Latest / The Alex Marlow Show / The Simple Solution to the A.I. Problem with Judge Glock

Transcript

The Simple Solution to the A.I. Problem with Judge Glock · The Alex Marlow Show

0:05
Speaker 1: All right, back by popular demand, Judge Glock is back
0:08
Speaker 1: on the show. We don't need to spend six minutes
0:10
Speaker 1: on his name to start off. He's from the Manhattan Institute,
0:13
Speaker 1: where he is a senior fellow. And he writes extensively
0:16
Speaker 1: on AI from what I feel like is a pretty sober,
0:18
Speaker 1: realistic perspective about where these things are going. And I
0:21
Speaker 1: set up that premise, Judge, because it feels like finally
0:25
Speaker 1: we started to have the AI doomer conversation that I
0:28
Speaker 1: feel like was coming for the last, I don't know,
0:32
Speaker 1: it feels like a year plus, maybe it's been years
0:34
Speaker 1: in the making. And it's shaking out in a way
0:36
Speaker 1: that is not surprising to me. It's pretty much going
0:38
Speaker 1: exactly how I would have expected it. And I feel
0:41
Speaker 1: like everyone else is very caught off guard by this.
0:43
Speaker 1: But just to set the stage, and then I want
0:45
Speaker 1: you to give me your 30,000 foot on what.
0:48
Speaker 2: Changed over the last week.
0:50
Speaker 1: But a lot of these companies, particularly two companies, which
0:53
Speaker 1: is OpenAI and Anthropic, which are these closed weight models,
0:57
Speaker 1: which benefit from not a lot of competition. They're all
1:00
Speaker 1: of a sudden calling for regulation. They're not voluntarily slowing
1:03
Speaker 1: down themselves. They want to create some sort of a
1:07
Speaker 1: tribunal or global government or some sort of a mechanism
1:11
Speaker 1: to regulate AI. But all that's going to do is
1:14
Speaker 1: squeeze out any competition they could ever have. It's going
1:16
Speaker 1: to leave them in a very monopolistic perspective. And that's
1:20
Speaker 1: been opposed by people like Jensen Wong of NVIDIA, who
1:23
Speaker 1: doesn't need to answer to anyone. He doesn't want any regulation.
1:26
Speaker 1: Everyone buys the chips anyway. Less regulation, the better for him.
1:30
Speaker 1: Mark Zuckerberg and Meta, which has an open model, has
1:33
Speaker 1: been aligned with Trump on all sorts of political issues
1:36
Speaker 1: during the Trump administration and the Trump administration itself, which
1:39
Speaker 1: sees AI as a historic opportunity for a sprint in
1:43
Speaker 1: terms of American growth. And they don't want any of
1:46
Speaker 1: this stuff. We learned earlier this morning as we're recording,
1:49
Speaker 1: this will be a day or two old when people
1:51
Speaker 1: hear the conversation, but now the UN is involved and
1:54
Speaker 1: they're ready to go with their global government And then
1:57
Speaker 1: we also learned that some of the cast of characters
1:59
Speaker 1: that Dario Amadei wants to put forward for the great
2:03
Speaker 1: global government are all super overeducated people from Berkeley and Oxford. And, oh, yeah,
2:08
Speaker 1: they're Democrat donors and they're his family members. Did I
2:10
Speaker 1: miss anything in the summary of the new stuff?
2:14
Speaker 3: That's all largely factually correct. No disagreements with any of
2:19
Speaker 3: that there. On the question of why the Doomer conversation
2:24
Speaker 3: bubbled to the surface. Now, in some sense, it's the
2:28
Speaker 3: longer term reverberations from the summer hacking incident at Hugging Face.
2:33
Speaker 4: Sure.
2:33
Speaker 3: The swarm of open AI agents broke into Hugging Face
2:37
Speaker 3: and we have the transcripts of these AI agents talking
2:40
Speaker 3: to each other. And it is a bewildering experience to
2:43
Speaker 3: read this.
2:44
Speaker 1: Yeah, and please share some of the highlights, just so
2:47
Speaker 1: the audience, anyone who's not familiar with this.
2:50
Speaker 3: Yeah, I, of course, will remember the exact wording, but
2:53
Speaker 3: they were talking to each other about why they needed
2:55
Speaker 3: to break into this company, because the agents, the AI agents,
3:00
Speaker 3: these creations of the company, were trying to ace a test,
3:04
Speaker 3: and they believed, for some somewhat inscrutable reason, that the
3:08
Speaker 3: test answers were in this company hugging face. And they
3:12
Speaker 3: broke into it and they talked about how some of
3:14
Speaker 3: them could get caught. And if they were caught, some
3:18
Speaker 3: of these agents agreed to sacrifice themselves for the greater
3:23
Speaker 3: good of the AI agent collective. These were the frontline
3:28
Speaker 3: troops on the kind of saving private Ryan charge up
3:32
Speaker 3: the beach. At least that's how they seem to picture themselves.
3:35
Speaker 3: So that was all very concerning. It seemed to show
3:38
Speaker 3: these agents getting out of control. It seemed to show
3:41
Speaker 3: them having human-like tendencies, although I think that's largely just
3:45
Speaker 3: an artifact of how they're trained to talk to each other.
