SummarizingDontReadHistoryForLessons

source Last Accessed 2024-08-2024

Summary

In this article, Cedric Chin argues that reading history for lessons can be detrimental because you risk misapplying the underlying concepts. To clarify, “lessons” are framed as broadly applied concepts that overlook the highly specific context of the anecdote from which they are extracted. Instead of reading history for “lessons”, Chin argues that we should read for “concept instantiations.” In his own words,

The goal of reading from history, then, is to expand the set of prototypes in your head.

Chin argues that this approach aligns with a central claim of Cognitive Flexiblity theory that

experts in ill-structured domains reason by comparison to fragments from prior cases that they’ve seen - one implication is that concepts are represented not as abstract principles in their heads, but a cluster of real world cases that serve as prototypes.

Takeaways

• One practice that Chin uses, that I appreciate and intend to mimic, is to mentally append “except when it doesn’t” to lists of historical lessons.
• Expanding upon that practice, I think a good follow up exercise is to frame the lesson as a theorem and approach it with counterexamples and edge cases.

Discussion

This was such a fascinating read for me because it is so opposite of my training. In mathematics, programming, and (as we tend to teach) physics, we work with first principles. Even if you’re simulating noisy data, you start with some core model and layer some noise on top of it. In contrast, professions like law, medicine, and paleontology operate off of case studies. It seems like it would be incredibly useful to be able to switch into a case-study mental mode for tasks where a first-principle approach isn’t appropriate.

The question, then, is when is a first-principles approach appropriate and when is a case-study approach appropriate? My instinct is that first-principles is more powerful but less robust than a case-studies approach. If the system is consistent between input and output, that seems like a viable candidate to look for first principles. Otherwise the case-study approach seems best. There are probably other requirements for a first-principle approach too.