Matt's Thoughts In Between - Issue #8

AI, hardware and geopolitics

I've written before about how AI is increasingly a national security issue. This new paper by Tim Hwang of MIT looks at the role of computational hardware in this, which seems a neglected angle. Hwang makes the point that control of the supply chain of computational power is the most likely place for international rivalry around machine learning to manifest itself.

Although AI appears on the surface to be the ultimate "placeless" phenomenon, the physical location of assets could come to matter if geopolitical conflict around AI boils over. Hwang notes that while the vast majority of the important GPU (the chip used for most machine learning) design takes place in the US, the opposite is true for actual manufacture: the vast majority takes place in China and Taiwan. (In this context, it's also worth reading Jeffrey Ding's new report about Chinese AI, which has a good section on hardware)

Where does Europe fit into this? Nowhere, if you ask Bruno Macaes, who has a scathing article in Politico on Europe's "AI delusion". Though Emanuel Macron, whose launch of France's AI strategy this week received positive reviews, may beg to differ.

Searching for young Einsteins (and Musks)

A topic I spend a lot of time thinking about at Entrepreneur First is how to identify extraordinary talent early. This is core to the EF model, so it was interesting to see the Economist look at the issue this week. They conclude that not only is it possible to identify exceptional talent at an early age, but that differences in ability among the most able 13 year-olds (i.e. those at the 99.75th percentile vs those at the 99th percentile) seem to predict substantial differences in outcomes as adults. That is, there is no "tailing off" effect as you move up the ability distribution, at least in mathematics.

The mechanism by which talent translates into achievement is complex and controversial (I wrote a bit about this a few weeks ago), but the best model I've seen recently comes from this paper on scientific achievement (which I came across via this excellent discussion of the same topic). It argues that achievement in science is a function of (a) choosing a project that turns out to be high impact, which is essentially random, and (b) having the talent to make the most of the opportunity if it turns out to be a very high impact one.

For what it's worth, this is also my (and perhaps Marc Andreessen's) mental model of entrepreneurial success. The data is certainly harder to collect, though I hope this is one area where EF will be able to make a big impact over time... (This is a topic I'm endlessly fascinated by, so if you know any good reading on the subject, please hit reply)

Jordan Peterson and secular religion

Last year I read a lot about critiques of and challenges to liberalism. One of the major themes was the void left by the decline of religion in the West, which is something I've been thinking about a lot. I've also been casually following the Jordan Peterson phenomenon over the last year, but frankly haven't taken it very seriously, as his work seemed to me a mixture of the nonsensical, the non-falsifiable and the objectionable.

So I was intrigued this week to read two excellent essays by people I admire (though rarely fully agree with) - Scott Alexander and Russ Roberts - that not only praise Peterson but point to his project as a potential plug for the religion-shaped hole in Western society, even if they don't sign up to it themselves. This is a topic I need to think about much more before I offer any firm conclusions. These pieces do, though, push me to update my beliefs in the direction that there is an enormous market for quasi-religious experience and that we'll see a lot of "innovation" in this space in the next decade. Last week we had, "Don't start a startup, start a country"; perhaps "start a religion" is next...

Quick links

Silk roads: Beautiful map of the trade routes of the medieval world, via @medievalhistory
Philosophical expression: Wittgenstein invented the emoji. Though his bedside manner left something to be desired.
Artists manqués: Why do so many scientists want to be filmmakers?
Electric sheep: Training machine learning models though "dreams" (fun interactive model). And a critique.
The crisis ten years on: Stunning charts telling the story of the global financial crisis - and perhaps pointing to the next one

Your feedback

If you enjoyed this, please share it, either by sharing the subscribe link on social media or by forwarding to the two people you think would most enjoy it. I'm always pleased to get your thoughts and suggestions - just hit reply.

Until next week,

Matt