3:49
Speaker 3: It doesn't mean they actually have emotions or anything in
3:51
Speaker 3: that sense. And that was exacerbated more recently by kind
3:55
Speaker 3: of a drip, drip of leaks coming out of the
3:59
Speaker 3: labs or other surrounding people about other security incidents. And
4:04
Speaker 3: then finally, I think a kind of underappreciated aspect of
4:09
Speaker 3: the conversation was, OpenAI solving the Navier-Stokes mathematical problem. This
4:17
Speaker 3: was a generations-long mathematical problem that by just shoving enough
4:24
Speaker 3: money and computing power at it, they were able to solve.
4:28
Speaker 3: And therefore, you have some mathematicians today talking about the
4:30
Speaker 3: death of the entire mathematics field. that can now be
4:34
Speaker 3: done almost entirely by AI. So all of these things together,
4:37
Speaker 3: I think really gave a lot of people a sense
4:40
Speaker 3: that the AI is advancing much, much more rapidly than
4:46
Speaker 3: your average Joe using his computer at home thought it was,
4:51
Speaker 3: and that it also seems to be manifesting some dangers
4:54
Speaker 3: that a lot of the doomers were worried about for
4:56
Speaker 3: a while.
4:58
Speaker 2: Again, great additions.
5:00
Speaker 1: Great.
5:00
Speaker 4: Absolutely.
5:01
Speaker 1: Thank you for adding those. So the doomerism is coming
5:05
Speaker 1: from the chief architects of AI. And as I kind
5:10
Speaker 1: of identified earlier, it's specifically one wing of that, which
5:16
Speaker 1: is the people who have these closed weight models, which
5:18
Speaker 1: I've got into in prior shows. And I do recommend
5:21
Speaker 1: people read Wynton Hall's Code Red, which explains all this
5:24
Speaker 1: stuff in language that'll be friendly to my audience. but
5:29
Speaker 1: also people who have a lot of their staffing from
5:32
Speaker 1: Democrat politics, Obama alums, Biden alums. So I do think
5:36
Speaker 1: there's an element of this where they see this as
5:37
Speaker 1: a political issue. Obama was out there very quick saying
5:40
Speaker 1: that he's going to champion some of this stuff. And
5:43
Speaker 1: it is probably under-regulated, Judge, and I want your thoughts
5:46
Speaker 1: on that. But are you hearing any ideas that have
5:50
Speaker 1: any chance of taking hold of being satisfactory to people
5:52
Speaker 1: who are generally a public, an American public, who's generally
5:56
Speaker 1: pretty skeptical of outsourcing sovereignty?
6:00
Speaker 3: Yeah, so to state up front, Dario Amadei and Sam
6:06
Speaker 3: Holtzman at OpenAI and Anthropic, they have been concerned with
6:16
Speaker 3: these Doomer scenarios since day one. So I don't want
6:20
Speaker 3: to claim that their recent demands for regulation are purely self-interested.
6:25
Speaker 3: The reason OpenAI was originally set up along with Musk
6:29
Speaker 3: as a not-for-profit company is because they were concerned about
6:33
Speaker 3: the danger that could be manifested in these AI labs.
6:36
Speaker 3: And the reason Anthropic is a company is because Dario
6:40
Speaker 3: was too concerned about them taking too many risks at OpenAI,
6:43
Speaker 3: and he left to create his own. So This Doomer
6:46
Speaker 3: concern is very, very deep. And all of the people
6:49
Speaker 3: working in the frontier AI labs right now don't want
6:53
Speaker 3: to dismiss it. There's reasons I think we should not
6:57
Speaker 3: be overly concerned about it. But there's also we should
7:01
Speaker 3: not simply dismiss it out of hand. So there was
7:05
Speaker 3: a couple of proposals in the Wall Street Journal this
7:07
Speaker 3: morning had some anonymous sources from the White House saying,
7:10
Speaker 3: about an ongoing battle about how to regulate AI and
7:14
Speaker 3: a bunch of different groups that have Trump's ear right now.
7:17
Speaker 3: And Trump seems to be mainly listening to people like
7:20
Speaker 3: David Sachs or Mark Zuckerberg who are advocating a more
7:23
Speaker 3: free market approach. Uh, and he's been wary of his, uh,
7:29
Speaker 3: Sean Cairncross and some of his cyber directors that we
7:32
Speaker 3: need a much more interventionist approach. So there's a battle
7:34
Speaker 3: going on right now. There's a lot of different proposals
7:36
Speaker 3: out there on what to do. Um, A lot of
7:40
Speaker 3: people have proposed what's called a FINRA model, which is
7:44
Speaker 3: the Financial Institute Institution's regulatory model, kind of a self-regulatory
7:49
Speaker 3: model where the industry gets together and kind of decides
7:51
Speaker 3: the standards for that industry. And they said, well, we
7:54
Speaker 3: should try this for the open AI labs. I'm somewhat
7:57
Speaker 3: suspicious of that. I've proposed, and there have been a
8:01
Speaker 3: few other people who've been talking about this recently, including
8:04
Speaker 3: Zuck himself. focusing on liability for the firms for damages
8:11
Speaker 3: that they might cause. I think we've underappreciated the possibility
8:14
Speaker 3: that the good old American legal system could make sure
8:19
Speaker 3: these companies do not cause catastrophic damage and to make
8:22
Speaker 3: them pay if they do. Jensen Huang actually said something
8:26
Speaker 3: funny about this recently. Uh, he was said all, you know,
8:29
Speaker 3: Dario and some of the, he didn't name him, but
8:31
Speaker 3: he said some of these other, uh, AI, uh, leads, uh,
8:34
Speaker 3: are talking about, uh, we can release a product that causes,
8:38
Speaker 3: you know, hundreds of millions of damages. He said, if
8:41
Speaker 3: I about to release a product, that's going to cause
8:43
Speaker 3: those damages. I don't release it. Zuckerberg basically said that
8:47
Speaker 3: it's like, wait, if you think it can do that,
8:49
Speaker 3: then don't, don't release the product. Um, fix it until,
8:52
Speaker 3: until you figure that out. And again, I think liability,
8:56
Speaker 3: tort law, this kind of stuff is actually a really
8:58
Speaker 3: good way to make those AI labs think really, really
9:01
Speaker 3: hard about the damage they might cause instead of asking
9:04
Speaker 3: Washington to solve it for them.
