Arnold Kling on Twitter, FTX, and ChatGPT

Arnold Kling: Thanks. I’m chasing Mike Munger, I think.

Arnold Kling: Yeah. Well, I think this issue of content moderation is a very interesting problem. I think it would be an interesting problem in the abstract because you can make two types of errors. You can allow content on that you really shouldn’t, or you can ban stuff or downweight stuff that you’re wrongly banning or downweighting. I don’t know how to avoid that dilemma, that you’re going to make mistakes either way.

What’s weird about it at this point in history is it’s being done in the context of–there’s at least one group of people who are just very determined to have certain content banned and they think it ought to be banned. So, the free speech culture has been on the decline. And, the phrase ‘free speech’ itself, which used to be automatically a positive thing–a motherhood and apple pie thing–is being questioned really for the first time in my lifetime. So, it’s a weird time to try to deal with that, the dilemma.

Arnold Kling: Yeah. Well, I think the latest revelations are portrayed as showing pretty close to collusion between government and Twitter, so that, I think, brings the whole free speech issue right back into play.

As far as a private entity, I don’t know. It’s a great attempt by libertarians to create a distinction. But nowadays–I mean, let’s say, are universities private entities? I mean if they’re getting government funding?

So I think, in terms of a free-speech culture, that people have to believe that the mere expression of opinion is not a threat, it’s not violence. And, the term ‘hate speech’ is something that I don’t think should be thrown around very much, if at all. So, I’m one of these people–we’re both Jewish–if somebody wants to deny the Holocaust, go ahead, in my view. The right answer isn’t to ban them from saying it: it’s to present evidence that reasonable people would agree supports the view that you shouldn’t deny the Holocaust.

Arnold Kling: As people on the Left are discovering, right? They were hoping they could leave Musk in the dust, and that’s just not so easy to do.

Arnold Kling: Yeah. I feel more on solid ground myself when I talk about the economics of trying to support social media. So, if you want to switch over to that, I’d be happy to–

Arnold Kling: Yeah. So, a problem that’s been very acute in the Internet but actually precedes it, is a problem that: a lot of things that you want to sell don’t have the standard economic textbook–it’s called a U-shaped average cost curve, where you produce output up to a certain quantity and the average cost is declining. And then, you get beyond that point and average cost is increasing. And so, the solution to that, in most cases, the optimum is to set price equal to marginal cost.

Arnold Kling: It’s the cost of producing the next unit.

Arnold Kling: Yeah. As somebody once put it: you can’t grow all the world’s wheat in a single teapot. So, like you say, if you have to expand land, at some point the cost of growing wheat is going to be increasing.

The other type of good–it’s often called a natural monopoly–it has a very high fixed cost and low marginal cost. So, a classic pre-internet example would be a pharmaceutical. Well, let’s say you can produce a pill for pennies, but the fixed cost of getting it developed and tested might be hundreds of millions of dollars. So, if you price it at the marginal cost of pennies, you’ll never recover the fixed cost.

And that problem is very widespread on the Internet, because the cost of distribution is approximately zero. Cost of copying is approximately zero. So, you’ve got a lot of goods. Maybe the standard circumstance on the Internet is for there to be high fixed cost and low marginal cost.

And, that’s very much true of these social media outfits. In fact, it’s even more extreme than that because, in some sense, the cost of an additional user is negative. You want to have more and more users if you’re Facebook. So, you can’t get away with charging a marginal cost of zero or negative and recover the high fixed cost of developing Facebook or Twitter.

Arnold Kling: Right. Anyway, and that’s only one mechanism.

So, I go back to what I guess you could call a ‘classic’ book–although something that comes out around 1999 or 2000, can you call it classic?–Information Rules by Hal Varian and Carl Shapiro. And, they point out sort of all the ways that you can support something with high fixed cost, low marginal cost. Advertising is just one of them.

The one that I find most interesting, because you observe it in so many places–in fact, so many places that I have this catchphrase, ‘Price discrimination explains everything’–that we observe price discrimination. And, for something like Twitter, I think that’s a very natural thing.

