This week I came across this interesting paper
on the optimal way to experiment with new ideas when the outcomes are power law distributed (We talked about power laws in TiB 113
and TiB 127
). Interestingly, the authors conclude that the right strategy looks a lot like Lean Startup
methodology: try lots of things and double down on outlier results, even when there’s a lot of noise in the data.
It’s interesting to think about how this idea applies to talent discovery. It seems plausible that talent has a “fat tailed
” distribution where, at least in some fields, even small differences among the most talented are associated with large changes in real-world outcomes (see TiB 125
for some suggestive evidence from the world of maths). The challenge is that in many domains, it’s difficult to identify exceptional talent by testing. Most of the time, you have actually to observe the activity you care about - but that tends to be time consuming and expensive.
Given this, one conclusion we might draw from the paper is that we should run many more experiments that allow talent to be revealed. Entrepreneur First
is one such approach in the startup domain - and some smart people have suggested that similar models might be impactful in science, art and other fields (see, e.g., Sam Altman here
). It’s an area where I’d love to see more work.