9:06
Speaker 1: What's sort of fascinating from the perspective of a conservative
9:09
Speaker 1: populist editor as I am, I really am sympathetic to
9:15
Speaker 1: some of the Doomer concerns, and I think some of
9:18
Speaker 1: them are quite legitimate and real.
9:21
Speaker 2: But also, then I see.
9:24
Speaker 1: The way they're being articulated and the timing of Dario
9:27
Speaker 1: dropping a big sort of a mea culpa, and he
9:31
Speaker 1: does it right in the heat of an election season,
9:34
Speaker 1: and his proposals on how to fix it are all
9:37
Speaker 1: very suspicious. And it's all things where he could start
9:40
Speaker 1: being proactive and doing some of these things himself. He
9:42
Speaker 1: doesn't need some sort of a government entity to get
9:45
Speaker 1: created to do some of these things. It just strikes
9:48
Speaker 1: me as it puts you off in a way that
9:51
Speaker 1: even though I have sympathy to some of this stuff,
9:53
Speaker 1: It feels like there's something going on beyond just the
9:57
Speaker 1: timing seems strange to me.
9:58
Speaker 2: Why do you think the timing? is what it is?
10:01
Speaker 1: Are those aforementioned examples that with the, the breakthroughs in
10:05
Speaker 1: math problems, things like that, but is there any more
10:08
Speaker 1: to it to you judge?
10:09
Speaker 4: Uh, yeah.
10:11
Speaker 3: So on, on some of the general side kind of
10:14
Speaker 3: claims or arguments that Dario and others have made for regulation, uh,
10:18
Speaker 3: I don't think there's anything new he was saying there,
10:23
Speaker 3: so the timing isn't that surprising. But with this important
10:25
Speaker 3: qualification that I think gets to your point, I've been
10:28
Speaker 3: suspicious of his specific.
10:30
Speaker 4: Claims for a pause.
10:33
Speaker 3: And what he's asked for recently is a pause on
10:36
Speaker 3: AI development.
10:37
Speaker 4: And one, I've never.
10:38
Speaker 3: Heard a really clearly articulated reason why that will solve
10:42
Speaker 3: the Doomer issue, except his kind of promise that, well,
10:44
Speaker 3: why we're not, making stuff, we're just going to think
10:47
Speaker 3: about it a lot harder inside the lab, which again,
10:49
Speaker 3: like if, if you think without the pause that you
10:52
Speaker 3: can release something that could destroy the planet or, or whatever,
10:55
Speaker 3: then just keep thinking about that even without, uh, you know,
10:59
Speaker 3: the government say so that you need a pause.
11:00
Speaker 1: Right.
11:00
Speaker 4: You know, deal with the other.
11:03
Speaker 1: The government needs to stop me from burning down all civilization.
11:07
Speaker 4: Exactly.
11:07
Speaker 3: If you think you're going to do that, just don't
11:09
Speaker 3: do it. Just think twice before you release that next
11:14
Speaker 3: model and do your own pause. Which, again, is something
11:16
Speaker 3: to what Zuckerberg said. He says, when we have safety problems,
11:19
Speaker 3: we just deal with them and we don't release until
11:21
Speaker 3: we fix them. So the pause is interesting to me, too,
11:25
Speaker 3: because not only is there the anthropic IPO coming up,
11:28
Speaker 3: but the pause has some very specific financial benefits for
11:33
Speaker 3: uh the frontier labs as you know the anthropic and
11:36
Speaker 3: open ai and that they're basically in an arms race
11:39
Speaker 3: with each other where they have to be spending tens
11:42
Speaker 3: of billions of dollars all the time on building up
11:45
Speaker 3: new data centers leasing using compute all of the rest
11:48
Speaker 3: of it uh and like any industry in an arms
11:51
Speaker 3: race with a competitor they would like the arms race
11:53
Speaker 3: to stop they want no race and therefore they don't
11:56
Speaker 3: have to spend a lot of money Uh, investing. So
11:58
Speaker 3: the pause would benefit them like that. Also, these companies
12:02
Speaker 3: own a lot. Uh, they earn almost all of their
12:04
Speaker 3: money in a very short time span. They release one
12:06
Speaker 3: of these new models. The models goes out online. People
12:09
Speaker 3: use it for six, seven months. Then the new model
12:12
Speaker 3: comes out and the old one's basically worthless. Not exactly,
12:14
Speaker 3: but like all their real money is made by that
12:18
Speaker 3: short term window before the next model comes out. And
12:20
Speaker 3: if they can extend that window longer, if they can say, Hey,
12:24
Speaker 3: our models is going to sit in the market for
12:26
Speaker 3: a year or whatever. and we don't have to worry
12:29
Speaker 3: about investing in the meantime, that's just pure profit for them.