So, he wanted to charge $8 a month for the blue check. I would almost guarantee you that he can charge $8 a month for something, because Twitter is a bundle of things and it’s very valuable to some people, and you can find some bundle of privileges that people will pay for. My initial guess was: having, let’s say, more than 1,000 followers, plus the ability to direct message people. There’s some bundle of rights and privileges that people would pay more for. And, yet you still can keep the casual user on–because you want to do that: the marginal cost of keeping them on is zero or even negative. You want them on. So, you want to keep the casual user and charge something to the user who has what we call ‘inelastic demand’–that is, they’re going to be willing to pay. And so, you want to charge more for the people who are willing to pay.

And, I’m confident that if they were to experiment and keep trying, they could find a way to price discriminate and make the service profitable.

Arnold Kling: Yeah. My understanding is that the advertising model on Twitter is unlike something like Google AdWords where it’s a turnkey thing. You go on to Google, you say: I want to buy 1,000 impressions of the word ‘Arnold Kling’ or ‘AP Statistics’, or whatever I might feel like buying the AdWords for. On Twitter, because it’s not a mass audience where you can easily identify your target, the ad campaign actually has to be designed professionally, really by the help of people from Twitter. So, some of these people–and maybe they weren’t fired or maybe they were–were involved in designing ad campaigns, and so they were revenue producers.

Arnold Kling: ‘Promoted tweet,’ or whatever. Yeah.

Arnold Kling: Yeah, I’ve never liked the advertising model on the Internet in general. I think there are better ways to try to support content, but somehow the advertising model became the standard thing. But, my view is that Google gets x percent of advertising, Facebook gets y percent, and everybody else collectively is fighting for 1 minus x minus y percent. And that’s not going to–you really shouldn’t try to play that game. You should try to find some other revenue model.

Arnold Kling: Yeah. Okay.

Arnold Kling: Okay. As best we know–okay, so let me take a little bit of a step back. So, everyone should understand that I have never been a crypto optimist. And, one of my lines is that I think the worst thing that happened to crypto was the speculative boom in Bitcoin and other currencies, because that gives people the impression that the use-case for crypto is wild speculation. And, you don’t want that to be your primary–there are other use-cases that seem like there’d be more promising–more that you’d want to build your reputation on.

Arnold Kling: Yeah. So, I mean, just to keep digressing a little bit: I think the interesting use-cases are to get around badly managed financial systems/regulations. So, if you’re in Argentina and I’m your relative living in America, and I want to send money to you, Argentina might–if I do it above board–Argentina might set an artificial official exchange rate, let’s say, of two pesos to the dollar, when it really should be five pesos to the dollar, like, in a free market. So, if I send you a remittance and do it above board, I’m sacrificing over half of the value by transacting at this artificial official exchange rate.

So, in that case, Bitcoin becomes–you know, keeping it, your remittance, outside of the view of the official government, but keeping it with something that can retain more of its value. That’s a real use-case. So, remittances are a real use-case.

The other is if you have a banking system that just doesn’t work. So, the example that I’ve read, supposedly, of this is Vietnam, where if you put your money in your bank, that’s actually riskier than putting it in crypto. And, that could either be because the banks are poorly regulated, or because the government doesn’t stop inflation, or what have you. So, that’s another use-case.

But–and this gets to my view of it–is that, all these use-cases are somewhat adjacent to criminal activity. That is, that you’re getting out of the regulated world, out of the formal world into the informal world, into the underground economy. So, you’re lying down with dogs, so to speak, and you’re probably going to wake up with fleas.

Maybe that’s wrong. Anyway, that’s–

Arnold Kling: Yeah, go ahead.

Arnold Kling: No, you finish.

Arnold Kling: Yeah. No, I think if the point is that if the financial system’s working well, it will at some point figure out an efficient way to do cross-border money transfers. There’s an incentive to do that. You can use computers. You don’t have to use computers in the really complex, resource-intensive way that Bitcoin uses it.

