Loading summary
Transcript79 lines
- [00:00]
Matt Welch
Matt Welch is a Luddite. I'm not sure you've ever used ChatGPT.
- [00:05]
Bob Cesca
Oh, just, you know, you do the Google search now. They do a thing at the top
- [00:08]
Matt Welch
that says AI, not ChatGPT.
- [00:10]
Bob Cesca
Yes, yes.
- [00:11]
Matt Welch
Yeah, so. And do you know what Claude is?
- [00:13]
Bob Cesca
That's one of the things.
- [00:14]
Matt Welch
Yep. So that's the starting point. We know of new methods of attack. I mean, you are a professional, Mr. Right.
- [00:26]
Camille Paglia
The Trojan Hog.
- [00:26]
Matt Welch
Media Pioneer, the Fifth Column. This is true. I'm aware of this. And I don't remember the first time I became aware of it, but it's
- [00:35]
Bob Cesca
always been like that. That's why.
- [00:36]
Matt Welch
Well, this is possible. I remember meeting you for the first time, not seeing you for the first time. Cause I'd experienced your content online. Probably something featuring Matt Welch at a conference someplace, some fancy schmancy conference. I don't know what it was. I won't say.
- [00:51]
Camille Paglia
It was before we met at Satori. We met at the Satori thing a few years ago.
- [00:55]
Matt Welch
I feel like we met before that. Some. We definitely met a Satori. And Satori is actually one of my favorites.
- [01:00]
Bob Cesca
It's a Tori thing. Conferences, sex, right?
- [01:02]
Matt Welch
Yes.
- [01:03]
Camille Paglia
Group sex.
- [01:04]
Matt Welch
Yes. Great group sex. By the way. He is amazing.
- [01:11]
Bob Cesca
Well, we started rolling 90 seconds ago, so we're just hitting here. No, we were talking beforehand, Camille and I, and I'm sure there is a large subset of listeners out there in the world who associate Bob. Right? Robert. Right. Depending on where they find you.
- [01:30]
Matt Welch
As I call him Bobby.
- [01:32]
Bob Cesca
Bloggingheads guy. Yeah, he's a blogging heads guy. He started this, what, 20 years ago? Holy.
- [01:37]
Camille Paglia
21.
- [01:38]
Bob Cesca
21 years ago. Just far enough ahead of video blogging to not get rich podcasting, which is fantastic.
- [01:46]
Camille Paglia
I definitely managed to not get rich, but it took work. It wasn't easy.
- [01:50]
Bob Cesca
Mickey Kaus helped.
- [01:52]
Camille Paglia
I'm sure Mickey's guidance helped me avert riches. Yes.
- [01:55]
Bob Cesca
The other thing that people know you for is just having to be exasperated constantly by Mickey Kaus, my long lost friend in Los Angeles, who has some, let's say, paranoid feelings about immigrants over the years. So people know you as this guy or someone who's written for New Republic and Slate and kind of, you know, the Kinsley esque wing of journalistic opinion liberalism. But you've actually not. You've been writing versions of this book now, now, now, for a while.
- [02:30]
Camille Paglia
Are you accusing me of recycling?
- [02:32]
Bob Cesca
That's not an accusation.
- [02:33]
Camille Paglia
I'm talking more about AI.
- [02:34]
Bob Cesca
No, not an accusation. That is a compliment.
- [02:36]
Camille Paglia
Oh, thank you, freelance journalist.
- [02:39]
Bob Cesca
Every part of the cow, at least four different sales. But no, you've been writing about intersection of like technology and philosophy and God on some level for a really long time. And you've been writing about AI back in the 80s.
- [02:52]
Matt Welch
So tell us about that.
- [02:52]
Camille Paglia
Yeah, not. But it's true that in 1983. Not that I'm old enough for this to be true, but in 1983, as a young journalist, I interviewed Geoffrey Hinton, who is now commonly called the godfather of AI, for a journal called the Wilson Quarterly, which is sadly, normally I
- [03:10]
Bob Cesca
wrote the Wilson Quarterly at least once. Yep, yep, I'm that old too.
- [03:16]
Camille Paglia
No, you're not. The. So, yeah, and his. At that time, he had this maverick approach to AI, and I talked about it, but it was not at all mainstream. And then his approach prevailed and is responsible for ultimately large language models, the so called deep learning revolution. And I hadn't really kept track of him. Every once in a while his name would show up in something about AI. And then suddenly, like in 2023 when ChatGPT became a big thing, I saw the headline in the New York Times, godfather of AI quits Go Google to warn of AI risks or something. That's a paraphrasal. So yeah, I went back and read my piece and understood what I had failed to understand. When I wrote the piece about the potential power of this approach, I'd really, really totally missed the boat.
