Thoughts in Between

by Matt Clifford

TiB 177: The causes of exceptional performance; When founders die; AGI; 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.

Where does exceptional performance come from?

Probably the best known idea about where exceptional performance comes from is the “deliberate practice” theory. This was popularised by Malcolm Gladwell in Outliers as 10,000 hours of practice in a narrow field. But this finding has proven hard to replicate and has been challenged by a number of researchers. A fascinating new paper conducts a meta analysis of over 50 studies covering more than 6,000 athletes and reaches a quite different conclusion: “deliberate practice” is highly correlated with exceptional performance at junior level, but not among the very highest performers - world class senior athletes.

In fact in this group, the Roger Federers and Michael Phelps of the world, we observe almost the opposite: these people are more likely to have played multiple sports as adolescents, to have started their primary sport later and to have progressed more slowly than “mere” national champions. As the paper notes, this has big implications for talent development programmes, which typically select athletes very early and may “reinforce rapid junior success at the expense of long-term senior success”.

It also suggests that (pace our discussion in TiB 125 about mathematics talent) we should be wary of assuming that a fast rate of initial progress is the best predictor of long term success. Intriguingly, the authors point out that we observe a similar effect in a very different field: science. They note that Nobel prize winners take longer to reach full professorships and are less likely to have won undergraduate scholarships than slightly less successful scientists. This feels like a research field worthy of more attention.

Another step towards AGI?

Some of the most striking developments in AI in recent years have involved training artificially intelligent agents to exceed human performance in games. DeepMind’s triumph in Go is the most famous example, but there have been successes in games as diverse as Starcraft II and chess. Until now, though, agents have had to be trained to play specific games, one at a time. DeepMind just released a paper on “generally capable agents”, which pushes beyond this: they train AI agents to learn to play arbitrary games, including games they’ve never encountered before, in an open ended environment. This video summarises some of the results.

This is potentially a big deal. Being able to train AI agents with general and adaptable capabilities is likely an important milestone along the path to artificial general intelligence. The paper has generated a lot of chatter in the AI safety world - see this comment thread for lots of discussion, some quite technical. I thought this analogy was potentially quite helpful:

This is the GPT-1 of agent/goal-directed AGI; it is the proof of concept. Two more papers down the line… and we'll have the agent/goal-directed AGI equivalent of GPT-3

(See here for our previous discussions of GPT-2/GPT-3 and its importance in AI) Not everyone is so impressed. It’s worth reading Rohin Shah’s write up in the excellent AI Alignment newsletter. He notes that the capabilities of these agents may be less generalisable outside their specific training environment than it appears.

One thing I found striking as I read the paper was how many training “behaviours” the successful AIs had in common with the top performing athletes discussed in the section above: the importance of training in multiple sports, the importance of self-play, the lack of immediate progress, etc. This may be an echo without meaning, but I wonder if there is something interesting there. Certainly, this paper represents a line of research worth keeping an eye on.

When founders die - or, why startups are special

One peculiar (anecdotal) observation about startups is that they’re much harder to replicate than they should be. Even when a founding team is still very small, the product not yet complicated and any network effects non-existent, it seems to be quite hard to simply observe what a startup is doing and copy it (There are exceptions, of course). One explanation is that there is something we might call “organisational capital” that is created when talented founders develop company- and idea-specific ways of working together that turn out to be highly productive.

This sort of explanation risks tautology (productive firms are more productive because… they’re more productive!), but a new and fascinating study suggests a (morbid) way to test it: look at what happens when a founder die unexpectedly. If startups are simply the sum of the talents within them, unexpected loss of a team member shouldn’t have a long term impact; after an initial shock, a comparably talented person can be hired in their place. Using a novel dataset the authors find large, persistent negative effects from losing a founder or early employee: jobs growth, productivity and revenue all fall relative to peers, and the effects persist for a decade.

The authors report some interesting second order findings too: this loss of organisational capital is twice as large for founders as early employees (but significant for both); larger in smaller founding teams; and larger in B2B startups. They also perform some helpful robustness checks to rule out, for example, that they’re measuring the effect of emotional trauma or loss of labour. If this is an area of interest for you, I highly recommend the paper. It provides impressive empirical backing for something most practitioners feel instinctively: when a great co-founding team comes together it creates something almost magical beyond the sum of its parts.

Quick links

  1. Don't nake typos. They have surprisingly large and negative effects on people's assessments of your intelligence.
  2. It helps a lotto. "Vaccine lotteries" seem to have a positive impact on take up rates and very high ROI.
  3. You pays no money and you takes your choice. Striking age-related variation in vaccine choice in Hong Kong (politics or policy?)
  4. Hopefully WittGANstein is next. An excellent demo of AI's photo generation capabilities - Bertrand Russell as you've never seen him before.
  5. Stag-nations? Europe's post-pandemic economic recovery looks very worrying.

No podcast episode this week, sorry. Normal service should resume from next week…

How you can help

Thanks for reading all the way to the end. If you enjoy, Thoughts in Between, please share it with someone else who might. The easiest way is just to forward the email and social media shares very welcome too.

I'm always happy to hear from readers - 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.