TiB 187: The State of AI; extreme talent; monasteries and crypto; and more...

Welcome new readers! Thoughts in Between is a newsletter (mainly) about how technology is changing politics, culture and society and how we might do it better.

It goes out every Tuesday to thousands of entrepreneurs, investors, policy makers, politicians and others. It’s free.

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

What's the state of AI?

Nathan Benaich and Ian Hogarth published their annual State of AI report last week and, as in previous years, it's a must read. If you want to read one document to get a comprehensive overview of what's cutting edge in AI - across the technology itself, its industrial applications and its real-world consequences - this is it. It's 188 slides, so block out some real time to digest.

The report includes excellent deep dives on a number of favourite TiB topics. I particularly recommend the sections on talent, which looks at the rise of Chinese talent and the challenge of academic brain drain into industry, and on politics, which explores the present-day military applications of AI and the growing attention paid to AI safety (though, in absolute terms, this remains tiny: the report shows that there are just 100 AI safety researchers across seven leading AI research organisations). Do also check out the section on EleutherAI, a startup that's building open source alternatives to GPT-3.

One great feature of State of AI since the first edition in 2018 has been Nathan and Ian's willingness to make public predictions about what will happen in AI over the next 12 months - and then to evaluate them a year later. Their 2020 predictions look pretty good (five and a half out of eight), so it's worth paying attention to their outlook for next year. They predict a big year for Anthropic (see TiB 167) and for ASML (see TiB 174) and a breakthrough in the physical sciences from DeepMind, following its AlphaFold success last year.

How to evaluate extreme talent

We’ve talked a lot before about the value of extreme talent, and some of the challenges involved in discovering it (see e.g. TiB 125 and 130). Rohit Krishnan, whose essay “On Medici and Thiel” we looked at in TiB 176, has a new piece on this topic, which explores why it’s so difficult to design selection processes for outliers. Krishnan outlines the core problem in his opening paragraph:

In a world where selection is hard, we resort to ever more stringent measurement. If measurement is too strict, we lose out on variance. If we lose out on variance, we miss out on what actually impacts outcomes

I have a lot of sympathy for this, as it’s a challenge we face at Entrepreneur First (EF). We learned the hard way (fortunately a long time ago) that overly prescriptive selection criteria produce great cohorts of founders on paper, but seldom yield the outliers that our business model requires. Krishnan’s piece is full of interesting insights on why this is the case. He cites as one challenge Berkson’s Paradox, which is that two traits can be positively correlated in the population as a whole, but negatively correlated within an extreme subset; Krishnan’s example is talent and attractiveness: positively correlated in the population; negatively among celebrities.*

Krishnan’s advice - which I endorse - is to do less selection up front and “rely on [observing] actual performance to select for the best”. A key part of the puzzle, though, is whether organisation’s cost / benefit function with respect to talent permits this. For EF it's inexpensive to give a prospective founder a chance, both relative to the capital we’ll put behind them if they succeed (the initial cost is around 1% of the lifetime total) and, especially, to the potential upside of discovering that they’re extraordinary. I suspect people pay too much attention to selection at the margin and not enough to reducing the cost of taking more bets.

* It’s tempting to think that the relationship between intelligence and success might be like this, but it doesn’t seem to be true

Monasteries, assets and economic revolutions

A favourite if infrequent TiB theme is the very long run impact of historical events (see e.g. TiB 80, 99, 152). This week I stumbled upon this paper that looks at the role of the dissolution of the monasteries in the English Reformation in laying the ground for the industrial revolution (see tweet-length summary here). The core argument is that land seized from monastic endowments and resold was free of feudal encumbrances and so enabled a new class of gentry innovators who were able to innovate and invest in their new assets. The authors show that formerly monastic land was much more likely to see early investment in water mills, for example.

I find this sort of thing fascinating in its own right, but I wonder if there is a more general point here. Perhaps all economic revolutions require an "opening up" of new assets which innovators can experiment with freely. Whenever factors of production become too encumbered by the constraints of the legacy economy, a step change in productivity requires new spaces; the existing ones are too weighed down in what in a software context we'd call technical debt to allow for anything other than incremental change.

This, of course, was and is the great promise of the internet: the opening up of new spaces for innovation, unencumbered by (some / much of ) the legacy economy. The frontier nature (see TiB 179) of these spaces allows for the emergence of a new socio-economically successful class with different values and worldviews. As the gentry were to the 16th century, so are internet entrepreneurs to the 21st. Some variant of this is my bull case for crypto / Web3 / DeFi: it represents the great opening up a new set of assets to innovators. There is and will be rampant fraud, speculation and silliness, but the long-run impact of these moments tends to be large and positive.

Quick links

Survival of the fittest? Covid is bad, even for professional footballers.
You're not Gen-Z, you're young. The concept of generations has little explanatory power and is bad science. Good thread.
New things to worry about. China's hypersonic weapons test. And particularly this thread: "strategic stability is slipping away".
The rich get richer and the poor get... richer? Wages for (US) lower paid jobs are rising faster than since the late 1990s.
Failure is worse than it looks... you can't even learn much from it, according to this study.

How you can help

I don't think when I started writing TiB I thought I'd still be doing it 187 editions in... so thank you for reading. It makes it worth it.

If you'd like to help, just forward this to a friend or share it on Twitter / LinkedIn.

And if you have comments, questions or recommendations, just hit reply.

Until next week,

Matt Clifford

PS: Lots of newsletters get stuck in Gmail’s Promotions tab. If you find it in there, please help train the algorithm by dragging it to Primary. It makes a big difference.