- [04:07]
Bob Cesca
So if you're saying that was one of the approaches, what was the road not taken?
- [04:12]
Camille Paglia
Well, the road not taken was more like kind of formal logic in a sense. You know, you represent things in symbols and then manipulate the symbols and it's a completely deterministic system. You would understand how it works. And these models, they don't entirely understand how they work.
- [04:38]
Matt Welch
They being the people who built them. Architects.
- [04:41]
Camille Paglia
Yeah, yeah, nobody entirely understands them. I mean, the key. The thing I realized when I went back. So I had an example of a neural network. Neural networks are at the heart of his approach of the deep learning revolution. All these huge models are neural networks, so called neural networks. And I was describing one that had actually been proposed by a former collaborator of Hinton's, but it didn't really represent what Hinton, it turns out, had in mind. But anyway, in this model, you know, it did was designed to handle language. It wasn't a working model. It was just kind of sketched out. And the key thing is you would have to. To get the model up and running, you would have to take the human understanding of the meaning of words and Implant them, transplant them to that machine. It had a way of representing the meaning of words. But everyone, most people assumed that humans would, would have to tell the machine what the wor. And it turns out they found a way for the machines to kind of figure it out, so to speak, and come up with a way of representing the meaning of words kind of on their own. And what they're doing is kind of reverse engineering, I would say human cognitive equipment, or at least the functional equivalent of human cognitive equipment, a way of representing the meaning of words, which presumably we have somewhere in our brain. And that's, that's what they do across the board through deep learning is they work backwards. They show them data that humans produce or data that comes into the human brain and the data that comes out of it in response or whatever. And they make the machines good at identifying pictures or predicting next word or whatever the machine has to kind of reverse engineer. I would put it this way, not everyone would, but I think it's fair to say reverse engineers cognitive functionality. And that's why it's so powerful. I mean, as you know, I think there's going to be an absolute earthquake because in principle, there's no part of the human mind that cannot be kind of functionally replicated.
- [06:57]
Bob Cesca
Before we get to the earthquake thing, just quickly, when you talk about meaning, part of it is, and you should know, if you don't already understand this, that I'm the dumb person in this conversation and Camille is the smart one. So I'm asking for the dumb people questions. But by meaning, you kind of mean like context and background and possible different inflections here and all of just the built up kind of human understanding of things. They're not doing any of that. They're doing pattern recognition. Is that a way of looking at it?
- [07:32]
Camille Paglia
I think that's misleading to do the pattern matching. See, here's the thing, people, the training. You've heard about training, pre training, whatever of these big models, they get them better and better, predicting the next word in the case of next token prediction. And people often refer to that as learning. I think in many ways it's more like evolution. It's more like natural selection and mutations happen and they're preserved. And at the end of it you actually have cognitive equipment. So it's not. It is in order to be good at pattern matching, they have to build something deeper. And when I say meaning, I mean a good example. It's just, I mean, you know, the thing that allows us to use words. Well, so there have been Theories of how the brain represents meaning. So for one theory is, for example, that you have like the dimensions of the meaning words. So suppose like, you know, you got tiger and rattlesnake and you just map them along two dimensions. One is how lethal they are and they both rate. They both have a high number. Another is how fast they are, and one has a much lower number than the other. And the theory has long been that maybe the human brain, you know, has. Has like super high dimension. It breaks these words down into the features of their meaning. And it turns out that that's what these, these large language models actually do do. But they to some extent do it on their own. I mean, I won't get into any more details, but that's what I'd really emphasize. So all the ways of minimizing what they're doing. Stochastic parrots, even pattern matching, although that's technically accurate, I just think Mrs. Leaves out the most important part of what's happening.
- [09:19]
Bob Cesca
Suppressed evolution, as you put it.
- [09:21]
Camille Paglia
Yeah.
- [09:21]
Bob Cesca
Like in a very short.
- [09:22]
Camille Paglia
And also learning. So when they learn to speak English, they're learning something that we learn in the course of a lifetime. But when they build the system for representing the meaning of words, which they also do, they're doing something that evolution did in the course of our evolutionary history. They're doing both at once. I would say.
- [09:40]
Matt Welch
There's a lot of stuff going on here. I'm a little afraid that we've kind of dove in at the deep end and that we've not sufficiently underscored the fact that Matt Welch is a Luddite who does not care about AI is allergic to it. I'm not sure you've ever used ChatGPT.
- [09:58]
Bob Cesca
Oh, just, you know, when you do the Google search now, they do a thing at the top that says AI, not ChatGPT. Yes, yes.
- [10:04]
Matt Welch
Yeah, so. And do you know what Claude is?