So, anyway, back to FTX. If–one of the challenges that crypto faces to do legitimate use-cases, like cross-border between the United States and the United Kingdom–is that it’s not easy to learn how to use. There are all these stories about people losing track of their key, so that they’ve effectively lost their crypto. I know people who say that they’re very technically competent but cannot–don’t feel competent to actually own their own crypto.

So, this reminds me of the state of the worldwide web around 1994, when what was holding back mass adoption was that it was very hard for an individual to figure out how to get the Internet protocols onto their computer.

So, the mass adoption problem tends to be solved–unfortunately, from my idealistic point of view–by large, centralized entities. So, as the Internet became hard to search; Google kind of takes over. As people became interested in this social media phenomenon, Facebook takes over.

The dream as of the early 1990s was that–because the Internet structurally was very decentralized, you know, just in terms of the underlying technology–that the kind of social systems and economic systems that would be built on top of the Internet would be equally decentralized and we would all be in this kind of, sort of libertarian utopia of a decentralized system.

Well, that idealistic dream has come back with crypto. But, my experience leads me to believe that you’ll need these centralized entities to–that that will be a necessary condition for mass adoption and for overall efficiency.

So, in crypto, what that means is third parties, like exchanges, become very important. And, that’s what, kind of, FTX was supposedly positioned as: as an exchange. An exchange can make money–either it can make money by charging a fee. Like you say, a bank can charge a fee for trading dollars for pounds. Or, it can make money on what’s called a ‘dealer spread,’ which is again, like, charging you one exchange rate to go from dollars to pounds and a different exchange rate to go from pounds to dollars, so that it in effect makes a profit on both sides of the transaction.

So, that’s how an exchange should work.

But, in the world of crypto, I guess these exchanges are holding your crypto. They’re not just sort of instantly making a transaction between these two. Maybe they should. Maybe there are exchanges that do that: they don’t really hold onto it.

But, FTX was doing that. So, in that sense, it’s acting like a bank, right? I’ve got my–supposedly, I’ve got my cryptocurrency being stored there, and I can take it out any time I want.

But, when you’re running a bank, there’s always the possibility that you’ll loot the bank doing whatever. In this case, they had the–he had over a hundred different entities that he was the owner of or founder of, one of them being this Alameda firm.

And, as far as we can tell, he took a lot of people’s money that they’d put into the FTX exchange and kind of did a round trip trade with Alameda that somehow made both of them seem to have better balance sheets than they really had.

And, somewhere in this process, either Alameda or FTX caused a lot of this money to disappear. And sort of, could be bad investments, maybe things that–maybe consumption by Sam Bankman-Fried and his cronies. Who knows where the money went? It’s mysterious how so much money could, kind of, disappear. I mean, I can’t begin to tell you how it actually happened, and I’m not sure at this point anyone can.

Arnold Kling: Yeah. When you mentioned all his [Sam Bankman-Fried] media tour, some of that could be that he’s just addicted to that. He was a great–he was superb at hyping himself in media terms. Or in some sense he’s preparing to plead insanity. Like, who would be crazy enough to give a media tour if people are suspecting you of fraud? So, ‘I must be crazy. So, go easy on me.’

Arnold Kling: Yeah, I always hated that.

Arnold Kling: I mean, I think months ago that the effective altruism stuff that he was pushing, I thought that that’s committing the sin of believing that you know how to fix the world. It’s the same sin that a politician–an authoritarian politician commits–of saying, ‘My central planning is superior to the market’s trial and error.’ It’s really the same thing. And, I kind of called him out on that months ago, said I just didn’t like the guy for that reason.

Arnold Kling: Yeah. He does seem like a very disturbed individual. And, but somehow he’s disturbed in such a way that he appeals personally to a lot of people.

