Links & What I've Been Reading Q4 2020
Arpit Gupta on prediction markets vs 538 in the 2020 election: "Betting markets are pretty well calibrated—state markets that have an estimate of 50% are, in fact, tossups in the election. 538 is at least 20 points off—if 538 says that a state has a ~74% chance of going for Democrats, it really is a tossup." Also, In Defense of Polling: How I earned $50,000 on election night using polling data and some Python code. Here is a giant spreadsheet that scores 538/Economist vs markets. And here is a literal banana arguing against the very idea of polls.
Markets vs. polls as election predictors: An historical assessment (2012). Election prediction markets stretch back to the 19th century, and they used to be heavily traded and remarkably accurate despite the lack of any systematic polling information. Once polling was invented, volumes dropped and prediction markets lost their edge. Perhaps things are swinging in the other direction again?
Metaculus is organizing a "large-scale, comprehensive forecasting tournament dedicated to predicting advances in artificial intelligence" with $50k in prize money.
Philippe Lemoine critiques Flaxman et al.'s "Estimating the effects of non-pharmaceutical interventions on COVID-19 in Europe" in Nature.
However, as far as can tell, Flaxman et al. don’t say what the country-specific effect was for Sweden either in the paper or in the supplementary materials. This immediately triggered my bullshit detector, so I went and downloaded the code of their paper to take a closer look at the results and, lo and behold, my suspicion was confirmed. In this chart, I have plotted the country-specific effect of the last intervention in each country:
Alex Tabarrok has been beating the drum for delaying the second dose and vaccinating more people with a single dose instead.
Sound pollution decreased due to COVID-19, and "birds responded by producing higher performance songs at lower amplitudes, effectively maximizing communication distance and salience".
Scent dog identification of samples from COVID-19 patients: "The dogs were able to discriminate between samples of infected (positive) and non-infected (negative) individuals with average diagnostic sensitivity of 82.63% and specificity of 96.35%." [N=1012] Unfortunately they didn't try it on asymptomatic/pre-symptomatic cases.
And a massive update on everything Covid from Zvi Mowshowitz, much of it infuriating. Do approach with caution though, the auction argument in particular seems questionable.
Innovations and Innovation
Peer Rejection in Science: a collection of "key discoveries have been at some point rejected, mocked, or ignored by leading scientists and expert commissions."
Somewhat related: if GAI is close, why aren't large companies investing in it? NunoSempere comments with some interesting historical examples of breakthrough technologies that received very little investment or were believed to be impossible before they were realized.
Deepmind solves protein folding. And a couple of great blog posts by Mohammed AlQuraishi, where he talks about why AlphaFold is important, why this innovation didn't come from pharmaceutical companies or the academy, and more:
First, from 2018, AlphaFold @ CASP13: “What just happened?”:
I don’t think we would do ourselves a service by not recognizing that what just happened presents a serious indictment of academic science. [...] What is worse than academic groups getting scooped by DeepMind? The fact that the collective powers of Novartis, Pfizer, etc, with their hundreds of thousands (~million?) of employees, let an industrial lab that is a complete outsider to the field, with virtually no prior molecular sciences experience, come in and thoroughly beat them on a problem that is, quite frankly, of far greater importance to pharmaceuticals than it is to Alphabet.
Then, in 2020, AlphaFold2 @ CASP14: “It feels like one’s child has left home.”:
Once a solution is solved in any way, it becomes hard to justify solving it another way, especially from a publication standpoint.
"These improvements drop the turn-around time from days to twelve hours and the cost for whole genome sequencing (WGS) from about $1000 to $15, as well as increase data production by several orders of magnitude." If this is real (and keep in mind $15 is not the actual price end-users would pay) we can expect universal whole-genome sequencing, vast improvements in PGSs, and pervasive usage of genetics in medicine in the near future.
Extrapolating GPT-N performance: "Close-to-optimal performance on these benchmarks seems like it’s at least ~3 orders of magnitude compute away [...] Taking into account both software improvements and potential bottlenecks like data, I’d be inclined to update that downwards, maybe an order of magnitude or so (for a total cost of ~$10-100B). Given hardware improvements in the next 5-10 years, I would expect that to fall further to ~$1-10B."
Fund people, not projects I: The HHMI and the NIH Director's Pioneer Award. "Ultimately it's hard to disagree with Azoulay & Li (2020), we need a better science of science! The scientific method needs to examine the social practice of science as well, and this should involve funders doing more experiments to see what works. Rather than doing whatever is it that they are doing now, funders should introduce an element of explicit randomization into their process."
It will take more than a few high-profile innovations to end the great stagnation. "And if you sincerely believe that we are in a new era of progress, then argue for it rigorously! Show it in the data. Revisit the papers that were so convincing to you a year ago, and go refute them directly."
Why Are Some Bilingual People Dyslexic in English but Not Their Other Language? I'm not entirely sure about the explanations proposed in the article, but it's fascinating nonetheless.