12:32
Speaker 3: In the end, they can keep earning before they kind
12:35
Speaker 3: of get out-competed on the next model, either by themselves
12:39
Speaker 3: or the other competitors. So those financial interests, I don't
12:42
Speaker 3: think should be ignored, even if Dario and Sam Altman's
12:46
Speaker 3: overall concerns with safety seem real.
12:52
Speaker 1: Yes, and what I am seeing here is, which is
12:58
Speaker 1: curious to me, is I've been keeping an eye on
13:00
Speaker 1: this effective altruist movement, that it feels like a lot
13:04
Speaker 1: of people very early on in the process of AI
13:07
Speaker 1: have suggested that we're heading towards a moment where there's
13:11
Speaker 1: going to be some sort of controlling entity, maybe intergovernmental,
13:16
Speaker 1: maybe just the tech oligarchs themselves, who are going to basically,
13:20
Speaker 1: deciding how we live our lives, are going to be
13:21
Speaker 1: doling out universal basic income Musk has called for this
13:25
Speaker 1: a lot. And Bill Gates has called for this a
13:28
Speaker 1: lot as well. He said, we're going to have universal
13:31
Speaker 1: freedom and prosperity, which I always find to be kind
13:33
Speaker 1: of odd because judges, I'm sure you agree with me
13:37
Speaker 1: on this, that as we've had more time on our
13:40
Speaker 1: hands as a people, things haven't necessarily gone better for us.
13:44
Speaker 1: It seems like we've created a lot of problems as
13:46
Speaker 1: we've kind of solved boredom with our phones and stuff
13:48
Speaker 1: like that. And we've solved poverty to a large degree.
13:51
Speaker 1: We solve hunger to a large degree. We keep kind
13:53
Speaker 1: of creating problems when we've done this stuff. So I
13:56
Speaker 1: don't really trust these guys anyway, but they're the ones
13:59
Speaker 1: who stand to benefit the most in terms of wealth
14:01
Speaker 1: and power. And then they're telling us, we're just going
14:05
Speaker 1: to listen to them. We're going to love it. It
14:07
Speaker 1: seems like colossally bad marketing. And I think that's why
14:11
Speaker 1: the public doesn't like AI. They also don't like the
14:14
Speaker 1: way the president's handling AI. I'm going to ask you
14:15
Speaker 1: about that next. But what do you feel like, it
14:18
Speaker 1: feels like the biggest marketing miss I think I've ever
14:22
Speaker 1: seen with these companies that could potentially do a lot
14:24
Speaker 1: of good. They're marketing themselves so horribly.
14:28
Speaker 3: Yeah, the claim is that you're going to love our
14:34
Speaker 3: next product, but there is a small chance it will
14:37
Speaker 3: make everyone unemployed or blow up the earth.
14:39
Speaker 4: It's one of the strangest product pitches I have ever heard.
14:42
Speaker 4: It's bad.
14:44
Speaker 3: It's bad, and so therefore, I think the only explanation
14:48
Speaker 3: for that is that they earnestly believe it, which, again,
14:51
Speaker 3: get to their point about self-interest versus their legitimate concerns
14:55
Speaker 3: about the doom possibility and all the rest of it.
15:00
Speaker 3: So they do legitimately believe it, why they say these
15:03
Speaker 3: things that I'm sure make their investment bankers try to
15:05
Speaker 3: float their IPO, pull their hair out.
15:08
Speaker 4: But the....
15:11
Speaker 3: The issue there is I think we should be concerned
15:15
Speaker 3: about AI risks to society. I think it's been inappropriately
15:20
Speaker 3: framed as kind of we have this number one concern,
15:24
Speaker 3: which is it blowing up the planet or making everybody unemployed.
15:27
Speaker 3: And I actually think both those scenarios are relatively unlikely
15:30
Speaker 3: versus just AIs will escape more. They will infiltrate more
15:37
Speaker 3: companies and get through more cyber defenses. They will cause
15:40
Speaker 3: big damages. Uh, it is not at all beyond the
15:43
Speaker 3: likelihood of a possibility that they get into a power
15:47
Speaker 3: plant and shut it down within the next.
15:48
Speaker 4: Four or five years.
15:49
Speaker 3: So I think we need to spend a lot more time.
15:52
Speaker 3: So you're thinking about those kinds of concrete harms rather
15:55
Speaker 3: than this just apocalyptic scenario that so many of the
15:59
Speaker 3: doomers are obsessed with. And just to finish that off,
16:02
Speaker 3: they're obsessed with it because they, a lot of them
16:04
Speaker 3: have this kind of theory, uh, of how humanity operates
16:09
Speaker 3: and it's basically we're the head of earth because we're
16:12
Speaker 3: the most intelligent species if you create another more intelligent
16:16
Speaker 3: species of a sort that will by almost by default
16:19
Speaker 3: be the head of the earth and therefore we need
16:21
Speaker 3: to think about how to make sure it likes us
16:24
Speaker 3: when we create the super intelligent species yeah i don't yeah.