And, in addition to his personal, quote, “charisma,” he threw a lot of money around to people in the media in order to get good publicity. Including, like, supporting their publications, giving them, like, a quarter of their funding, or what have you. And, you don’t know to what extent he bribed individual journalists. Michael Lewis spent a lot of time with him. Was that because Sam Bankman-Fried plied Michael Lewis with lots of money and other stuff? Or was that genuine interest on Michael Lewis’s part? Those are the–

Arnold Kling: Right. But, oh, the potential for conspiracy theories here, I think, is just huge. I mean, I think this is–

Arnold Kling: Well, again, we don’t know what extent it was deliberate fraud. This guy, I think it’s Marc Cohodes, who called out FTX back in October, just said, ‘It’s a bunch of interns running around trying to do an exchange. They have no more business–they shouldn’t be running a lemonade stand, much less an exchange,’ was one of his lines. He thinks that one of the main attorneys for Bankman-Fried was really running the whole show.

Well, so you get the picture that he was using Bankman-Fried as this frontman where he was deliberately pulling off fraud.

So, there’s all these potential conspiracy theories. So, the fact that he gave so much money to politicians and then was invited to weigh in on regulation. And, he’s been invited to testify in front of Congress. Usually, when somebody like that is invited to testify in front of Congress, it’s the chance for the Congressman to get to show up on TV yelling at the guy. Who knows? Maybe these are all people he’s paid for and he’s going to be treated well. I don’t have a specific conspiracy theory, but I guarantee you there are going to be a lot of plausible conspiracy theories coming out of this.

Arnold Kling: Yeah, she’s on the cover of lots of magazines and clearly charmed Henry Kissinger and other dignitaries–I think it was Kissinger, was one of the people on her Board. I hope I’m not maligning him.

Arnold Kling: Well, yeah. So, if you’d looked at the very start, I would have been skeptical–equally skeptical–of Jeff Bezos as an entrepreneur and Sam Bankman-Fried as an entrepreneur. As they start out, they’re doing wildly ambitious things. They’ve got a lot of personal charisma, and you can’t believe they know what they’re doing. Like, my line on Amazon as they were getting going is, ‘Okay, so what Walmart has to do to compete with Amazon is develop a website. What Amazon has to do to compete with Walmart is develop a whole logistics system for international shipping of goods and so on. So, who’s going to win? Well, it’s clearly Walmart if all they have to do is develop a website.’ And, that didn’t turn out to be the case at all.

So, it turned out that Jeff Bezos had tremendous substantive skill–I think amazing substantive skill–as a business leader. I mean, there are always people who will denigrate anybody, but to me he’s, like, the most amazing business leader you could find. And, Sam Bankman-Fried turned out to be a total fraud, or a total incompetent, or some combination of those two. And, if you’re a venture capitalist, it’s probably hard to tell them apart at the beginning. I mean, clearly there would be people who would. But, the ambition–a venture capitalist once described it as, ‘I’m not going for singles and doubles. What I want is not even a home run, but a home run that goes outside the stadium and rolls down the street.’

For an ordinary investor like you or me, if we’re going to put money into a particular stock–which we typically don’t: we typically buy index funds–but if we’re going to do that, sort of we’d say, ‘Well, I really want to avoid losing everything.’ And, the upside doesn’t have to be unlimited. And, with a venture capitalist, it’s the other way around: ‘Oh yeah, I lose everything on a lot of my investments, but every once in awhile I hit the jackpot.’ And, that kind of mentality means you want to back a founder who has ridiculously high ambitions and who has very grandiose visions.

And so, as they’re starting out, Sam Bankman-Fried has a grandiose vision, or he promulgates one, and Jeff Bezos gives you a grandiose vision. And it seems like you should back both of them. And in fact, I think what a lot of venture capitalists feel if they’ve been around a long time is, if they missed Amazon, that’s the kind of error they don’t want to make again. Whereas, betting on Sam Bankman-Fried is, like, ‘Okay, made that mistake. Lost some money. Happens all the time.’ So, it’s a very different perspective than what you and I would have looking at an individual investment.