- [10:07]
Bob Cesca
That's one of the things.
- [10:08]
Matt Welch
Yep. So that's the starting point. And Robert, I think it's important for you to perhaps, and this is different, make the case for someone who is not really paying attention to this stuff except just hearing the term over and over again for why they should care. Because your book, your new book, which we haven't held up yet, Robert, writes the God Test has a rather provocative title. It's grand, it's interesting. Artificial Intelligence and our coming Cosmic Reckoning. Wow, that sounds like a big deal. We all remember the racial reckoning and how wonderfully that went and how there were no problems associated with that at
- [10:46]
Bob Cesca
All Camille's very bad at sarcasm.
- [10:48]
Matt Welch
But this seems important. So help Matt Welch and his ilk understand why this is at all important. Absent perhaps the technical explication of how the technology works.
- [11:03]
Camille Paglia
Yeah, I think. And the book is an argument to this effect, and people can judge the argument that this is going to be the most abrupt transformation of human society ever. The most dramatic. It's going to be a complete earthquake. And we need to do certain things pretty soon to get through it in good shape, including, I think, you know, the reason it's called the God Test is there are certain things I think we have to do. I mean, the term has various meanings, the title has various meanings, but
- [11:44]
Matt Welch
I
- [11:45]
Camille Paglia
think this is gonna call for a cohesive global community, for example, and that would call for doing less in the way of fighting wars and stuff. But the reason I went into the fact that this thing, you can just kind of turn it loose and it will reverse engineer parts of the mind, so to speak, is because that's why I just don't think there's an end in sight to its progress. That's why I think its impact is gonna be so great. And, you know, I just. There's almost no part of life that I think won't be affected. So you. Now, you don't use. You don't. Do you not use large language models? Because you would find that they're amazing.
- [12:27]
Bob Cesca
I'm sure they could do some things
- [12:28]
Camille Paglia
for you, tools for certain kinds of
- [12:29]
Bob Cesca
things, for certain kinds of research and, you know, baseball stats and stuff. That's really important. But I haven't yet got to a place where it was. It seemed like nothing that I needed to do anything with.
- [12:41]
Camille Paglia
Yeah, I think you'd be surprised. I mean, I just, like two days ago, for the first time, finally tried this vibe coding thing. Cause I want to do a website for the book. And it's kind of shocking. It's kind of shocking what it can do in that realm. And I was already impressed with the research capabilities. I mean, you do have to keep an eye on it. It can hallucinate, but can.
- [13:05]
Matt Welch
Yes, lots of but.
- [13:08]
Bob Cesca
So, like, all right, you're talking about an earthquake. You're talking about, like, within the next decade.
- [13:12]
Camille Paglia
Where are we?
- [13:12]
Bob Cesca
Are we at, like a 3.4 on the Richter scale? Are we already up to 5.7? Like, is it all around me in a way that I'm not. I'm just too stupid to realize. And that's probable, I don't think.
- [13:24]
Camille Paglia
How does it hit in very big ways. I Mean, you know, young computer programmers coming out of college are having more trouble finding work. There are things like that. There are these, you know, isolated cases of parents saying that an LLM led to the suicide of their child. Who would, you know, who had the AI had become the kid's best friend and kind of walled it off. The parents would say from others and then finally from human friends and finally gave it some bad guidance in one sense or another. So there are a lot of, you know, when people complaining that it's, it's breaking up their marriages for a different reason because their spouse, it tells their spouse that they're always right in the marital arguments. So I mean, you're getting a lot of little things. I do think the impact on jobs is gonna be big and ultimately is gonna be a big impact on international relations. I mean, the threats that people have been warning about are becoming real or in terms of, you know, the possibility of kind of catastrophic things like somebody using it to help build a bioweapon. The models are getting, you know, more and more, the guardrails really need to work to avoid something like that. And the guardrails have a history of not really working in the long run. So I don't know, are you agnostic or no?
- [14:57]
Matt Welch
I mean, I'm bullish on the tech. I use it a lot. I have subscriptions to all of the frontier models. Pretty much I pay, you know, for the hundred dollar Claude subscription. I think I'm still paying for the $200 ChatGPT subscription. So I'm using them regularly in my work in a bunch of different contexts. I have built some apps. I am persuaded that these are useful tools, but I'm also aware of the news and the various critics of the industry who are deeply skeptical of what Anthropic and OpenAI are up to at the moment. Specifically, it's people who talk about just these industries where they've got this glut of compute that they've purchased, they have built out all these data centers and they are not anywhere near profitable. And they are finding that it is really, really difficult for a number of the organizations who have these subscriptions to these models to actually figure out just how much more productive are we getting. And in fact, you mentioned people having difficulty finding jobs. I think it also seems to be the case that a lot of folks over hired during COVID and have used AI as an excuse for getting rid of a bunch of people. But a lot of those teams are beginning to hire again because they're finding that there are all sorts of interesting, unexpected ways that the tech is impacting people's bottom line. So I do think it's useful to perhaps tell the story about why that skepticism is perhaps misguided or maybe even to just contextualize it, because I tend to think about it as something like pets.com like there was a time when the Internet was in its earlier stages where it seemed like everything was about to boom. And then we saw a bust and a massive reorganizing. But there are ways in which AI is not quite like.
