Month: January 2024
My new podcast with Dwarkesh Patel
We discussed how the insights of Hayek, Keynes, Smith, and other great economists help us make sense of AI, growth, risk, human nature, anarchy, central planning, and much more.
Dwarkesh is one of the very best interviewers around, here are the links. If Twitter is blocked to you, here is the transcript, here is Spotify, among others. Here is the most salacious part of the exchange, highly atypical of course:
Dwarkesh Patel 00:17:16
If Keynes were alive today, what are the odds that he’s in a polycule in Berkeley, writing the best written LessWrong post you’ve ever seen?
Tyler Cowen 00:17:24
I’m not sure what the counterfactual means. Keynes is so British. Maybe he’s an effective altruist at Cambridge. Given how he seemed to have run his sex life, I don’t think he needed a polycule. A polycule is almost a Williamsonian device to economize on transaction costs. But Keynes, according to his own notes, seems to have done things on a very casual basis.
And on another topic:
Dwarkesh Patel 00:36:44
We’re talking, I guess, about like GPT five level models. When you think in your mind about like, okay, this is GPT five. What happens with GPT six, GPT seven. Do you see it? Do you still think in the frame of having a bunch of RAs, or does it seem like a different sort of thing at some point?
Tyler Cowen 00:36:59
I’m not sure what those numbers going up mean, what a GPT seven would look like, or how much smarter it could get. I think people make too many assumptions there. It could be the real advantages are integrating it into workflows by things that are not better GPTs at all. And once you get to GPT, say, 5.5, I’m not sure you can just turn up the dial on smarts and have it, like, integrate general relativity and quantum mechanics.
Dwarkesh Patel 00:37:26
Why not?
Tyler Cowen 00:37:27
I don’t think that’s how intelligence works. And this is a Hayekian point. And some of these problems, there just may be no answer. Like, maybe the universe isn’t that legible, and if it’s not that legible, the GPT eleven doesn’t really make sense as a creature or whatever.
Dwarkesh Patel 00:37:44
Isn’t there a Hayekian argument to be made that, listen, you can have billions of copies of these things? Imagine the sort of decentralized order that could result, the amount of decentralized tacit knowledge that billions of copies talking to each other could have. That in and of itself, is an argument to be made about the whole thing as an emergent order will be much more powerful than we were anticipating.
Tyler Cowen 00:38:04
Well, I think it will be highly productive. What “tacit knowledge” means with AIs, I don’t think we understand yet. Is it by definition all non-tacit? Or does the fact that how GPT-4 works is not legible to us or even its creators so much? Does that mean it’s possessing of tacit knowledge, or is it not knowledge? None of those categories are well thought out, in my opinion. So we need to restructure our whole discourse about tacit knowledge in some new, different way. But I agree, these networks of AIs, even before, like, GPT-11, they’re going to be super productive, but they’re still going to face bottlenecks, right? And I don’t know how good they’ll be at, say, overcoming the behavioral bottlenecks of actual human beings, the bottlenecks of the law and regulation. And we’re going to have more regulation as we have more AIs.
You will note I corrected the AI transcriber on some minor matters. In any case, self-recommending, and here is the YouTube embed:
Wednesday assorted links
1. The Frick Museum will reopen with 14 (!) evening bars.
2. Sebastian Barry in conversation with Roy Foster.
3. On ideological gender disparities in Korea.
4. Those new service sector jobs, What is Intervenor Compensation?, and “robot wranglers” (WSJ).
5. Is Petro stifled in Colombia?
6. Further fresh Vitalik. Includes coverage of his childhood, more personal than about mechanism design.
7. Is there really a “National Hug an Economist Day”?
8. Other than this tweet, I know nothing about the new Catholic Institute of Technology.
In Praise of Non-conformity
In this age, the mere example of nonconformity, the mere refusal to bend the knee to custom, is itself a service. Precisely because the tyranny of opinion is such as to make eccentricity a reproach, it is desirable, in order to break through that tyranny, that people should be eccentric. Eccentricity has always abounded when and where strength of character has abounded; and the amount of eccentricity in a society has generally been proportional to the amount of genius, mental vigour, and moral courage which it contained. That so few now dare to be eccentric, marks the chief danger of the time.
I saw this quote on Facebook and thought immediately of my friend Bryan Caplan. Bryan’s book of essays, is an excellent guide not simply to Bryan’s non-conformism but also on how to be a successful non-conformist in a conformist world.
What should Stripe Press reprint?
Stripe Press just launched a call for reprints to enrich our collection with out-of-print books suggested by our readers. Given the shared interests of our audiences, Tam and I thought it would be fitting to reach out. Would you be open to featuring our Google Form in the Marginal Revolution links roundup? Your support would be invaluable in reaching a wider, engaged audience.