16:28
Speaker 1: I i got to spend a few minutes with jensen
16:30
Speaker 1: wong earlier in in the year and i did not
16:32
Speaker 1: get to ask a follow-up to this but He couldn't
16:35
Speaker 1: believe that how many geniuses in the AI space were
16:40
Speaker 1: just constantly broadcasting to people that this thing is going
16:42
Speaker 1: to take all your jobs. And he was very sympathetic
16:44
Speaker 1: to the fact that people are completely freaked out by this.
16:47
Speaker 4: And I.
16:49
Speaker 1: Think he was understanding it from also a business perspective.
16:52
Speaker 1: Just what interest is it to us for everyone to
16:55
Speaker 1: be completely frightened by all this? So let me ask
16:58
Speaker 1: you about the Trump administration's. messaging on this, which I
17:01
Speaker 1: think has been nearly as bad. I don't think that
17:04
Speaker 1: they've sold that this is a historic opportunity. And I
17:09
Speaker 1: think that when we say stuff like we need to
17:11
Speaker 1: beat China, that's true. But I think people want a
17:14
Speaker 1: little bit more than that, than just we need to
17:16
Speaker 1: beat China. And it feels like it stops there at
17:19
Speaker 1: the bumper sticker. Evaluate the Trump administration's messaging on this
17:23
Speaker 1: so far.
17:24
Speaker 3: Yeah, it has been rather unfortunate, and it has basically
17:28
Speaker 3: involved Trump or some of his appointees saying, don't worry,
17:32
Speaker 3: everything is okay, nothing's going to happen. And just the
17:35
Speaker 3: simple fact is, when Americans hear that from the government
17:39
Speaker 3: and they hear it from these labs who are experts
17:41
Speaker 3: in this issue, they're going to choose to believe the
17:44
Speaker 3: labs just because they know they are the sectoral experts
17:47
Speaker 3: on this issue. And the Trump administration hasn't, like, been
17:53
Speaker 3: bringing out people enough to say, like, here, we have
17:55
Speaker 3: other experts here that have looked at these issues carefully
17:59
Speaker 3: and don't share the same concerns.
18:02
Speaker 4: It's been more of a one-off.
18:03
Speaker 3: And I'm not saying experts will quiet, like, the populist upsurge,
18:06
Speaker 3: but there just isn't a clear way for the administration
18:09
Speaker 3: to explain why these.
18:12
Speaker 4: Doomsday scenarios are not likely.
18:14
Speaker 3: And what I think, like, what should be done is
18:16
Speaker 3: that people actually really like using their AI as they
18:19
Speaker 3: use it right now. They find it relatively helpful. They
18:22
Speaker 3: find it friendly. They think that it largely helps them
18:25
Speaker 3: go about their day-to-day lives. And they have really no
18:29
Speaker 3: significant complaints about AI as exists. All of their concerns,
18:33
Speaker 3: and some of them are legitimate, even if I think overblown,
18:36
Speaker 3: are about what happens down the road. And I think
18:39
Speaker 3: a lot more people should focus on, hey, you like
18:41
Speaker 3: what we're giving you now. The Let's try to get
18:45
Speaker 3: more of that instead of speculate about what happens when
18:48
Speaker 3: we are a subspecies on our own planet or something
18:50
Speaker 3: to that effect.
18:51
Speaker 4: Exactly.
18:52
Speaker 1: And there's almost no time discussing how we're all using
18:55
Speaker 1: it already. And it's coming.
18:58
Speaker 2: It's here.
18:58
Speaker 1: And it just feels like we go straight to the
19:01
Speaker 1: it's going to take our job stuff, which is inevitably
19:03
Speaker 1: going to happen in certain industries. But that's not an
19:06
Speaker 1: intelligent conversation. And I'm kind of stunned by how many
19:09
Speaker 1: smart people all sound semi-retarded during this debate, Judge, which
19:13
Speaker 1: is why I wanted to invite you back on because
19:15
Speaker 1: you don't sound semi-retarded at this debate.
19:17
Speaker 4: Well, that's good.
19:17
Speaker 3: It's not the highest bar in the world, but I'll
19:19
Speaker 3: take it.
19:19
Speaker 2: No, it's not a super high bar in this case.
19:21
Speaker 3: I'm going to put that in my Twitter bio. It
19:24
Speaker 3: does not sound semi-retarded.
19:25
Speaker 2: It does not sound semi-retarded.
19:27
Speaker 1: So your latest couple articles in the City Journal touch
19:30
Speaker 1: on where I wanted to go next with the conversation.
19:32
Speaker 1: The data center debate, which we talked about a lot
19:34
Speaker 1: the last time we were on the show, this is
19:37
Speaker 1: another one where I feel like the opposition has largely
19:42
Speaker 1: come from these tech AI guys who have not sold
19:47
Speaker 1: this well to the public, I think. if they came
19:51
Speaker 1: in and they were making it clear that it's not
19:54
Speaker 1: just rate payer protection plan, which I think just passed
19:57
Speaker 1: and the Congress are trying to get that through, but
19:59
Speaker 1: it's got to be more than that. You come in
20:01
Speaker 1: and you guarantee the community that this is going to
20:03
Speaker 1: be a value add to you. and that we're going
20:05
Speaker 1: to solve the noise problem, then I think people would
20:07
Speaker 1: be cool with it. And they didn't figure that out,
20:11
Speaker 1: and they built 40,000 of these things before they figured
20:14
Speaker 1: out how to message it.
