Expertise and History

It’s odd: I’m not a startup founder or indeed any kind of manager, but I enjoy the Common Cog blog. Recently, they posted an article that should have retained its original title, which I believe was Stop Worrying About Survivorship Bias. (They ruined it by appending “With This One Weird Trick” at the end.) The context is reading histories and biographies about business and famous business operators. I think it’s worth reading in full, even though it’s long—or at least I think so.

Histories and biographies relating to business deal almost exclusively with successes or with unusual failures. But many—maybe most—businesses fail for boring reasons and are never written about. The writing therefore suffers from “survivorship bias.”

The main body of the article discusses expertise and how its study evolved starting in the 1980s. I had been aware of the work up to that point, beginning with de Groot’s study of master chess players in the 1940s. I was aware of Ericsson’s concept of deliberate practice. I was not aware of the idea of ill-structured domains, which this article introduced me to. It claims (plausibly) that business is such a domain, which helps justify reading history and biographies even though they do not reflect the field of business at large. (That, and the fact that as practitioners we don’t want an all-encompassing theory of the domain—only enough to be successful in it.)

Medicine is called out as another ill-structured domain, but I think there are many others, including the one I am most interested in professionally: software development. If true, this explains a lot. For one thing, we’ve had any number of attempts to impose paradigms (to name a few: Structured Programming, Object-Oriented Programming, Functional Programming, Agile Development, DevOps), each initially embraced as the Next Big Thing that would make everything better, only to be discovered later to be, at best, a way of avoiding some of the mistakes made before. In an ill-structured domain, there are no universal truths—or at least few really interesting universal truths—only things that sometimes work, depending on context.

To develop expertise in an ill-structured domain, you need to be able to adapt to novel situations by recognizing patterns you’ve seen before and understanding how to fit them together in new ways. Personally, I think of this as training our own neural network, and similar considerations apply. Fortunately, the human neural network is able to learn from far fewer examples than the artificial sort.

This doesn’t mean I’m going to start reading business biographies. Most are closer to hagiography. Instead, it means I’m going to read widely – about business as well as other subjects. The idea is that valuable expertise in the future will be less about nuts-and-bolts knowledge of particular technologies, techniques, or frameworks (to the extent that it ever was) and more about understanding overall context, and when and how to apply those tools.

One last supposition: most areas where people will still have jobs in the future – other than ones where our bodies are useful in ways that robots generally are not – will be ill-defined domains. It’s been true for a long time, maybe even 60 years, that “routine” jobs have been automated away. That same trend is going to affect jobs requiring specialist knowledge, especially where “specialist” still amounts to something like book learning, even if the books are too advanced for most humans.