The Centre for Applied Eschatology: "CAE is an interdisciplinary research center dedicated to practical solutions for existential or global catastrophe. We partner with government, private enterprise, and academia to leverage knowledge, resources, and diverse interests in creative fusion to bring enduring and universal transformation. We unite our age’s greatest expertise to accomplish history’s greatest task."
Labor share has been decreasing over the past decades, but without a corresponding increase in the capital share of income. Where does the money go? This paper suggests: housing costs. Home ownership as investment may have seemed like a great idea in the past, but now we're stuck in this terrible equilibrium where spiraling housing costs are causing huge problems but it would be political suicide to do anything about it. It's easy to say "LVT now!" but good luck formulating a real plan to make it reality.
1/4 of animals used in research are included in published papers. Someone told me this figure is surprisingly high. Unfortunately there's no data in the paper breaking down unpublished null results vs bad data/failed experiments/etc.
@Evolving_Moloch reviews Rutger Bregman's Humankind. "Bregman presents hunter-gatherer societies as being inherently peaceful, antiwar, equal, and feminist likely because these are commonly expressed social values among educated people in his own society today. This is not history but mythology."
@ArtirKel reviews Vinay Prasad's Malignant, with some comments on progress in cancer therapy and the design of clinical trials. "The whole system is permeated by industry-money, with the concomitant perverse incentives that generates."
@Cerebralab2 reviews Nick Lane's Power, Sex, Suicide: Mitochondria and the meaning of life. "The eukaryotic cell appeared much later (according to the mainstream view) and in the space of just a few hundred million years—a fraction of the time available to bacteria—gave rise to the great fountain of life we see all around us."
Is the great filter behind us? The Timing of Evolutionary Transitions Suggests Intelligent Life Is Rare. "Together with the dispersed timing of key evolutionary transitions and plausible priors, one can conclude that the expected transition times likely exceed the lifetime of Earth, perhaps by many orders of magnitude. In turn, this suggests that intelligent life is likely to be exceptionally rare." (Highly speculative, and there are some assumptions one might reasonably disagree with.)
How to Talk When a Machine is Listening: Corporate Disclosure in the Age of AI. "Companies [...] manage the sentiment and tone of their disclosures to induce algorithmic readers to draw favorable conclusions about the content."
On the Lambda School stats. "If their outcomes are actually good, why do they have to constantly lie?"
On the rationalism-to-trad pipeline. (Does such a pipeline actually exist?) "That "choice" as a guiding principle is suspect in itself. It's downstream from hundreds of factors that have nothing to do with reason. Anything from a leg cramp to an insult at work can alter the "rational" substrate significantly. Building civilizations on the quicksand of human whim is hubris defined."
There is a wikipedia article titled List of nicknames used by Donald Trump.
We Are What We Watch: Movie Plots Predict the Personalities of Those who “Like” Them. An amusing confirmation of stereotypes: low extraversion people are anime fanatics, low agreeableness people like Hannibal, and low openness people just have terrible taste (under a more benevolent regime they might perhaps be prohibited from consuming media).
A short film based on Blindsight. Won't make any sense if you haven't read the book, but it looks great.
The History of the Decline and Fall of the Roman Empire by Edward Gibbon. Fantastic. Consistently entertaining over almost 4k pages. Gibbon's style is perfect. I took it slow, reading it over 364 days...and I would gladly keep going for another year. Full review forthcoming.
The Adventures and Misadventures of Maqroll by Álvaro Mutis. A lovely collection of 7 picaresque novellas that revolve around Maqroll, a cosmopolitan vagabond of the seas. Stylistically rich and sumptuous. It's set in a kind of parallel maritime world, the world of traders and seamen and port city whores. Very melancholy, with doomed business ventures at the edge of civilization, doomed loves, doomed lives, and so on. While the reader takes vicarious pleasure in the "nomadic mania" of Maqroll, the underlying feeling is one of fundamental dissatisfaction with what life has to offer—ultimately the book is about our attempts to overcome it. Reminiscent of Conrad, but also Herzog's diaries plus a bit of Borges in the style.
Pandora’s Box: A History of the First World War by Jörn Leonhard. A comprehensive, single-volume history of WWI from a German author. It goes far beyond military history: besides the battles and armaments it covers geopolitics and diplomacy, national politics, economics, public opinion, morale. All fronts and combatants are explored, and all this squeezed into just 900 pages (some things are inevitably left out - for example no mention is made of Hoover's famine relief efforts). Its approach is rather abstract, so if you're looking for a visceral description of the trenches this isn't the book for you. The translation isn't great, and it can get a bit dry and repetitive, but overall it's a very impressive tome. n.b. the hardcover edition from HUP is astonishingly bad and started falling apart immediately. (Slightly longer review on goodreads.)
To Hold Up the Sky by Liu Cixin. A new short story collection. Not quite at the same level as the Three Body Trilogy, but there are some good pieces. I particularly enjoyed two stories about strange and destructive alien artists: Sea of Dreams (in which an alien steals all the water on earth for an orbital artwork), and Cloud of Poems (in which a poetry contest ultimately destroys the solar system, sort of a humanistic scifi take on The Library of Babel).
