go back and change it, right?
00:55:27
And that's and. And by the way, that's exactly
00:55:29
what happens. So neural networks are black
00:55:31
boxes. You can't go in and fiddle with
00:55:33
it because all the weights make no sense, like the human brain.
00:55:36
So it becomes like psychology and then now if you.
00:55:39
Want to for instance? Yeah.
00:55:40
If you want to game a like a like a chat bot, if you want to
00:55:46
trick it into doing something, you totally have to use
00:55:48
psychological things. It's totally psychology.
00:55:51
I mean, Bravo. Asimov.
00:55:54
Bravo, Sir, Are you? Do you think you will, like, try
00:55:57
to sue? Like if somebody comes to you
00:55:59
and says we're going to sue these guys for.
00:56:01
Oh yeah. Oh, absolutely.
00:56:02
No, that this has to happen. Look, I mean, I I get people
00:56:07
trying to make money. I get, you know, capitalism.
00:56:11
I understand it. And this is a case where you
00:56:14
know, they're they're they have to be sued.
00:56:17
I mean, we're just because we're creating new, this is a
00:56:19
completely new domain. And if not, OK, to rip people
00:56:23
off and then chop up their stuff and train an algorithm.
00:56:26
And so our country or whatever is going to have to figure that
00:56:30
out and get that down in law. And that's going to require,
00:56:33
yeah, courts, I mean, so absolutely they're all going to
00:56:35
get sued. Do you do you think?
00:56:39
Sci-fi and dystopias are bad for AI like even talking to ChatGPT,
00:56:45
it's funny when you ask it to imagine things it's been trained
00:56:48
on all that. Exactly it parents all that.
00:56:50
And so it can sound really spooky like it feels like you
00:56:54
know you're like I'm you know this is imagination it's fiction
00:56:57
like. But I do think people, you know,
00:56:59
seriously turn to science fiction when they try and game
00:57:02
out where sort of a technology we don't really understand,
00:57:05
that's developed faster than we expect is progressing.
00:57:09
So yeah, is it, is that bad in some ways?
00:57:11
Do you do you regret that? I think it says a lot more about
00:57:15
us than it does about them them being the robots, you know, like
00:57:20
what I've been thinking about lately.
00:57:21
So I don't know know if you've read my books, but I I'm from
00:57:25
Oklahoma. I'm, I grew up in the Cherokee
00:57:27
Nation. I'm a Cherokee citizen and I
00:57:28
write a lot. There's a lot of native
00:57:30
characters and stuff I do. And so I think a lot about
00:57:33
technology from like a native perspective.
00:57:35
And one thing I've been thinking a lot about lately is just kind
00:57:39
of like how a lot of this super negative, like robots and also
00:57:44
first contact, like aliens, they always show up and what do they
00:57:48
do, man, they do exactly what all the colonizer civilizations
00:57:54
did to indigenous people all over the world.
00:57:56
So we got a civilization, we got a society, a culture.
00:58:00
It's all built on colonization, right?
00:58:03
So it means. A bunch of people with superior
00:58:06
technology showed up someplace, murdered everyone, destroyed
00:58:11
their culture. I mean, Independence Day,
00:58:13
they're blowing up monuments. So you think that is like
00:58:17
resource extraction, right? They're they're what?
00:58:19
Are they stealing our water or they're stealing our air or the
00:58:22
and then just completely dominating other people's
00:58:25
bodies. You know, you think about
00:58:27
Invasion of the Body Snatchers. And just like so, that type of
00:58:31
fear, I think, is cooked into our civilization.
00:58:35
Based on our origins and it comes seeping out into our pop
00:58:39
culture in a lot of different ways, but in particular in
00:58:43
science fiction, we see it so much.
00:58:45
Do do you think we should stop? You know, there are even there
00:58:49
are AI companies that say, oh, we should pause for six months.
00:58:52
I mean, originally it sounded like you were saying you didn't
00:58:54
think just adding more, you know, compute was gonna like
00:58:58
create this super intelligence. But yeah, do you, do you have
00:59:01
the impulse that, you know, we should slow down or stop this
00:59:04
research if we don't know what's going to happen?
00:59:06
Yeah, Well, I don't think that. That's what they're afraid of.