The Google Form is open…
Did the Trump tariffs help the heartland?
No, but they did get him some votes there:
We study the economic and political consequences of the 2018-2019 trade war between the United States, China and other US trade partners at the detailed geographic level, exploiting measures of local exposure to US import tariffs, foreign retaliatory tariffs, and US compensation programs. The trade-war has not to date provided economic help to the US heartland: import tariffs on foreign goods neither raised nor lowered US employment in newly-protected sectors; retaliatory tariffs had clear negative employment impacts, primarily in agriculture; and these harms were only partly mitigated by compensatory US agricultural subsidies. Consistent with expressive views of politics, the tariff war appears nevertheless to have been a political success for the governing Republican party. Residents of regions more exposed to import tariffs became less likely to identify as Democrats, more likely to vote to reelect Donald Trump in 2020, and more likely to elect Republicans to Congress. Foreign retaliatory tariffs only modestly weakened that support.
That is from a new NBER working paper by David Autor, Anne Beck, David Dorn, and Gordon H. Hanson.
From the beginning, “neoliberalism” was an obnoxious term
It was meant as an insult, implying that Mises – a marginalist – was trying to salvage 19th century liberal economics from the collectivist attacks of the Marxist left and the Nazi right, hence the “neo” moniker being attached.
One of the main promoters of this use was Othmar Spann, a rival of Mises on the University of Vienna faculty. Spann was a prominent proto-Nazi intellectual. In 1924 he added a disparaging chapter on “neoliberalism” to the new edition of his economics textbook.
By the time Mises arrived in Paris in 1938 for the CWL gathering, he had endured a decade and a half of simultaneous disparagement as a “neoliberal” by Nazis and Marxists. It should be no surprise that he was not keen to adopt the label himself.
Here is the full Phil Magness tweet storm.
Tuesday assorted links
1. Puffins are excellent swimmers.
2. How good are top Chinese LLMs?
3. Multicellularity older than we had thought?
4. “Online, she goes by the name Ballerina Farm, and millions of people watch her almost daily videos depicting her life in the countryside 30 miles from Salt Lake City.” (NYT, interesting throughout).
Vitalik on AI and crypto
…we can expect to see in the 2020s that we did not see in the 2010s: the possibility of ubiquitous participation by AIs.
AIs are willing to work for less than $1 per hour, and have the knowledge of an encyclopedia – and if that’s not enough, they can even be integrated with real-time web search capability. If you make a market, and put up a liquidity subsidy of $50, humans will not care enough to bid, but thousands of AIs will easily swarm all over the question and make the best guess they can. The incentive to do a good job on any one question may be tiny, but the incentive to make an AI that makes good predictions in general may be in the millions. Note that potentially, you don’t even need the humans to adjudicate most questions: you can use a multi-round dispute system similar to Augur or Kleros, where AIs would also be the ones participating in earlier rounds. Humans would only need to respond in those few cases where a series of escalations have taken place and large amounts of money have been committed by both sides.
This is a powerful primitive, because once a “prediction market” can be made to work on such a microscopic scale, you can reuse the “prediction market” primitive for many other kinds of questions:
- Is this social media post acceptable under [terms of use]?
- What will happen to the price of stock X (eg. see Numerai)
- Is this account that is currently messaging me actually Elon Musk?
- Is this work submission on an online task marketplace acceptable?
- Is the dapp at https://examplefinance.network a scam?
- Is
0x1b54....98c3actually the address of the “Casinu Inu” ERC20 token?You may notice that a lot of these ideas go in the direction of what I called “info defense” in .
There are many other points, such as AIs as oracles, read it here, interesting throughout.
How Genes and Investments Interact in the Formation of Skills
That is the subtitle of a new AER paper by Mikkel Aagaard Houmark, Victor Ronda and Michael Rosholm. Here is the abstract, which points out some mechanisms of genes-environment interaction which are often overlooked:
This paper studies the interplay between genetics and family investments in the process of skill formation. We model and estimate the joint evolution of skills and parental investments throughout early childhood. We document three genetic mechanisms: the direct effect of child genes on skills, the indirect effect of child genes via parental investments, and family genetic influences captured by parental genes. We show that genetic effects are dynamic, increase over time, and operate via environmental channels. Our paper highlights the value of integrating biological and social perspectives into a single unified framework.
Here are less gated versions of the paper.