Arnold Kling: So, yeah, there’s just a different incentive and a different kind of experience base that venture capitalists are dealing with.

Arnold Kling: Yeah, so it’s susceptible to a charismatic goofball in a way that the rest of us wouldn’t be.

Arnold Kling: Yeah. Basically, if you’re in middle management and you propose some very potentially spectacular project for your company, you’re going to have to sit in meetings. And, they are going to be 10 people in the room, and if one of them says, ‘This won’t fly,’ that’s it. You’re done. And, that’s actually rational from the corporation’s point of view, because as a middle manager, you have very little skin in the game. Suppose your project is going to cost $20 million; the upside is $500 million or a billion dollars worth of value. You’re not going to get very much of that upside.

On the other hand, if the corporation loses the $20 million, you’re going to lose zero of the downside.

So, you don’t have much skin in the game when you’re playing with the corporation’s money. And so, a corporation that doesn’t set up a bureaucracy that’s skeptical of ideas that come from middle management is just going to end up throwing money at lots of projects that it can’t–that don’t turn out well, and it just squanders its money.

So, a vetocracy, in short, is a rational thing to have. But, that’s why–so, that’s why innovations typically start out outside of big corporations, because it’s just rational for them to not approve projects for people who don’t really have that much skin in the game.

Arnold Kling: Yeah. Yeah, you need the upside, and you’re right. And, the founder does have a lot of skin in the game, typically.

Arnold Kling: Yeah, that’s ELIZA, right?

Arnold Kling: The artificial psychiatrist.

Arnold Kling: Yeah. Yeah, I think right now it’s at the level of an undergraduate BS artist who hasn’t studied enough for the test. Now, computer–

Arnold Kling: Right. So, I think what’s impressive–what strikes people–is that it can sound like a human being, and you can even have it imitate a particular human being. Like, if somebody has a lot of content out there, whether it’s Shakespeare or Tyler Cowen or somebody, and you say, ‘Write an essay about X in the style of Shakespeare, or in the style of Tyler Cowen,’ it will imitate that. So, it has a great facility with putting together words. It’s a great BS artist. But a lot of the stuff it does is BS.

One of the first things I did, is I said: ‘Write an essay about Arnold Kling, economist.’ And it proceeded to tell a bunch of lies. It said I was born in New York; I was actually born in St. Louis. It said I was born in 1960; it was actually 1954. And so on. It said I wrote The Myth of the Rational Voter; that’s written by Bryan Caplan.

And, it just said this with a straight face–like, you know–and along with some things that were true about me, but it was just so–and somebody else described, they had asked it for an essay on Thomas Hobbes and it came back with something that was a description of the philosophy of John Locke, which it attributed to Thomas Hobbes, even though Locke is different–in some ways the opposite of Hobbes.

So, I get the sense that it sort of looks for connections related to the words that you include in your prompt, and then its real skill is assembling them into a good-sounding narrative–a narrative that sounds like what a human would put together.

Yeah, it may evolve into something different. There’s this long essay by Gary Marcus, who is a real skeptic of the whole method of these sort of neural networks.

Well, so let me go back– this may be a weird thing–but when I was with Freddie Mac in about 1991, 1992, 1993, we were looking for ways to automate the underwriting process. We thought we could start to use computers to supplement or replace human underwriters. And, the approach we first looked at was known as expert systems. Basically, you would program a computer with a bunch of rules that mimicked what a human underwriter would use.

And then, at one point we heard a presentation on credit scoring, which operates–which was kind of an early version of big-data machine learning as we now do it, in that it just said, ‘Let’s just look for correlations with default.’ And, this wasn’t even default on mortgages: it was just default on credit cards and so on. And, they found variables that statistically were important that human underwriters weren’t even paying attention to. One was credit capacity. Like, if you have $30,000 of unused credit–that is you’ve got some credit cards: let’s say three credit cards each with a credit limit of $10,000 and you’re not using any of them–that’s a real strong sign that you’re not going to default. Conversely, making a lot of inquiries–like, if you’ve applied for five credit cards in the last two weeks, that’s a real red flag. Well, human underwriters never even knew that. The computers figured it out.