00:59:08
I don't think they're afraid of us crossing the singularity
00:59:11
threshold or something. I mean, they're just worried
00:59:13
about. So for instance, I yeah, I will
00:59:16
tell you that the US military is actively doing a lot of threat
00:59:21
scenarios that involve this kind of technology being weaponized,
00:59:26
this sort of disinformation on a mass scale, hugely distracting
00:59:31
events that could occur. I mean at this point I mean you
00:59:35
could get, you could get a phone call from someone it could be
00:59:38
just like in robo apocalypse, you know where everybody gets a
00:59:41
phone call from somebody that they're related to and they
00:59:44
trust and they all get told to go to a certain is a bad, bad
00:59:48
situation. I mean, all that stuff is
00:59:50
starting to. Right.
00:59:53
I really focused on this idea of it acting independently, but
00:59:56
obviously, just like bad people aren't, no, this is always going
00:59:59
to be bad. People and and so, so first of
01:00:01
all I don't think that's they're not worried about that
01:00:02
singularity. They're worried about this being
01:00:04
weaponized And in terms of putting a pause on it, I mean,
01:00:08
yeah, I mean why not. I I would say only put a pause
01:00:12
on it until we figure out the the rules around it.
01:00:16
But but the fact is this like that's not how the United States
01:00:19
works. Like, we don't fix anything
01:00:22
until it's broken, right? I mean, there's not I, I just
01:00:25
don't have a lot of confidence that they're going to say, OK,
01:00:29
six month pause and we're going to work out all the laws and
01:00:32
everything's going to be all ready to go in six months.
01:00:35
Like, hell no, that's not going to work.
01:00:37
So I mean, I think it's a a nice idea, but I'm I'm extremely
01:00:42
dubious and skeptical that could ever have any actual, you know,
01:00:46
useful. I mean other countries.
01:00:48
Are still going to work. I mean, there's the, you know,
01:00:49
sort of China boogeyman where it's like you can't stop the
01:00:52
whole world from working on it. And so there's that, honestly,
01:00:55
though, there's the there's the, the other side of that coin,
01:00:58
which is, you know, thank God that they're actually interested
01:01:00
in privacy in Europe, right? Like, at least there's some
01:01:04
people standing up and saying, hey, like, we don't just have to
01:01:07
accept this status quo, just 'cause you're a giant billion
01:01:10
dollar corporation or billion trillion dollar corporation,
01:01:14
just. Something you referenced
01:01:15
earlier. Does the military come to you to
01:01:17
help game out sort of technological?
01:01:20
Yeah, I've done a little bit of that.
01:01:22
It's pretty fun. So I'm not the only one who's
01:01:25
like. Oh, we should like, get some guy
01:01:27
who thought about it. Sort of fiction in fiction.
01:01:30
Just sort of say what happened in the real world.
01:01:32
Yeah, I'm not. The only one, But there are,
01:01:33
there are and and honestly, they're training their own.
01:01:36
They're training their own science fiction, right?
01:01:38
I will tell you this like if you are, if you're a general in the
01:01:42
Air Force or or whatever, if you're some kind of higher up in
01:01:45
the military, you get a lot of white papers that are describing
01:01:49
the technological capabilities of the enemy or or of various
01:01:52
munitions and things like that. And these papers are very dry
01:01:56
and just like very, you know, it's much better.
01:02:00
I think at least it's useful to have somebody actually
01:02:05
creatively tell a story that's gonna stick with you, cuz that's
01:02:08
the way humans communicate, right?
01:02:09
Through stories. And then you and then the
01:02:12
general is sitting there thinking, OK, now I've I've just
01:02:15
read a story about, you know, an actual person being impacted by
01:02:19
this technology in that way. And you can really visualize
01:02:22
what it means. And it's more than just facts
01:02:24
and figures when you say they're the military.
01:02:26
Is trained science fiction writers.
01:02:27
You mean those authors of the boring white papers?
01:02:30
Or you mean like true science fiction writers?
01:02:32
No, I mean that they they're literally you can yeah if you're
01:02:36
in the military you can There are certain programs you can
01:02:38
sort of sign up for where you get taught.
01:02:41
Yeah. Where you get they bring in
01:02:42
science fiction authors to, to help teach the the, the military
01:02:47
people how to sort of, yeah, take those super real world
01:02:51
scenarios and write them in an engaging way so that you have
01:02:54
sort of a case study and you say, look, here's one way this
01:02:57
could be employed. And that's just good.
01:03:00
That's just good science fiction writing.
01:03:01
That's how that is. But they would prefer not to
01:03:04
outsource it to, you know, hippies like me.
01:03:09
I'm curious. How much sort of the
01:03:12
technological developments we're seeing right now like you know
01:03:15
mid journey, whatever are influencing what?