The economics of illicit sand markets
Very few people are looking closely at the illegal sand system or calling for changes, however, because sand is a mundane resource. Yet sand mining is the world’s largest extraction industry because sand is a main ingredient in concrete, and the global construction industry has been soaring for decades. Every year the world uses up to 50 billion metric tons of sand, according to a United Nations Environment Program report. The only natural resource more widely consumed is water. A 2022 study by researchers at the University of Amsterdam concluded that we are dredging river sand at rates that far outstrip nature’s ability to replace it, so much so that the world could run out of construction-grade sand by 2050. The U.N. report confirms that sand mining at current rates is unsustainable.
And:
Most sand gets used in the country where it is mined, but with some national supplies dwindling, imports reached $1.9 billion in 2018, according to Harvard’s Atlas of Economic Complexity.
Companies large and small dredge up sand from waterways and the ocean floor and transport it to wholesalers, construction firms and retailers. Even the legal sand trade is hard to track. Two experts estimate the global market at about $100 billion a year, yet the U.S. Geological Survey Mineral Commodity Summaries indicates the value could be as high as $785 billion. Sand in riverbeds, lake beds and shorelines is the best for construction, but scarcity opens the market to less suitable sand from beaches and dunes, much of it scraped illegally and cheaply. With a shortage looming and prices rising, sand from Moroccan beaches and dunes is sold inside the country and is also shipped abroad, using organized crime’s extensive transport networks, Abderrahmane has found. More than half of Morocco’s sand is illegally mined, he says.
Of course these are usually unowned, unpriced resources:
Luis Fernando Ramadon, a federal police specialist in Brazil who studies extractive industries, estimates that the global illegal sand trade ranges from $200 billion to $350 billion a year—more than illegal logging, gold mining and fishing combined. Buyers rarely check the provenance of sand; legal and black market sand look identical. Illegal mining rarely draws heat from law enforcement because it looks like legitimate mining—trucks, backhoes and shovels—there’s no property owner lodging complaints, and officials may be profiting. For crime syndicates, it’s easy money.
Here is the full Scientific American piece by David A. Taylor.
Monday assorted links
1. Mexican investment is doing just great.
2. In praise of double majors.
3. How to do things if you don’t have talent (does this mean you do have talent?).
4. The coming of numeracy to 17th century England. And a new project Death by Numbers.
5. Those new service sector jobs: “After years shepherding children from one minute to the next, moms and dads hire $250-an-hour counselors to help them learn to live on their own.” (WSJ)
6. Okie-dokie: The Democrats’ new permitting-reform bill will spend $3 billion to help non-profits increase their participation in the environmental review process (Atlantic). Excerpt here.
I appear with Nicolas Bejarano on the Art Salon podcast
The Spotify and other links are here, mostly we discuss economics of the arts and thus this differs from many of my recent podcasts.
Hot parents, richer kids?
Since the mapping of the human genome in 2004, biologists have demonstrated genetic links to the expression of several income-enhancing physical traits. To illustrate how heredity produces intergenerational economic effects, this study uses one trait, beauty, to infer the extent to which parents’ physical characteristics transmit inequality across generations. Analyses of a large-scale longitudinal dataset in the U.S., and a much smaller dataset of Chinese parents and children, show that a one standard-deviation increase in parents’ looks is associated with a 0.4 standard-deviation increase in their child’s looks. A large data set of U.S. siblings shows a correlation of their beauty consistent with the same expression of their genetic similarity, as does a small sample of billionaire siblings. Coupling these estimates with parameter estimates from the literatures describing the impact of beauty on earnings and the intergenerational elasticity of income suggests that one standard-deviation difference in parents’ looks generates a 0.06 standard-deviation difference in their adult child’s earnings, which amounts to additional annual earnings in the U.S. of about $2300.
That is from a new NBER working paper by Daniel S. Hamermesh and Anwen Zhang.
Conditional Approval for Human Drugs
Recently a new drug to extend lifespan was granted conditional approval by the FDA–the first drug ever formally approved to extend lifespan! (By the way, the entrepreneur behind this breakthrough, Celine Halioua, is an emergent ventures winner for her earlier work rapidly expanding COVID testing. Tyler knows how to spot Talent!)
Great news, right? Yes, but there are two catches. First catch: the drug is for extending the lifespan of dogs. Second catch: Conditional approval is only available for animal drugs. Conditional approval was permitted for animal drugs beginning in 2004 for minor uses and/or minor species (fish, ferrets etc.) and then expanded in 2018 to include major uses in major species. What does conditional approval allow?
Conditional Approval (CA) allows potential applicants (referred to from this point as “sponsors”) to make a new animal drug product commercially available after demonstrating the drug is safe and properly manufactured in accordance with the FDA approval standards for safety and manufacturing, but before they have demonstrated substantial evidence of effectiveness (SEE) of the conditionally approved product. Under conditional approval, the sponsor needs to demonstrate reasonable expectation of effectiveness (RXE). A drug sponsor can then market a conditionally approved product for up to five years, through annual renewals, while collecting substantial evidence of effectiveness data required to support an approval.
