Links & What I've Been Reading Q3 2021
1. Was the Industrial Revolution The Industrial Revolution? A fascinating look at the industrial revolution in the UK, including some explanations of slow/zero growth in various periods before WWII.
From 1760 to 1800, the contribution of the steam engine was .004 percent per year to capital deepening and .005 percent to TFP growth. Not until after 1850 had the high-pressure engine become widespread and efficient enough to be deployed in factories and on rail engines that these numbers each rose to .2 percent. A century passed between James Watt’s patent—the first revolutionary “general purpose technology”—and its maximum realization in TFP growth.
Soaring population growth in the late eighteenth and early nineteenth centuries threatened the island with a Malthusian demographic catastrophe. [...] Without an Industrial Revolution, Mokyr reasons that GDP per capita in Britain could have been twenty percent lower in 1830 than in 1760.
Britain became modern, and then it got rich.
2. Leo Aschenbrenner on his favorite Chad Jones papers. "Most of all, Chad’s papers showed me what beautiful economic theory looks like. Simple models that capture a few essential forces, guided by broad empirical trends. These can often reveal insights that totally non-obvious ex ante—but are strikingly intuitive and powerful once found."
3. The Ant and the Grasshopper: Seasonality and the Invention of Agriculture A fascinating (and speculative) paper from 2019 which argues that agriculture was invented because changes in the earth's orbit caused an increase in seasonality!
4. Evidence of Fraud in an Influential Field Experiment About Dishonesty. Fairly brazen data fabrication, though it's still not clear whether it was Ariely or the company that was in charge of collecting the data.
5. Some evidence suggesting that the Sputnik vaccine paper used fake data. I'd note that real-world data shows the vaccine working pretty well regardless of whether there was fraud in the trial.
6. Predicting and reasoning about replicability using structured groups: predicting replicability using the IDEA protocol (‘Investigate’, ‘Discuss’, ‘Estimate’ and ‘Aggregate’) for generating and combining predictions seems to work very well, achieving 84% classification accuracy in this sample. Still waiting on the SCORE results.
successful replications led to an increase in yearly citations of around 5% and that unsuccessful replications led to a decrease in yearly citations of around 4%. For the average article in my sample, which has roughly eight citations per year, this would imply a change of ±1 citation every 2 to 3 years.
As I was saying, replications don't really matter, so it's better to go for forward-looking reforms instead of trying to fix the past.
8. A survey on questionable research practices s from the Netherlands. ~4% fabrication, ~50% frequently engage in QRPs.
9. A clever paper uses the shutdown of a journal (due to an "exogenous shock" in economese) to measure the prevalence of strategic citations. Citations drop by about 20% after discontinuation.
10. Is the Value Premium Smaller Than We Thought? A look at the various decisions that go into constructing a risk factor, and how they affect the end result. "The results suggest that the original value premium estimate is upward biased because of a chance result in the original research decisions."
11. Text-generating models are sometimes used to plagiarize papers by back-and-forth translation, or to generate new (nonsensical) papers. This study looks for "tortured phrases" like "profound neural organization" (ie deep neural network) and "haze figuring" (ie cloud computing), and finds many published papers that appear to have been computer-generated.
12. Arcadia Science is a for-profit research institute, with a biology lab opening in Berkeley next month. "No work produced or funded by Arcadia will be published in journals."
13. Simpson's paradox and Israeli vaccine data. On stratification by age and calculating vaccine effectiveness.
14. Tamiflu for covid? Looks pretty good, ~50% decrease in risk of hospitalization. Costs $700 though.
15. Alignment Problems With Current Forecasting Platforms. A look at some issues with GJO/CSET/Metaculus. It's not easy to incentivize people to provide their true forecasts at all times, share information, etc.
16. Facebook's new forecasting platform lasted about a month.
17. Hedgehog, blockchain prediction market from "Futarchy Research Limited".
18. Razib Khan has a relatively positive review of Harden's The Genetic Lottery, but the Steve Sailer review is a lot more entertaining. It's amusing that the BBEG for these people is still Charles Murray rather than, say, David Reich who has said much worse things.
19. George Church is bringing back the woolly mammoth.
20. ADS: Become a Billionaire.
Surveying the top Y Combinator companies, I find that around the top 50 are valued at over $1,000,000,000. They won’t all exit successfully, and the founders won’t all own enough equity to emerge with tres commas to their net worth, but this already gets us to a much more practical and optimistic heuristic to life:
- Try very hard to get into YC
- Conditional on acceptance, try very hard to become a billionaire
The odds really aren't that bad. Also from ADS, Does Moral Philosophy Drive Moral Progress?
22. Felix Stocker on Will MacAskill's longtermist plans: Reflecting on the Long Reflection. "I'm struggling to see the Long Reflection as anything other than impossible and pointless: impossible in that we cannot solve all x-risks before any s-risks, or avoid race dynamics; pointless in that I don't believe that there is a great Answer for it to discover."