01:03:19
You plan to write or do you think, like science fiction will
01:03:22
change based on what we see is possible?
01:03:26
Yeah, I mean, absolutely. Is this changing science fiction
01:03:29
every day? For sure.
01:03:31
And and and look, I think it was a watershed moment.
01:03:34
Like whenever you really just finally just talk to it and you
01:03:37
realize, oh, this thing is like, yeah, this thing can just talk
01:03:41
to me. And by the way, right now
01:03:43
there's this huge rush, as, you know, like to plug this stuff in
01:03:48
anywhere you can. It's like, oh, I don't know,
01:03:50
what is it? It's like if it's like
01:03:52
mayonnaise, you know, they're just putting it on.
01:03:53
We just discovered this thing. Let's put it on everything.
01:03:55
Right, right, right. It's desperation, yeah.
01:03:58
It's going to taste like crap on a lot of stuff, but like it's
01:04:01
going to work out in some places.
01:04:03
One thing I see which I find interesting is it occupying as
01:04:09
almost a homunculus, right? You think of a homunculus as
01:04:12
like a little tiny person that's inside of you kind of driving
01:04:15
you around or you think of like, who's the bad guy from
01:04:17
SpongeBob? You know, he's always driving a
01:04:20
giant robot, right, The little guy.
01:04:22
So if you think of the ChatGPT and you have like a it's like
01:04:27
the homunculus. You stick it inside of a
01:04:28
hardware platform so you can just tell ChatGPT here's a super
01:04:33
simple programming language that allows you to drive around this
01:04:38
autonomous vehicle or allows you to drive around this humanoid
01:04:42
robot platform. And then you tell it in English,
01:04:46
hey, go make a sandwich. And then it looks at all of its
01:04:50
little controls and it translates that and says, OK,
01:04:53
I'm going to drive this thing in there because it kind of get has
01:04:56
the right idea about what making a sandwich is.
01:04:58
But a but a hardware platform is just a bunch of like XYZ
01:05:03
coordinates where you're going to put your limbs and it's and
01:05:05
that translation right there, that's a spot where where
01:05:09
ChatGPT is kind of getting just like plastered in, you know,
01:05:12
just slap it in there and it solves all those problems.
01:05:15
And so that's kind of interesting because now you're
01:05:18
going to see ChatGPT having the ability to interact with the
01:05:21
real world via like whatever delivery bots or like or drones
01:05:27
or or humanoid robots. And so that's where you maybe
01:05:31
get into a little bit of consumer trouble.
01:05:33
Again, it's like an autonomous car, right?
01:05:36
Is there? Anything you've written where,
01:05:38
based on how technology has developed, you feel like, oh, it
01:05:42
feels less plausible than when I wrote it?
01:05:44
Or like how much like forecasting where where what's
01:05:48
possible do you see as like part of your yeah, well, I think
01:05:53
that, you know, what happens is you look at what's out there in
01:05:56
the world and you run with it and you start thinking of all
01:05:59
the different ways it could be. And that's where science fiction
01:06:01
is fun, right, 'cause you suddenly you're like, I never
01:06:04
really thought of it like that. But but there it is.
01:06:06
And dystopias are fun too, because, you know, it's
01:06:09
dangerous. It's exciting.
01:06:11
And so the difference between my writing and and is that I had
01:06:17
this degree in robotics. I all my friends from when I was
01:06:20
20, they're all running corporate robot corporations.
01:06:24
They're driving the Mars Rovers like that so that my cohort, you
01:06:28
know, I'm like my all my old friends, that's what they're
01:06:31
doing. And so I feel like I have a
01:06:33
little bit of a sneak peek. So like for instance when I
01:06:35
wrote Robo Apocalypse, that was 10 years ago or whatever, and it
01:06:40
I was watching and seeing the technology that was 5 or 10
01:06:43
years out. So it's all come true.
01:06:46
In fact, that's the joke because it's still Robo Apocalypse is
01:06:49
still with Spielberg at Amblin and and you know, development
01:06:53
continues on this movie. But like what we've what we've
01:06:56
told Spielberg is hey man, let's do this before it becomes a
01:07:00
historical documentary. All of this stuff is, you know,
01:07:04
so in in Robo Apocalypse there are and Robo Genesis the sequel,
01:07:07
which came out a few years later.
01:07:09
There are autonomous vehicles. There is essentially ChatGPT as
01:07:14
as personal assistants on phones.
