Why Greatness Cannot Be Planned

IN PROGRESS (https://readwise.io/bookreview/50377846)

A straightforward argument, made by very loose analogy to evolution & machine learning, that directly aiming for “greatness” via objectives does not work. Seeking out novelty is more effective.

If much of the map/loss landscape is unknown, it’s often better to pursue the next most novel action rather than directly optimizing some global objective. Such novelty search allows you to learn more about the map than a direct optimization step.

Evolution does not work by shooting directly for some ultimate being. Rather, it is a process of trial and error that produces organisms as well-adapted as possible to a particular set of conditions.

I did like the framing of how global competition (akin to a global loss in ML) reduces diversity whereas local competition encourages it.

Implicit in this idea is that any objective that can be aimed & planned for is almost definitionally not “great” because it is visible/essentially incremental.

Provenance

No idea! Probably somewhere around the Pathless scene