So, long story short: The statistical-based or machine-learning type–what we would now call machine-learning approach to automating underwriting–won out and the expert systems lost.

And, what’s happened in AI, I think, has been something similar. But, Gary Marcus claims that these sort of purely correlation-oriented, statistically-oriented forms without any domain knowledge entering the picture, that we’ve reached their limit. His claim is they’re not going to get a lot better; and you’re going to have to bring in other types of learning.

And, again, my experience would lead me to believe it’s very hard to integrate them, actually. It’s very hard to integrate two different types of learning, two very different methods, without one method hurting the other.

Just like, I think, in chess, it’s become very hard to integrate a human’s decision with the computer’s decision. When the computers were first starting to get really good at chess, a human could spot a situation: ‘Well, here’s a situation where based on the layout of the board, the computer is going to make a mistake, so I’m going to override it.’ But then, you reach the point where the human override on average ends up worse than not overriding.

Anyway, so it’s very hard to integrate two completely different methods of learning. And so, we may have reached a point or come close to reaching a point where the ChatGPTs of the world don’t get better.

Arnold Kling: Yeah, it’s way early–way early to tell.

Arnold Kling: So, you’re more bullish than I am?

Arnold Kling: Yeah. No, I think the most immediate impact I can see is that it will create a big issue of deception and of detecting deception in terms of, first of all, content.

Like, the biography of me that it put together immediately raises the issue of, you’d have to look through it to find out what’s true and what’s not.

And then, the issue of deception, self-deception of: is this who it says it is, right? So, if it gets really good at imitating Russ Roberts, could there be 100 Russ Roberts tweets out there and only one of them really comes from you? I mean, just the ability to deceive other people or deceive people about who you are as a bot.

And, naturally, the response is going to develop a bot that can detect–tell the difference between a bot and a human. So, you’ve got this arms race going.

You’ve got, on the one hand, the bot is trying to super-pass the Turing test and convince you that it’s a person and a particular person. And, on the other side you’ve got a bot that’s trying to distinguish between the two because the human effort involved in determining the difference between a bot and a human is too difficult.

I mean, that’s I think why Elon Musk’s $8 to get the blue check didn’t work. It probably costs more than $8 a person to verify that the person is who they say they are.

So, all of a sudden we’re going to have this arms race over something that wasn’t so much of an issue before, of phony identity–either phony identity as a human or phony identity as a particular human. So, people trying to achieve that, or computers trying to achieve that–the ability to mimic a particular human being–versus how do you detect and stop that?

And, the only way you can do that at scale is with another bot. So, that’s just a whole new arms race that I think is going to be the most immediate impact of these open AI projects.

Arnold Kling: ‘Works for me.’

Arnold Kling: Well, it depends. If, on the job, they’re going to need to talk fluently in English with people, it’s probably a bad thing. No, I mean, that’s going to be it. There are going to be use-cases where it’s okay and use-cases where it’s terrible, and being able to separate the two is going to be really hard.

No, but, I do agree with you. It’s an amazing development. It’s amazing how fast it is. A week ago, I was in Austin and met this kid who probably started shaving six months ago and says, ‘I’m really into machine learning.’ I said, ‘Well–‘ I wish I could buy shares in the guy. I mean, it’s clearly going to be the big trend for a while.

And, one anecdote is there was somebody who found a 16-question IQ [intelligence quotient] test and they gave it to the ChatGPT, and he said it scored about as much as a human who has an IQ of 100. My experience with computers is, once they reach a certain point and they’re on the up-curve, it’s not like they stop. So, will it take less than six months for the thing to come back with an IQ of 130? Probably. So, watch this space. I’m sure I’ve said at least one thing discussing it that’s going to look idiotic a few months from now.

Arnold Kling: Okay, thanks.