Forwarded this email? Subscribe here. If you like it, please forward it to a friend.

One important hypothesis that we’ve discussed before is that authoritarian regimes may have a structural advantage in developing technologies like AI - and that in turn these technologies may entrench such regimes in power. Henry Farrell has an excellent new essay (with a brilliant title) that challenges that assumption (Thanks Sam Watts for sharing).

Farrell (see previous coverage) argues that states’ desire for legibility has often reinforced planning errors, with disastrous effects - and machine learning models may exacerbate this. Such models may magnify biases in training data and lead authoritarian regimes to take actions that destabilise themselves. This is similar to the point discussed last week about the risks authoritarians face when they break feedback channels that show what’s actually going on (But do also note the interesting point in the comments that even when it doesn’t work, technology can reinforce authoritarianism!)

It’s worth comparing this argument with this fascinating paper by Bryan Caplan that looks at the impact of technology on totalitarianism. Caplan argues that bad decisions don’t undermine totalitarian states, bad - i.e. insufficiently totalitarian! - successors do (e.g. Khrushchev after Stalin). So the technology we should fear is not AI, but life extension for dictators or genetic engineering to select for more orthodox successors… 

One of the winners of this year’s Nobel Memorial Prize in Economics was Michael Kremer, who is best known for his work on randomised control trials in development studies (see write up here). One of his most intriguing ideas is what he called (back in 1993) the “O-Ring theory of economic development“ (named after the famous O-Rings that Feynman showed played such a catastrophic role in the Challenger disaster).

The core idea is that in many processes that create economic value, the weakest link is very important: e.g. a company with a great product but a terrible sales function will fail. One consequence of this model is “assortative matching”: great people coalesce in the same firms (so that lower performers don’t “break the chain”) and so are much more productive and earn higher wages than they would elsewhere. There are good explanations here and here that discuss the implications.

It strikes me that startup ecosystems are a lot like this. If a city has an active seed scene, but poor Series A coverage or no scaling talent, its startups will fail regardless of their quality - so great startups cluster. This helps explain both why valuations are so high in Silicon Valley and so low everywhere else. Perhaps this is just me talking my own book, but if that’s right, it feels like the value of startup-focused institutions that enable founders to transcend ecosystem boundaries may be underrated. 

This week six European countries announced they were joining Instex, the special purpose vehicle set up - against US wishes - to permit (limited) trade with Iran following the Trump Administration’s withdrawal from the Iran nuclear deal. 

It’s an example of two important trends that remain under-discussed: one, growing calls for “European sovereignty” across a range of policy areas (see previous coverage of Macron on technological sovereignty); and, two, the difficulty of extraterritorial regulation in a multipolar world. Steve Randy Waldman wrote a prescient short essay on the topic last year, which is worth revisiting. The key idea is that US withdrawal from the Iran deal gave its European partners both cover and strong incentives to push back on the US’s enforcement of financial regulation outside its borders.

We now have at least three global powers that are competing (with mixed success) to exercise extraterritorial control across different domains. The US wants, as Waldman notes, to enforce its rules wherever dollars are involved; the EU, as it showed with GDPR, wants to be the global tech regulator of first resort; and China wants to project its values wherever it does business (and Singapore wants to enforce fake news flags on social media!). This theme is just taking off; it’s going to vastly complicate national regulatory debates, especially in tech, in the coming years.  

It’s great to welcome so many new readers this week - thanks for subscribing. And long time readers, thanks for sticking with it! If you enjoy TiB, I’d love you to share it with a friend. The easiest way is just to forward it.

And, of course, feel free to hit reply to say what you liked, say hello or follow me on Twitter.

Until next week,

Matt