Thoughts in Between

by Matt Clifford

TiB 114: WTF happened in 1971; the power of machine learning in science; the court case that could destroy the EU; and more...

Welcome new readers! Forwarded this email? Subscribe here. Enjoy this email? Forward it to a friend.

What went wrong in 1971?

Something went badly wrong in 1971 - or at least that’s the strong impression created by a fascinating collection of charts curated at the new and aptly named On a range of indicators - inequality, median income, debt, incarceration - things seem to have taken a turn for the worse starting in the early 1970s.

There are some interesting attempted explanations from across the ideological spectrum in the replies to this tweet, which range from Vietnam and counterculture, the environmental movement, the Nuclear Non-Proliferation Treaty and the decline of science. Lots of answers from the libertarian right point to the end of the Bretton Woods system and the rise of fiat money (but see Adam Tooze’s argument, in his excellent interview with Tyler Cowen, that “no such system ever existed”).

One of the most interesting arguments from the left is Chris Dillow’s view that until the 1970s the interests of capital and labour were better aligned - "General Motors needed a large well-paid working class; Goldman Sachs, not so much” - and that capital won the ensuing power struggle. The puzzle I always return to, though, is Scott Alexander's claim - discussed last year - that the real turning point was a decade earlier, around 1960 - “the year the Singularity was cancelled” and global hyperbolic economic growth started to falter. Whatever the answer, it's an important question to ponder. 

Machine learning-driven science, revisited

Eric Schmidt, the former CEO of Google, has a new paper, co-authored with Maithra Raghu, on the use of deep learning for scientific discovery. It’s aimed at people who are relatively new to machine learning and is an excellent overview of the state of the field. 

It reminded me of one of my favourite ever papers (which we discussed last year), which uses Natural Language Processing (NLP) techniques to "read" a huge corpus of materials science journals and discover new compounds. That paper is less than a year old, but Google Scholar suggests it’s already been cited over 90 times - a signal of the extraordinary pace of machine learning as a field. The papers that cite it suggest promising results from applying machine learning to mathematics, physics and, inevitably, coronavirus - as well as impressive new findings in materials science.

We’ve talked several times about the causes and (very bad) consequences of the apparent slowdown in scientific progress. There are, crudely, two kinds of explanations: those that point to bad incentives in our institutions and those that suggest that important new ideas are just getting harder to find. If you’re in the latter camp, machine learning techniques that give us radically better discovery tools seem like very good news. 

Could an obscure court case destroy the EU?

Few topics are less glamorous than German constitutional law, but sometimes world historical events have obscure triggers. I wonder if last week’s German Constitutional Court ruling on the European Central Bank (ECB)’s bond buying programme may be one of them. The German court ruled that the ECB is acting outside its jurisdiction by pursuing massive quantitative easing (another previously obscure but now important topic). 

This matters because it puts the EU on a potentially fatal collision course with Germany, its largest and richest member. The EU legal system is built on a fudge, as this superb thread explains: all EU members agree that the European Court of Justice is superior to national courts on matters of EU law - but who gets to decide whether a given area is a matter of EU law is much murkier. Here the German court is claiming this authority - but if all member states did this (and the more autocratic ones are increasingly tempted), the whole EU project would collapse.

I’ve argued before that this may be a century of secessions: the models of political economy that gave today’s polities coherence are being undermined by technology and globalisation. That was the core argument in this case: what’s good for Germany may not be good for the EU and vice versa. The EU was quick to smack the ruling down - and the most likely outcome is another legal fudge. But fudges can hold only when no one has the incentive to probe them too carefully - and this court case suggests that era is long gone.

Quick links

  1. You are what you tax. Thread on different countries' corporate tax codes that somehow manages to be funny (actually) and insightful (Thanks Jonny for the link)
  2. The unreasonable effectiveness of machine translation. Fascinating study that suggests automated translation increased exports on eBay by >10%(!)
  3. Buy earplugs. A 10dB noise increase sees productivity fall 10%.
  4. 9th century word processor substitutes. What monks did because they lacked Microsoft Word (Semi-related: Renaissance wrapping paper)
  5. Coronavirus round up. Sobering graphic showing years of life lost to coronavirus. Coronavirus sends hourly wages soaring... because so many low paid jobs have been destroyed. Amazing paper on how coronavirus "borrows" computational power from our cells. Quantifying the risk of day-to-day activity.

Your feedback

Thanks for reading; if you enjoy TiB, please forward it to a friend or, even better, share it on Twitter / LinkedIn. If it's getting stuck in your email's Promotions tab, please help train the algorithm by moving it to Primary :)

I'm always happy to hear from readers. You can find me on Twitter or just hit reply.

Until next week,

Matt Clifford