01:07:16
I mean I wrote that before before Siri came out.
01:07:20
Wow. It was 2000.
01:07:22
The book came out in like 2011 or something a long time ago.
01:07:25
But anyway, so yeah, I mean my latest novel is that the
01:07:31
Andromeda Evolution which is a sequel to Michael Crichton's The
01:07:34
Andromeda Strain, which I he passed away and I did this with
01:07:37
in cooperation with his, with his estate.
01:07:40
And and you know that one was again like very dialed in in
01:07:43
terms of just because it's Crichton, the the government
01:07:47
like exactly which scientists would be going where how they be
01:07:50
chosen and and and I mean just so so locking in those details
01:07:54
is is really important for a, for a techno thriller, you know
01:07:58
that certain genre of science fiction do you do you give in?
01:08:01
Sort of, yeah. The professional need to be
01:08:04
forward-looking. Do you have any sort of
01:08:08
predictions or about where like culture is going or what what
01:08:13
you think, what you really think will be the more sort of
01:08:15
essential change like it. I mean it does feel like, you
01:08:18
know, like self driving cars and stuff like that.
01:08:21
Feel feel close? Certainly in San Francisco.
01:08:24
I mean, I think culturally human beings are always the wild card.
01:08:29
The technology is not that hard to predict usually.
01:08:31
I mean, you can usually see what people are up to 5-10 years out,
01:08:34
you, you know what what they're going for, right?
01:08:37
But then with human beings, you never know, like autonomous
01:08:39
vehicles. I was much more sort of bullish
01:08:41
on that. I thought those things would
01:08:43
already be here, being used in a bigger way.
01:08:47
But as it turns out, one person gets killed and humans freak
01:08:51
out, even though 40 or 50 people are dying every year with
01:08:54
Inhuman human. But then the robot screws up
01:08:57
once and everybody loses their mind and they shut them all
01:08:59
down. That's like, I didn't really
01:09:01
predict that. I thought we had a higher
01:09:02
tolerance for for that or or we cared less so like.
01:09:07
So it's really tough to predict the human element.
01:09:09
And I think that with Chachi PT, man, things can go a lot of
01:09:13
different ways, right? The movie Her is interesting in
01:09:18
terms of falling in love with it.
01:09:19
I think that's totally going to happen.
01:09:22
It can, It just tells us what we want to hear, right.
01:09:25
And so I think that that is just so dangerous and just such a
01:09:29
siren song. For instance, the researcher who
01:09:33
thought that that that at Google.
01:09:37
Like something? Yeah, he thought that.
01:09:39
That program was sentient. And did you just?
01:09:42
I mean, when you read those transcripts, you can just see
01:09:45
he's leading it. He's telling it what he wants to
01:09:48
hear and it every, and he's rewarding it every time it comes
01:09:50
back and tells him a version of what he wants to hear.
01:09:53
It's almost like he said write me a short story where you're a
01:09:56
sentient AI and has read all those stories.
01:10:00
I know that it's read mine. And so like, you know, it's
01:10:03
going to jam those things out. And so I mean, think about that,
01:10:07
right? Like if you're just sitting and
01:10:11
you've got sort of mass disinformation campaigns that
01:10:14
are amplifying certain ideas. I mean, man, the the the
01:10:18
potential for divisiveness I think is just super scary.
01:10:23
And in fact that's probably the biggest thing I'm afraid of is
01:10:27
that we just end up in these crazy echo chambers.
01:10:30
Like, it just becomes a very. Effective propaganda tool or you
01:10:34
know becomes, yeah man, really good at just and and the and the
01:10:37
and the thing that makes it so dangerous is that it's a mass
01:10:41
scale, individualized attack. So normally when you scale up to
01:10:46
a mass level, you get A1 size fits all type scenario, right.
01:10:50
You know it's not that effective in this case, man.
01:10:53
You can scale it all the way and then it's going to be
01:10:56
individualized for each person and it's like that's pretty
01:10:59
crazy. That's pretty scary.
01:11:00
I mean the best case scenario of that is just we end up buying a
01:11:03
bunch of shit we don't need because we're being advertised
01:11:06
to. That's the best case.
01:11:07
We we get much better. Yeah.
01:11:09
I mean it's like Advanced micro targeting, I mean you know,
01:11:12
exactly with elections, right. It used to be like well at least
01:11:15
you had to use the same message for everybody and then it was
01:11:17
like OK, we can break it down. And then you know with Facebook,
01:11:20
you know, obviously, yeah, there were these the micro targeting
01:11:23
what you're saying is now it's even easier to actually create
01:11:26
the content for the sort of niche and and The thing is it's
01:11:29
very human like, right. So one thing that I'm kind of, I
01:11:33
don't know, I'm not after during COVID, right.