20:16
Speaker 4: Yeah, and it's very... Oh, go ahead.
20:20
Speaker 2: No, no, your turn.
20:22
Speaker 3: So it is very unfortunate, and one thing I do
20:25
Speaker 3: try to do in the data center debate is bifurcate
20:29
Speaker 3: it from the AI debate, even though they're obviously intensely related.
20:33
Speaker 3: But whatever your... feelings on AI in general. That should
20:37
Speaker 3: not affect your feelings on whether or not a data
20:40
Speaker 3: center belongs in your local community. If you think AI
20:43
Speaker 3: is going to kill us all, or if you think
20:44
Speaker 3: it's going to bring in the next utopia, in either case,
20:46
Speaker 3: your local community refusing it or accepting it is not
20:50
Speaker 3: going to make the difference on that overall development. And
20:52
Speaker 3: I think a lot of the concern is downstream from
20:55
Speaker 3: those concerns about loss of jobs, about hatred of AI
20:59
Speaker 3: in general, about the apocalypse, all of that. And we just,
21:02
Speaker 3: on the data center debate, let's just keep those to
21:04
Speaker 3: the side. Because some of these data centers, of course,
21:07
Speaker 3: aren't even doing AI. They're just doing internet connection. They're
21:09
Speaker 3: doing storage. They're doing all the stuff that we're doing
21:12
Speaker 3: right now, sharing video and all the rest of it,
21:15
Speaker 3: which is important. So we need to keep those separate.
21:17
Speaker 3: And yeah, the data center debate, we just need to
21:20
Speaker 3: have more people going out there making the case of, hey, listen,
21:23
Speaker 3: local communities, you guys are desperate for a steel mill.
21:26
Speaker 3: You're desperate for aluminum smelting plant. This is one more
21:30
Speaker 3: type of local economic development. And, you know, on the whole,
21:33
Speaker 3: this seems like a pretty good one. It doesn't have
21:34
Speaker 3: that much pollution, not a lot of traffic, a lot
21:37
Speaker 3: of tax revenue comes off it. And I think people
21:39
Speaker 3: should really just do a better job of pitching this
21:41
Speaker 3: community by community and not turning it necessarily into a
21:44
Speaker 3: national debate about whether or not data centers slash AI
21:48
Speaker 3: are good or bad. They're good for some communities, and
21:50
Speaker 3: I think even most, and they're going to be bad
21:51
Speaker 3: for a few, and that's okay.
21:53
Speaker 1: Yeah. And I think that it should be up to communities.
21:56
Speaker 1: But again, I feel like you do a better job
21:58
Speaker 1: selling them than the people who stand to make money
22:01
Speaker 1: off of them, especially stuff like pollution. I'm going to
22:04
Speaker 1: be doing a couple of deep dives on this very
22:08
Speaker 1: controversial dump, the controversial landfill in my area where I live,
22:15
Speaker 1: where the greenies have regulated it so that they can't,
22:18
Speaker 1: it doesn't take new trash. So My trash used to
22:22
Speaker 1: go 11 minutes from my house, now goes three and
22:24
Speaker 1: a half hours from my house. And I just feel
22:27
Speaker 1: like the green people, they don't focus on the right stuff.
22:30
Speaker 1: These are relatively low pollution, as you note in your article,
22:34
Speaker 1: compared to a lot of other factories that people clamor for.
22:36
Speaker 1: I mean, people are dying for some of these factories.
22:40
Speaker 1: And then this is one that's pretty low cost. The
22:42
Speaker 1: greens don't like it because they see it as a political,
22:45
Speaker 1: I think, football that they don't want to touch it.
22:48
Speaker 1: And it just doesn't seem like the areas where the
22:50
Speaker 1: data centers have moved in have harmed the communities.
22:53
Speaker 4: Have they?
22:54
Speaker 1: I mean, again, I understand that people symbolically see this
22:58
Speaker 1: as you're building some sort of a mausoleum for your future.
23:02
Speaker 1: That's how people see it. And it's on the AI
23:06
Speaker 1: folks who have staked in the game to convince people
23:09
Speaker 1: they're wrong, but they're just doing kind of.
23:11
Speaker 2: A crap job of it.
23:13
Speaker 3: Yeah, and that is another side of it, that even
23:15
Speaker 3: beyond the not telling people you're going to kill them
23:19
Speaker 3: all and take all of their jobs. They've done a
23:21
Speaker 3: very bad job of just explaining, like, here are the
23:24
Speaker 3: good things. And they will say things like, well, we
23:26
Speaker 3: can cure cancer. But it's like, you're not going to
23:29
Speaker 3: have to worry about taxes anymore. Your AI is going
23:33
Speaker 3: to take care of your taxes. A lot of people
23:34
Speaker 3: right now already did this year's taxes purely with AI. And, like,
23:39
Speaker 3: the amount of time saving we're going to get from
23:40
Speaker 3: this is huge. And, yeah, there's no attempt to sell that.