23. Alexey Guzey on Bloom et al's Are Ideas Getting Harder to Find? The paper has a bunch of problems, but the more general section on TFP is the most interesting:
France’s TFP in 2001 was higher than in 2019. Italy’s TFP in 1970 was higher than in 2019. Japan’s TFP in 1990 was higher than in 2009. Spain’s TFP in 1984 was higher than in 2019. Sweden’s TFP in 1973 was higher than in 1993. Switzerland’s TFP in 1974 was higher than in 1996. United Kingdom’s TFP in 2003 was higher than in 2019.
24. ACX on the FDA: Adumbrations Of Aducanumab The Moldbuggian aspects of this are still underappreaciated. Bureaucracy and bureaucrats are isolated from the consequences of their actions; the idea of equality before the law is a complete joke in the modern regulatory state, and the incentive vectors point in exactly the wrong direction. Scott ultimately blames it on the incentives of the politicians—the people seem to accept infinite costs to prevent certain bad things from happening; but if we take the people as a given, isn't ultimately the system of governance at fault? Plus ACX on missing school: Kids Can Recover From Missing Even Quite A Lot Of School.
25. Herding, Warfare, and a Culture of Honor: Global Evidence. "The culture of pre-industrial societies that relied on animal herding emphasizes violence, punishment, and revenge-taking". Highly speculative (the approach of extracting culture of honor from folklore seems doubtful for various reasons) and those scatter plots are not entirely convincing, but also intuitively appealing.
26. Exploiting an exogenous shock in birth control prices, The Children of the Missed Pill looks at the causal impact of the pill: "As children reached school age, we find lower school enrollment rates and higher participation in special education programs." The eugenic effect of abortion/contraception is both underrated and understudied.
27. A primer on olivine weathering as a cheap method of carbon capture; looks like it could sequester a tonne of CO2 for less than $20. Geoengineering is very cheap compared to most proposed "green" solutions. The OECD has 120 euros per tonne as its "central estimate" of carbon costs in 2030, implying an extremely high ROI for geongineering.
28. DeepMind: Generally capable agents emerge from open-ended play. "We find the agent exhibits general, heuristic behaviours such as experimentation, behaviours that are widely applicable to many tasks rather than specialised to an individual task. This new approach marks an important step toward creating more general agents with the flexibility to adapt rapidly within constantly changing environments."
29. Unintentionally hilarious paper about AI spotting race in chest x-rays: "Our findings that AI can trivially predict self-reported race - even from corrupted, cropped, and noised medical images - in a setting where clinical experts cannot, creates an enormous risk for all model deployments in medical imaging: if an AI model secretly used its knowledge of self-reported race to misclassify all Black patients, radiologists would not be able to tell using the same data the model has access to."
30. "Pain Reprocessing Therapy" "centered on changing patients’ beliefs about the causes and threat value of pain" more effective than usual care for back pain, at least if you think you can trust people's responses in surveys.
31. Yet another piece of evidence against the efficacy of advertising: TV Advertising Effectiveness and Profitability: Generalizable Results From 288 Brands. "...negative ROIs at the margin for more than 80% of brands, implying over-investment in advertising by most firms. Further, the overall ROI of the observed advertising schedule is only positive for one third of all brands."
35. An animated explainer of Robin Hanson's grabby aliens model: Humanity was born way ahead of its time. The reason is grabby aliens.
36. Did you know that Milla Jovovich released an album in 1994 and it's...not bad at all? Sounds like Kate Bush and Peter Gabriel. Check out Clocks. [NSFW cover art]
37. Plus some great krautrock: Et Cetera - Kabul.
The History of the Pelopponesian War, by Thucydides. Re-read. What was that Coleridge quip? "All men are born Herodotians or Thucydideans"? Something like that. Anyway, I was definitely born a Herodotian. Thucydides is a historian with the soul of an accountant. Still, there are things to appreciate in that attitude: while most ancient historians never saw an army smaller than 400,000, he's happy to tell you about engagements with 60 hoplites and 20 archers. And keeping track of a myriad engagements, covering Asia Minor, Greece, and Italy, over the span of multiple decades is extremely impressive.
How prescriptive is Thuc's realpolitik? I'm not entirely sure, it certainly didn't do the Athenians any good. He's obviously a skeptic when it comes to the supernatural, and there's very little room for morality in his history; is this an artifact of the lack of morality in the way the Athenian went about their affairs, or is this something Thuc projects onto them? One interesting point is that his story draws on the structure of tragedy: the hubris of the Sicilian expedition is ultimately punished; the players seem to lack any ability to change course. Perhaps morality plays no role in this history because Thuc views the path taken by each polis as deterministic. (This applies both to the "Thucydides trap" specifically, and also more generally).