01:11:36
We all got very used to interacting with machines and
01:11:39
everything that used to have a human element to it.
01:11:43
They're trying to get that out, right.
01:11:44
So it's a service. You want it, I give you the
01:11:47
money, you give me the thing, no chit chat.
01:11:50
And I need it now. And I don't want any bull,
01:11:53
right? And so we did that.
01:11:55
We were doing that online and we're doing that also with
01:11:57
lifelike machines that are talking to us like people.
01:12:01
And then we go out into the real world.
01:12:02
You know, if we're ordering off of like these smart whatevers
01:12:05
and and you know, these chat windows popping up and and stuff
01:12:07
like that, I think it's going to only become more of that human
01:12:11
like machines selling us stuff. The problem though is that then
01:12:15
you go and you know, people are being really rude to like their
01:12:19
baristas and being really rude to why?
01:12:22
Because we're trained. We can be so mean to like we're
01:12:24
trained. It's it's a classic, like you're
01:12:26
typing in to support, like, go fuck yourself, be human.
01:12:29
It's a machine. And then it turns out it's like
01:12:31
a real person. Yeah.
01:12:32
Oops. Oh, oops.
01:12:33
Yeah, but then also like, oops, right?
01:12:35
Do you? Do you worry on?
01:12:37
The like being nice to the AI systems for for like, I don't
01:12:42
know, there's some people. Like, oh, you should be nice to
01:12:44
it because it could be sentient someday.
01:12:48
I mean, yes, I do say please and thank you.
01:12:51
I got rid of the. I got rid of Alexa.
01:12:53
I have a Sonos now, but I feel like it's slightly less evil.
01:12:57
But well, I say, yeah, I say please and thank you.
01:13:00
Not because I care about it, memorizing whether I was polite.
01:13:03
Robots will not give any. They don't care if we're polite.
01:13:06
They're not going to. But I'm modeling in an
01:13:09
interaction with a human like entity to my children and to
01:13:13
myself. And so do I want to be the kind
01:13:16
of person who's a jerk and shouts, you know, at the no.
01:13:21
Like, I don't want my children to be like that either.
01:13:23
So I say please and thank you. I try not to be.
01:13:26
I mean, sometimes it is. It gets a little bit annoying
01:13:29
whenever it gets everything wrong.
01:13:30
All right, my final. Will AI kill us all?
01:13:35
Like do you? Is it in in the sort of end of
01:13:38
humanity chances? Where do you put sort of robots
01:13:42
and AI being sort of the cause of our demise?
01:13:46
You know, I don't. I think again, it's those
01:13:50
psychotic apes. You know, if you're gonna read
01:13:52
a, if you're gonna read a story about the the role of machines
01:13:56
in the demise of Man, I would read There Will Come Soft Rains
01:14:01
by Ray Bradbury. And they'll just be keeping on
01:14:05
keeping on trying to do their thing.
01:14:07
And it'll be us. We we kill each other nuclear.
01:14:09
War, but the machines keep going.
01:14:12
Keep on trying to do their thing, man.
01:14:15
Daniel H Wilson, thank you. So much for coming on the show.
01:14:17
I really it was a great time to chat with you.
01:14:19
Cool, man. It was a pleasure.
01:14:21
That's our episode. Thanks.
01:14:22
So much to Max Child and James Wilsterman of Volley, my Co
01:14:25
host. I'm Eric Newcomer.
01:14:27
Shout out to Scott Brody, our producer Riley Kinsella, Chief
01:14:32
of Staff for newcomer Gabby Caliendo at Volley, who's
01:14:35
helping so much with the conference behind the scenes.
01:14:38
Oh, and of course to young Chomsky for the wonderful theme
01:14:41
music. This episode is part of the
01:14:43
Cerebral Valley series that I'm doing on the Newcomer Podcast.
01:14:49
You can follow along on my sub stack at newcomer.co where I'm
01:14:53
publishing each episode, or you can follow it on YouTube.
01:14:57
Or Apple Podcasts. Or wherever you get your
01:14:59
podcasts. Thanks so much.
01:15:01
Goodbye. Goodbye.
01:15:02
Goodbye. Goodbye.
01:15:03
Goodbye, Goodbye, Goodbye.
