4 ms·

I agree with this, so many problems can be fixed with proper incentive structures.

• setr 8y ago
  The issue is the same as ML: the damn things will figure out how to misinterpret the incentive into something perverse, if given the chance. Even worse, they have the opportunity to influence the incentives, or change them outright, to give themselves the chance; and since the incentives have to change over time, to keep up with changes in the environment, any stable set of incentives is likely to become unstable in due time. So incentives are alone not the answer; it needs to also somehow be self-stabilizing, and self-correcting, and also not be wrong (because such a structure would be difficult to fix, as its designed to be difficult to edit in the first place !)