Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

kodos96 comments on The Importance of Goodhart's Law - Less Wrong

75 Post author: blogospheroid 13 March 2010 08:19AM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Comments (113)

You are viewing a single comment's thread.

Comment author: kodos96 17 March 2010 08:56:17PM *  8 points [-]

Getting back to trying to propose practical mitigation strategies for goodhart's law, I propose a fairly simple solution: Choose a G*, evaluate performance based on it, but KEEP IT SECRET. This of course wouldn't really work for national scale, GDP-esque kind of situations, but for corporate management situations it seems like it could work well enough. If only upper management knows what G* is, it becomes impossible to optimize for it, and everyone has to just keep working under the assumption they're being evaluated on G.

Taking it a step further, to hedge against employees eventually figuring out G* and surreptitiously optimizing for it, you could have a bounty on guessing G* - the first employee who figures out what the mystery metric G* really is gets a prize, and as soon as it's claimed, you switch to using G**

Comment author: wnoise 17 March 2010 09:51:05PM 7 points [-]

The hedge is absolutely necessary, elsewise, a manager will just tell subordinates what G* is in order to look impressive for managing a high-performing group.

Comment author: froob 17 August 2010 01:46:21PM 6 points [-]

Andrew Grove (of Intel fame) wrote a book, High Output Management, suggesting that management needs two opposing metrics to avoid this problem. For example, measure productivity and number of defects, and score people on the combined results.

BTW the large nail/little nail joke has a third part. Soviet management eventually got a clue and started measuring by the value of the nails produced... and the result was the world's first solid-gold-nail factory.

Comment author: Sniffnoy 17 August 2010 11:09:47PM *  2 points [-]

Presumably finding arbitrarily many basic G*'s will be hard. Two ideas for dealing with this: 1. Even if you only have finitely many and they're all known, you could select one at random each time there's a switch. 2. Each time there's a switch, select a somehow-random linear (or some other sort, if you like) combination of your basic G*'s. (That would make guessing it in the first place quite hard, actually...)

Comment author: botogol 01 April 2010 03:15:07PM 0 points [-]

if management are doing that then are neglecting a powerful tool in their tool-kit, because announcing a G* will surely cause G* to fall, and experience says that to begin with a well-chosen G* and G remain correlated (because many of the things to do to reduce G* also reduce G). It is only over time that G* and G detach.