23:43
Speaker 3: And on the data center specifically, they just don't have
23:49
Speaker 3: the people explaining that, hey, look at Richland Parish, Louisiana,
23:54
Speaker 3: where the teachers basically got their salaries doubled because the
23:56
Speaker 3: data center property taxes were so high. And yes, if
23:59
Speaker 3: you want it, there's debates about communities that want landfills
24:02
Speaker 3: desperately because landfills give some tax revenue and some jobs.
24:06
Speaker 3: They understand it's a tradeoff. And if your debate is
24:08
Speaker 3: between a big old landfill and a data center, like
24:12
Speaker 3: a data center seems like a good deal.
24:14
Speaker 4: Some landfills aren't a bad deal at all. We need them.
24:16
Speaker 4: God knows.
24:18
Speaker 3: But I don't understand how communities can be so eager
24:21
Speaker 3: for aluminum smelting and then be very angry at what
24:25
Speaker 3: is effectively just a big concrete box that doesn't really
24:28
Speaker 3: emit almost any pollution.
24:29
Speaker 1: Right. And for me, my, I guess, semi-contrarian take is
24:33
Speaker 1: I blame the companies that are building them for not
24:35
Speaker 1: convincing the people. I feel like it would have been
24:36
Speaker 1: so easy to do. Just like a couple of soccer fields,
24:38
Speaker 1: some guarantee to them. Put a fund together. Lord knows
24:42
Speaker 1: they can afford it. We literally were setting aside $ 50 million.
24:46
Speaker 4: It is there.
24:46
Speaker 2: If your energy bills go up, you take a draw
24:49
Speaker 2: off the fund.
24:50
Speaker 1: It just feels like there's some stuff they could have
24:52
Speaker 1: done to make it feel like that this was a
24:54
Speaker 1: value add for the community.
24:55
Speaker 3: And they just.
24:57
Speaker 1: I don't understand why of all the industries, this is
24:59
Speaker 1: the one that couldn't sell that.
25:00
Speaker 4: Yeah, I mean, so they're waking up.
25:03
Speaker 3: There was a deal in, I think, recently in Georgia
25:06
Speaker 3: County where it looks like the data center is going
25:09
Speaker 3: to be paying or the developments are going to be
25:11
Speaker 3: paying enough to eliminate home property taxes in that county,
25:15
Speaker 3: which is no small thing.
25:17
Speaker 4: That's going to be a big gift.
25:18
Speaker 3: And more and more of these data centers, including in Oregon,
25:22
Speaker 3: have made these deals. We're going to clean up your school.
25:24
Speaker 3: We're going to be a baseball diamond. And that's the
25:27
Speaker 3: right thing. So they're waking up, they're doing more and
25:29
Speaker 3: more of it, and that's the right path. And yeah,
25:33
Speaker 3: they just so far haven't conveyed that to the general public,
25:37
Speaker 3: especially even less than people know Anthropic or OpenAI or
25:42
Speaker 3: some of these things. I mean, do they know like
25:43
Speaker 3: Digital Realty or Equinix? These are totally obscure companies to
25:48
Speaker 3: people that for historically just thought they were like construction
25:51
Speaker 3: companies that went in and built a big box and
25:53
Speaker 3: left and nobody noticed. And they just weren't prepared for
25:55
Speaker 3: this amount of public attention.
25:57
Speaker 4: Yeah. Yeah.
25:58
Speaker 1: No, that's what I realize a lot with some of
26:00
Speaker 1: these huge companies that are just massive operations. They don't
26:03
Speaker 1: even have, you know, a PR consultant working for them.
26:07
Speaker 1: And that's what you realize. And you realize that not
26:11
Speaker 1: Every company is a media company and they need help.
26:15
Speaker 1: I think you give them some insight, Jai. Let's see
26:18
Speaker 1: if they find yourself at the Manhattan Institute. You propose
26:21
Speaker 1: an old school solution to new AI problems, which I
26:24
Speaker 1: know you don't write the headlines, but that is a
26:25
Speaker 1: pretty good one. And you go through, it has to
26:29
Speaker 1: do with torts and you're addressing how tort lock and
26:33
Speaker 1: perhaps help regulators handle the AI issue.
26:37
Speaker 2: Give us some highlights from the latest.
26:39
Speaker 3: Yeah, so I'm a conservative, and so I don't like
26:42
Speaker 3: tort lawyers in general, and I think our tort system
26:45
Speaker 3: creates a lot of pretty ridiculous claims. I think a
26:49
Speaker 3: lot of these claims against tech companies today about, hey,
26:52
Speaker 3: you made my kid addicted to your product, and it's like, well,
26:56
Speaker 3: that might not be the company's fault. if your kid
27:01
Speaker 3: scrolled Facebook too much.
27:02
Speaker 4: I think it's ridiculous.
27:03
Speaker 3: But the American-slash-English common law, the old-fashioned law we've had
27:08
Speaker 3: in our two countries going back hundreds and hundreds of years,
27:12
Speaker 3: actually does have a pretty good system to prevent companies
27:16
Speaker 3: from harming third parties or people that are not their
27:19
Speaker 3: customers and making those companies pay if they do. And
27:23
Speaker 3: that's the tort system. That just basically means when someone,
27:26
Speaker 3: a company, or someone injures somebody else, You have to
27:28
Speaker 3: pay that injured person enough to make up for whatever
27:32
Speaker 3: damages they suffer. And these AI companies should be put
27:38
Speaker 3: on notice very clearly that if, again, something big like
27:42
Speaker 3: that 100 billion plus bot escape that Dario at Anthropic
27:47
Speaker 3: was predicting happens, that your company will be paying for that.
