TiB 103: Can machine learning save Communism; Coronavirus and authoritarianism; how to save science; and more...

Forwarded this email? Subscribe here. Enjoy this email? Forward it to a friend.

Coronavirus and the pitfalls of authoritarianism

Zeynep Tufekci has an interesting new piece that asks whether censorship and surveillance caused Xi Jinping to underestimate the seriousness of coronavirus. This is potentially an important rejoinder to the argument that AI, by enabling a “panopticon” state, is a pro-authoritarian technology (see previous discussion and this Henry Farrell piece).

Tufekci argues that one of the second order effects of a surveillance society is that people act strategically to minimise the chance of punishment - and so their behaviour (and the data it generates) ceases to be a good guide for policy. If there’s a strong incentive to suppress bad news, you won’t hear bad news, even when you need to (There are echoes of the problem of central planning, discussed below).

There’s an interesting connection here to Amartya Sen’s (contested) argument that famines don’t happen in democracies. According to Sen, famines aren’t about resources or technology, but are political failures of planning and responsiveness. China has demonstrated extraordinary powers in response to the virus, but perhaps its initial spread is a fascinating and tragic example of the costs of state over-capacity.

Will machine learning rescue central planning? (No)

One of the most important economic debates of the 20th century was the socialist calculation debate: are markets necessary or can a sufficiently smart central planner allocate resources efficiently? Markets won the century, broadly, but astronomical advances in compute power and datasets have led some to ask whether planning was merely technologically premature (e.g. “Digital socialism”).

A new paper by Jesus Fernandez-Villaverde re-examines whether machine learning can replace markets - and concludes "no". The argument is a variant of Hayek’s classic The Use of Knowledge in Society: the problem is not computational, but behavioural; under planning, people don’t have the incentive or even capability to share all their relevant information.

This isn’t just a critique from the right. One of my favourite essays is a magnificent 2012 piece by (left-leaning) Cosma Shalizi with the brilliant title “In Soviet Union, Optimization Problem Solves You”, which examines the computational problem of planning (The essay itself is riffing on the superb fictionalised account of Soviet planning, Red Plenty, which is worth a read). Machine learning will do many things, but fixing communism isn’t one of them.

How to save science

This week two excellent new pieces on causes of - and possible cures - for the slowdown in scientific progress (previous coverage) were published. Both provide cause for optimism that we’ve not simply run out of good ideas.

The first by Alex Danco, looks at the problem of “positional scarcity” in science: brilliant people are forced to work years as postdocs to win the credential of publication in artificially space-constrained top journals. The second by Jay Bhattacharya and Mikko Packalen argues that the modern journal system discourages exactly the kind of “playful” exploratory research that is crucial to future scientific breakthroughs (which in turn reminds me of this discussion of "far analogies"). It’s perhaps the most interesting paper I’ve read this year; do read the whole thing.

Interestingly, both pieces see the challenge as at least in part a talent allocation problem. For Danco, the fact that postdocs waste years of their lives playing a positional game discourages talented people from academia. And for Bhattacharya and Packalen, academia’s preference for incremental over exploratory research puts off risk seekers (especially as science really does advance one funeral at a time). It’s exciting - if daunting - that these problems seem solvable; as I’ve said before, the world needs more experiments in institutional design.

Quick links

If you want a revolution, how come you're so content? Interesting data on life satisfaction.
Playing at inflation. More than you could ever want to know about the price of Lego.
Adventures in paleo-futurism. Ten past inventions of the "future" that we're still waiting for.
The AI ate my homework. Facebook AI breakthrough in using machine learning to solve equations.
Coronavirus is eating the world. The differential impact of the outbreak on Tencent and Alibaba.

Your feedback

Thanks for reading TiB. I'd love it if you would forward it to a friend or share it on social media.

It's always great to hear from you: feel free to get in touch on Twitter or by hitting reply.

Until next week,

Matt Clifford