On the question of direct democracy as a system of government things are a bit clearer as Thuc doesn't hide his views. He's short on alternatives though; the traditional polis obviously can't cope with the environment of the 4th century, but Thuc can't really see beyond it.
There are apparently some people who think Thucydides influenced the Neoconservatives, and I find that utterly absurd. Thuc is extraordinarily cynical when it comes to "spreading freedom"-style justifications for war, and if there's any realpolitik involved in spending trillions so that Afghan women can get gender studies degrees for 20 years before the Taliban come back, I'm not seeing it.
One of the things that stand out is how bad the Greeks are at war. Reading Thuc, you're constantly thinking "well of course these guys got rolled by the Romans". How did they beat the Persians so hard? Sieges seem to be a sticking point (something Phillip II turned out to be quite good at), so perhaps the open battles against the Persians played into their hands, or perhaps it was simply a matter of mismatched unit compositions. On the other hand the Athenians were extraordinarily persistent; even after the plague and the Sicilian disaster they still kept going for years, possibly only losing due to the Persian money flowing into the Spartan coffers.
If you haven't read any histories of the Pelopponesian war, this is highly recommended, just keep in mind it's very unfinished. Get the Landmark edition.
The Swerve: How the World Became Modern, by Stephen Greenblatt. This book has an incredibly ambitious thesis: it argues that the world became modern due to the rediscovery of Lucretius' De Rerum Natura. Unfortunately the evidence presented in favor of that thesis is pretty weak, and the book suddenly ends right as it starts to get into a groove. Still, it's fairly entertaining and has a ton of interesting anecdotes from the life of Poggio Bracciolini (the man who rediscovered Lucretius).
The Origin of Species, by Charles Darwin. A fairly dry read, its value today mainly lies in its documentation of the discovery of evolution, and in showing how Darwin could reason his way forward despite rather limited means (not even an inkling of DNA!). It was cool to see the role geology played in the development of evolutionary theory, and there's a very interesting passage (at the end of the chapter ON THE IMPERFECTION OF THE GEOLOGICAL RECORD) in which Darwin almost invents plate tectonics based on the geographical distribution of species. It's difficult to recommend: if you want to learn about evolution, pick up a modern textbook; if you're interested in the history of science you should probably read a historian; and if you just want to read something cool by Charles Darwin, pick up his Beagle adventure.
A Fire Upon the Deep, by Vernor Vinge. Some pretty cool worldbuilding, with a universe divided into zones where different levels of technology are possible (the highest one is filled with Gods who quickly commit --suicide). One of the main alien races is a sentient houseplant riding a roomba (seriously). But half the novel is wasted on a dull isekai story about some annoying kids stuck on a backwards planet with telepathic wolves, making the thing way overlong. And the resolution is not entirely satisfying.
Inhibitor Phase, by Alastair Reynolds. A new novel in the Revelation Space universe, unfortunately it's also the worst novel in the Revelation Space universe. It's a bit like a horror theme park, going from one ride to the next with little to no connective tissue between them. Even worse, many of the rides are completely nonsensical given the setting (humanity has almost been completely wiped out by the inhibitors). The two main characters are completely uninteresting, their dialogue is annoying, and the revelations of their backstory are completely predictable.
Crash, by J. G. Ballard. Holy mother of Christ, this is an experience. A blunt tool that beats you into submission through drone-like repetitiveness. Truly a novel that lives up to its reputation (one publisher's reader wrote: "This author is beyond psychiatric help. Do Not Publish!"). What images! A peerless examination of the intersection between sex and technology. The Cronenberg film gets the imagery right, but the languid, whispered tempo is completely wrong. Kermode, in a very positive review, described it as "glacial"! I feel the novel required a more in-your-face treatment.
He dreamed of ambassadorial limousines crashing into jack-knifing butane tankers, of taxis filled with celebrating children colliding head-on below the bright display windows of deserted supermarkets. He dreamed of alienated brothers and sisters, by chance meeting each other on collision courses on the access roads of petrochemical plants, their unconscious incest made explicit in this colliding metal, in the heamorrhages of their brain tissue flowering beneath the aluminized compression chambers and reactions vessels.
Lord of the Flies, by William Golding. Somehow managed to evade this as a kid. It's compelling and effective but I can't get on board with its overwhelming cynicism.
Don't Make Me Think, by Zero HP Lovecraft. The emoji gimmick doesn't work, but I loved the world-building.
Flashman and the Dragon, by George MacDonald Fraser. Flashman's in China this time, right in the middle of the Taiping Rebellion. This is the 8th book in the series, and things are starting to get repetitive, but the humor, deep historical research, and memorable characters manage to overcome the familiar plotline.
The Shadow of the Wind. by Carlos Ruiz Zafón. Bad audiobook of a bad airport novel filled with interminable exposition dumps in an awful style. Dropped it halfway through.