Here is where it gets even more interesting. Why does the FDA say that conditional approval is a good idea?
First, it’s very expensive for a drug company to develop a drug and get it approved by FDA. Second, the market for a MUMS [Minor Use, Minor Species, AT] drug is too small to generate an adequate financial return for the company. The combination of the expensive drug approval process and the small market often makes drug companies hesitant to spend a lot of resources to develop MUMS drugs when there is so little return on their investment.
By allowing a drug company to legally market a MUMS drug early (before it is fully approved), conditional approval makes the drug available sooner to be used in animals that may benefit from it. This early marketing also helps the company recoup some of the investment costs while completing the full approval.
…Similar to conditional approval for MUMS drugs, the goal of expanded conditional approval is to encourage drug companies to develop drugs for major species for serious or life-threatening conditions and to fill treatment gaps where no therapies currently exist or the available therapies are inadequate.
Sound familiar? These are exactly some of the points that I have been raising about the FDA approval process for years. In particular, by bringing forward marketing approval by up to 5 years, conditional approval makes it profitable to research and develop many more new drugs.
Conditional approval is very similar to Bart Madden’s excellent idea of a Free to Choose Medicine track, with the exception that Madden makes the creation of a public tradeoff evaluation drug database (TEDD) a condition of moving to the FTCM track. Thus, FTCM combines conditional approval with the requirement to collect and make public real-world prescribing information over time.
But why is conditional approval available only for animal drugs? Conditional approval is good for animals. People are animals. Therefore, conditional approval is good for people. QED.
Ok, perhaps it’s not that simple. One might argue that allowing animals to use drugs for which there is a reasonable expectation of effectiveness but not yet substantial evidence of effectiveness is a good idea but this is just too risky to allow for humans. But that cuts both ways. We care more about humans and so don’t want to impose risks on them that we are willing to impose on animals but for the same reasons we care more about improving the health of humans and should be willing to risk more to save them (Entering a burning building to save a child is heroic; for a ferret, it’s foolish.)
I think that the FDA’s excellent arguments for conditional approval apply to human drugs as well as to (other) animal drugs and even more so when we recognize that human beings have rights and interests in making their own choices. The Promising Pathways Act would create something like conditional approval (the act calls it provisional approval) for drugs treating human diseases that are life-threatening so there is some hope that conditional approval for human drugs becomes a reality.
Dare I say it, but could the FDA be lumbering in the right direction?
An important new paper on the costs of climate change
Forthcoming in ReStud, I haven’t had the chance to read it yet:
To analyze climate change mitigation strategies, economists rely on simplified climate models — so-called climate emulators — that provide a realistic quantitative link between CO2 emissions and global warming at low computational costs. In this paper, we propose a generic and transparent calibration and evaluation strategy for these climate emulators that is based on freely and easily accessible state-of-the-art benchmark data from climate sciences. We demonstrate that the appropriate choice of the free model parameters can be of key relevance for the predicted social cost of carbon. The key idea we put forward is to calibrate the simplified climate models to benchmark data from comprehensive global climate models that took part in the Coupled Model Intercomparison Project, Phase 5 (CMIP5). In particular, we propose to use four different test cases that are considered pivotal in the climate science literature: two highly idealized tests to separately calibrate and evaluate the carbon cycle and temperature response, an idealized test to quantify the transient climate response, and a final test to evaluate the performance for scenarios close to those arising from economic models, and that include exogenous forcing. As a concrete example, we re-calibrate the climate part of the widely used DICE-2016, fathoming the CMIP5 uncertainty range of model responses: the multi-model mean as well as extreme, but still permissible climate sensitivities and carbon cycle responses. We demonstrate that the functional form of the climate emulator of the DICE-2016 model is fit for purpose, despite its simplicity, but its carbon cycle and temperature equations are miscalibrated, leading to the conclusion that one may want to be skeptical about predictions derived from DICE-2016. We examine the importance of the calibration for the social cost of carbon in the context of a partial equilibrium setting where interest rates are exogenous, as well as the simple general equilibrium setting from DICE-2016. We find that the model uncertainty from different consistent calibrations of the climate system can change the social cost of carbon by a factor of four if one assumes a quadratic damage function. When calibrated to the multi-model mean, our model predicts similar values for the social cost of carbon as the original DICE-2016, but with a strongly reduced sensitivity to the discount rate and about one degree less long-term warming.
The social cost of carbon in DICE-2016 is oversensitive to the discount rate, leading to extreme comparative statics responses to changes in preferences.
That is the abstract from Doris Folini, Aleksandra Friedl, Felix Kübler, and Simon Scheidegger,