- [16:48]
Camille Paglia
Well, I think that's actually a good analogy. And you know, I do think that there's a real chance that there will be like, you know, a bubble will burst. You know, there was the dot com bubble burst, but the Internet and the web continued to evolve and become, for better or worse, more and more pervasive parts of our lives. And social media went on to happen. And so that was kind of a case of misplaced bets, specific bets. And I think that may be happening now. I mean, it's such a fluid kind of environment, like right now, only over the last month, the business model, I mean, the following has happened in about the last four or five months. First, anthropic kind of displaces OpenAI as the lead horse because their revenue is just absolutely skyrocketing factor of 10, year over year. And then more than that, apparently, although
- [17:48]
Matt Welch
they're still privately held, so we don't really have the public information on their.
- [17:51]
Camille Paglia
That's true. I mean, they're gonna have to disclose stuff soon with an ipo. But my main point is, over the last month there's been kind of a rethinking of that because so many companies are saying this is too expensive and they're starting to use open source models for a lot of their work. And save the real, which is free.
- [18:11]
Matt Welch
Right.
- [18:12]
Camille Paglia
And save the real work for just the top tier. But that's not an indictment of the technology. What that means is we don't yet know who is going to capture the value and what business models will. And as for jobs, it's true that there have been these layoffs in tech companies, and some people think it reflects Covid over hiring, some people think it reflects AI. But there's definitely a reason that Mark Zuckerberg, in the same week that he announced 8,000 layoffs, said, oh, and also I'm going to start monitoring the keystrokes of all the workers.
- [18:49]
Matt Welch
It didn't go over well with the workers.
- [18:51]
Camille Paglia
No, but it's because of what I just said all he has to do, if he has the data that comes into them and the data they put out, that a good worker puts out, then he can just generate in a machine, just show that the machine and go through the training and then fire those workers because it will have whatever functional cognitive capacities allowed them to do their work. So we'll see. I do think there may be a lot of turbulence in the financial markets and anthropic and OpenAI could be dead in a couple of years for all I know.
- [19:26]
Matt Welch
And that still wouldn't kill the tech. Like the tech wouldn't.
- [19:28]
Camille Paglia
No, that would be. No, I absolutely don't think it's going to kill the tech.
- [19:33]
Matt Welch
Yeah, I think that's an important point for people to get. But I do want to talk a little bit more about the expectations for AI because anyone who's paid attention and certainly Matt, you know this as well, you've got doomers and you've got some sort of utopians and they both believe the same thing essentially that God is being built in a box and it will either deliver us to a world of zero scarcity or become paperclip maximizer or turn us all into gray goo or otherwise the computers will enslave civilization. And it's funny I say something like that out loud and I think for people who aren't really paying attention, they may have this expectation. No, no, no, that's not. I mean, you're being hyperbolic. No, I'm literally explaining how a lot of the people in the AI industry actually think about this tech. It's one of those two things. It is going to be revolutionary, but revolutionary in that sort of insane way. I think you have a perspective that I don't know how to characterize it. I think it does have the expectation that it's going to have this huge impact. You are concerned about the negative scenarios, but you don't really consider yourself a doomer. But at the same time, the book suggests the prescription, as you mentioned earlier, is we have to have this kind of global cadre of technologists and politicians that come together, that set the rules, that make certain that we don't build the wrong kind of thing.
- [21:00]
Camille Paglia
Well, I don't want an elite class of techno geniuses running things. We need, you know, things in some cases just the functional equivalent of arms control treaties so that like we know what's going on in China, China knows what's going on here, so that. And around the world we know that nobody's building a bioweapon with it or nobody's building a self replicating super hacker that takes out, you know, all the satellites or something. We, we need that kind of thing and I think other kinds of international governance. You want as little as possible. You this technology makes it even more important to avoid concentrations of power because it can be abused by people who monopolize power. But you're right that. Yeah, I'm sure more on board with international governance than you guys are than most people are on the doomer thing. You're right about that.
- [21:56]
Matt Welch
2.
- [21:58]
Camille Paglia
When I say it's going to be an earthquake, I'm talking about relatively mundane and near term effects like three, four, five years.