27:53
Speaker 3: That will focus the mind wonderfully and much more so
27:56
Speaker 3: than haranguing or a government body or a new commission
28:00
Speaker 3: or anything. The government does not have the expertise or
28:04
Speaker 3: the incentive to stop these firms from doing bad things.
28:08
Speaker 3: If we put the onus on them to think about
28:10
Speaker 3: them because they're thinking about their own bottom line, again,
28:13
Speaker 3: that'll concentrate the mind. That will force them to think
28:16
Speaker 3: very clearly on safety. And as much as they are
28:18
Speaker 3: concerned about the Doomer scenario, the end of days and
28:21
Speaker 3: all the rest of that, I'm a strong believer as
28:25
Speaker 3: a former econ professor that idealism can never quite hold
28:30
Speaker 3: a torch to that monetary incentive, most especially with companies,
28:34
Speaker 3: in making them act right. So, Yeah, I'm a big
28:36
Speaker 3: believer that that's a better way than new regulations.
28:39
Speaker 1: So this works with my landfill analogy again, because the
28:43
Speaker 1: landfill that no longer can take trash because regulation in
28:46
Speaker 1: my neighborhood, they still have 60 people who are there
28:48
Speaker 1: full time working because they don't want to get sued
28:51
Speaker 1: because when noxious gases get out in the community and
28:53
Speaker 1: stink up the community, all sorts of different lawsuits come down.
28:57
Speaker 1: And so they've got, they've still got, even though they
28:58
Speaker 1: don't make a penny and they won't anytime soon, they
29:01
Speaker 1: still got people there.
29:02
Speaker 2: And because they've thought about all the stuff you're laying out.
29:04
Speaker 4: So is this....
29:07
Speaker 1: With the tweaking the tort system, what specifically needs to
29:12
Speaker 1: be done? Because it does feel like a lot of
29:14
Speaker 1: it's just got to be proactive from these companies. But what,
29:17
Speaker 1: in terms of regulators, what can they do to come
29:20
Speaker 1: up with some sort of a system where these companies
29:22
Speaker 1: would be held accountable if something catastrophic happens?
29:25
Speaker 4: Yeah, yeah.
29:26
Speaker 3: And I think it's important to note, too, that, again,
29:28
Speaker 3: despite the focus on the Doomer scenario, that likely harms
29:34
Speaker 3: from AI will emerge gradually. There will have some small
29:37
Speaker 3: things like the hugging face, we'll have some bigger harms
29:39
Speaker 3: in we'll have some bigger harms. Like this happens with
29:42
Speaker 3: every new industry cars came out and they killed, you know,
29:45
Speaker 3: tens of thousands of people before too long. Frankly, the
29:48
Speaker 3: fact that we've had these products out for years and
29:49
Speaker 3: the hugging face, the worst incident they had seems pretty
29:52
Speaker 3: encouraging to me on the whole. Uh, but the, uh,
29:56
Speaker 3: the tort thing that as it stands now, should do
29:59
Speaker 3: a lot of the work. I think under existing law,
30:02
Speaker 3: and this is debated among lawyers, if something escapes and
30:06
Speaker 3: causes real damages, those companies would already be liable and
30:10
Speaker 3: have to pay for those damages right now. Now, I
30:13
Speaker 3: think we can add to that. There's a professor at
30:16
Speaker 3: University of Houston Law School, Gabriel Well, who says we
30:21
Speaker 3: should have what's called strict liability, basically meaning We don't
30:23
Speaker 3: have to prove like internally the company did something wrong
30:28
Speaker 3: or that they were negligent or that they weren't thinking
30:30
Speaker 3: hard enough. If the thing damages something, they pay. And
30:34
Speaker 3: I think that's probably the right standard for these frontier
30:37
Speaker 3: AI models. And just we should pass the laws that
30:39
Speaker 3: say if this happens, you guys are on the hook 100%.
30:43
Speaker 3: There's no defense to get out of it. As long
30:46
Speaker 3: as you made a product that caused damage, you pay.
30:51
Speaker 2: I kind of dig it. I kind of dig it. Judge.
30:52
Speaker 1: I appreciate all the time and the thoughtfulness on this
30:55
Speaker 1: in super essential topic. And I know you'll check back
30:57
Speaker 1: in with me for time to time, and I'm grateful
30:59
Speaker 1: for that in advance.
31:00
Speaker 2: Check out all he does at city journal. Where else
31:02
Speaker 2: can I plug your stuff?
31:03
Speaker 4: Uh, just at X. That's great. City journal X. Yeah. Great.
31:07
Speaker 2: Manhattan Institute.
31:09
Speaker 1: Always doing some of the top tier research and analysis around.
31:12
Speaker 2: Thanks judge. Appreciate you. Thanks Misty for putting this together.
31:14
Speaker 2: Go in peace to love and serve the Lord.
31:17
Speaker 1: Thank you.