I think you are referring to Goodheart’s law, because all the measures your examples used as a proxy to achieve some goal were gamified in a way that the proxy stopped working reliably.
Hmm, this seems a little different from Goodhart’s law (or at least it’s a particular special case that deserves its own name).
This concept, as I understand it, is not about picking the wrong metric to optimize. It’s more like picking the wrong metric to satisfice, or putting the bar for satisficing in the wrong place.
I think you are referring to Goodheart’s law, because all the measures your examples used as a proxy to achieve some goal were gamified in a way that the proxy stopped working reliably.
Hmm, this seems a little different from Goodhart’s law (or at least it’s a particular special case that deserves its own name).
This concept, as I understand it, is not about picking the wrong metric to optimize. It’s more like picking the wrong metric to satisfice, or putting the bar for satisficing in the wrong place.