Creating Future People: The Ethics of Genetic Enhancement by Jonathan Anomaly. A concise work on the ethical dilemmas posed by genetic enhancement technology. It's written by a philosopher, but uses a lot of ideas from game theory and economics to work out the implications of genetic enhancement. Despite its short length, it goes into remarkable practical detail on things like how oxytocin affects behavior, the causes of global wealth inequality, and the potential of genetic editing to decrease the demand for plastic surgery. On the other hand, I did find it somewhat lacking (if not evasive) in its treatment of more general and abstract philosophical questions, such as: under what conditions is it acceptable to hurt people today in order to help future people?
The Life and Opinions of Tristram Shandy, Gentleman by Laurence Sterne. Famously the "first postmodern novel", this fictional biography from the 1760s is inventive, bawdy, and generally really strange and crazy. Heavily influenced by Don Quixote, parodies various famous writers of the time. Schopenhauer loved it and a young Karl Marx drew inspiration from it when writing Scorpion and Felix! I admire its ethos, and it's sometimes very funny. But ironic shitposting is still shitposting, and 700 pages of shitposting is a bit...pleonastic. At one point the narrator explains his digressions in the form of a line, one for each volume:
Illuminations by Walter Benjamin. Doesn't really live up to the hype. The essays on Proust and Baudelaire are fine, the hagiography of Brecht feels extremely silly in retrospect. The myopia of extremist 1930s politics prevents him from seeing very far.
Omensetter's Luck by William Gass. An experimental novel that does a great job of evoking 19th century rural America. Omensetter is a beguiling, larger-than-life figure, a kind of natural animal of a man. Nowhere near as good as The Tunnel and not much easier to read either. It's clearly an early piece, before Gass had fully developed his style.
The Silence: A Novel by Don DeLillo. Not so much a novel as a sketch of one. Not even undercooked, completely raw. It's about a sudden shutdown of all technology. Stylistically uninteresting compared to his other work. Here's a good review.
Little Science, Big Science by Derek John de Solla Price. Purports to be about the science of science, but really mostly an exploration of descriptive statistics over time - number of scientists, the distribution of their productivity and intelligence, distribution across countries, citations, and so on. Should have been a blog post. Nice charts, worth skimming just for them. (Very much out of print, but you can grab a pdf from the internet archive or libgen).
Experiment and the Making of Meaning: Human Agency in Scientific Observation and Experiment by David Gooding. Skimmed it. Written in 1990 but feels very outdated, nobody cares about observation sentences any more and they didn't in 1990 either. Some interesting points about the importance of experiment (as opposed to theory) in scientific progress. On the other hand all the fluffy stuff about "meaning" left me completely cold.
The Subjective Side of Science: A Philosophical Inquiry into the Psychology of the Apollo Moon Scientists by Ian Mitroff. Based on a series of structured interviews with geologists working on the Apollo project. Remarkably raw. Mitroff argues in favor of the subjective side, the biased side, of how scientists actually perform science in the real world. On personality types, relations between scientists, etc. There are some silly parts, like an attempt to tie Jungian psychology with the psychological clusters in science, and a very strange typology of scientific approaches toward the end, but overall it's above-average for the genre.
On Writing: A Memoir of the Craft by Stephen King. A pleasant autobiography combined with some tips on writing. The two parts don't really fit together very well. This was my first King book, I imagine it's much better if you're a fan (he talks about his own novels quite a lot).
The Lord Chandos Letter And Other Writings by Hugo von Hofmannsthal. A collection of short stories plus the titular essay. If David Lynch had been a symbolist writer, these are the kinds of stories he would have produced. Vague, mystical, dreamlike, impressionistic. I found them unsatisfying, and they never captured my interest enough to try to disentangle the symbols and allegories. The final essay about the limitations of language is worth reading, however.
Ghost Soldiers: The Forgotten Epic Story of World War II's Most Dramatic Mission by Hampton Sides. An account of a daring mission to rescue POWs held by the Japanese in the Philippines. The mission itself is fascinating but fairly short, and the book is padded with a lot of background info that is nowhere near as interesting (though it does set the scene). Parts of it are brutal and revolting beyond belief.
We Are Legion (We Are Bob) by Dennis Taylor. An interesting premise: a story told from the perspective of a sentient von Neumann probe. That premise is sort-of squandered by a juvenile approach filled to the brim with plot holes, and an inconclusive story arc: it's just setting up the sequels. Still, it's pretty entertaining. If you want a goofy space opera audiobook to listen to while doing other stuff, I'd recommend it.
Rocket Men: The Daring Odyssey of Apollo 8 and the Astronauts Who Made Man's First Journey to the Moon by Robert Kurson. Focused on the personalities, personal lives, and families of the three astronauts on Apollo 8: Frank Borman, William Anders, and James Lovell, set against the tumultuous political situation of late 1960s America. Written in a cinematic style, there's little on the technical/organizational aspects of Apollo 8. Its treatment of the cold war is rather naïve. The book achieves its goals, but I was looking for something different. Ray Porter (who I usually like) screws up the narration of the audiobook with an over-emotive approach, often emphasizing the wrong words. Really strange.