4 ms·

> The Silicon Valley touts itself as a meritocracy where people climb the economic ladder based on the power of their ideas. Maybe it's not worth pointing this

> The Silicon Valley touts itself as a meritocracy where people climb the economic ladder based on the power of their ideas. Maybe it's not worth pointing this out over and over again, but a meritocracy is not a system whereby you climb the economic ladder based on the "power of your ideas". It's a system in which you climb a hierarchy by scoring well in some externally defined "unbiased" evaluation process. The irony, which was intended in the coining of the term, is that such a test can never be unbiased and the institution that defines the test has power over everybody (read: The College Board). There cannot exist even a model of a society where people succeed based purely on how "good" they are that also doesn't discriminate, since sorting the "good" from the "bad" is itself an act of discrimination.

api 13y ago
> It's a system in which you climb a hierarchy by scoring well in some externally defined "unbiased" evaluation process. Getting deeply into evolutionary theory and learning theory destroyed my belief in naive meritocracy, and not for the typical reasons either. What you learn is that defining merit in any complex system is often as difficult as actually achieving it. If the fitness landscape is non computable then defining merit globally IS precisely as difficult as achieving it. The two reduce to the same problem! ... and of course any real world problem domain that is worth doing and profitable to solve is almost definitely non computable, complex, shifting, chaotic, all those nasty things ... Moreover, if you define merit via any "merit function" that does not properly model your problem domain then the system will optimize to your merit definition rather than for actual merit. I watched this happen over and over again with genetic algorithms, simulated annealing, and other learning algorithms. Sometimes it's comical... I watched a damn genetic programming system evolve to predict the behavior of my operating system's kernel thread scheduler to predict whether they would receive an "A" or "B" category item (I had a multithreaded fitness function) instead of classifying the item for example. Nature always prefers the lowest-complexity solution for fundamental second law of thermodynamics type reasons. I think this is why things like stack ranking, etc., result in superficial short-term improvements to company performance followed by long-term collapses in creativity and deep performance over ensuing years. When you implement stack ranking then your company's culture optimizes itself for stack ranking, not for the market. Stack ranking captures a small aspect of the overall fitness function of the market -- namely tenacity and work ethic -- but fails to capture anything more than that. So you get gerbils, not engineers. It's a hard problem. Be suspicious of sound-bite answers, fads, and simple ideologies in these waters.
nickff 13y ago
>"There cannot exist even a model of a society where people succeed based purely on how "good" they are that also doesn't discriminate, since sorting the "good" from the "bad" is itself an act of discrimination." I think this is a very important point, that optimization is inseparable from distribution. Daron Acemoglu also makes this point very well (when he talks about different political regimes).

api 13y ago

> It's a system in which you climb a hierarchy by scoring well in some externally defined "unbiased" evaluation process. Getting deeply into evolutionary theory and learning theory destroyed my belief in naive meritocracy, and not for the typical reasons either. What you learn is that defining merit in any complex system is often as difficult as actually achieving it. If the fitness landscape is non computable then defining merit globally IS precisely as difficult as achieving it. The two reduce to the same problem! ... and of course any real world problem domain that is worth doing and profitable to solve is almost definitely non computable, complex, shifting, chaotic, all those nasty things ... Moreover, if you define merit via any "merit function" that does not properly model your problem domain then the system will optimize to your merit definition rather than for actual merit. I watched this happen over and over again with genetic algorithms, simulated annealing, and other learning algorithms. Sometimes it's comical... I watched a damn genetic programming system evolve to predict the behavior of my operating system's kernel thread scheduler to predict whether they would receive an "A" or "B" category item (I had a multithreaded fitness function) instead of classifying the item for example. Nature always prefers the lowest-complexity solution for fundamental second law of thermodynamics type reasons. I think this is why things like stack ranking, etc., result in superficial short-term improvements to company performance followed by long-term collapses in creativity and deep performance over ensuing years. When you implement stack ranking then your company's culture optimizes itself for stack ranking, not for the market. Stack ranking captures a small aspect of the overall fitness function of the market -- namely tenacity and work ethic -- but fails to capture anything more than that. So you get gerbils, not engineers. It's a hard problem. Be suspicious of sound-bite answers, fads, and simple ideologies in these waters.

nickff 13y ago

>"There cannot exist even a model of a society where people succeed based purely on how "good" they are that also doesn't discriminate, since sorting the "good" from the "bad" is itself an act of discrimination." I think this is a very important point, that optimization is inseparable from distribution. Daron Acemoglu also makes this point very well (when he talks about different political regimes).