517da335fd0ec2f4a25ea139d5494163-Paper.pdf

Itisoften the responsibility of the agent designer toconstruct thistargetwhich,inrichandcomplexenvironments,constitutesaonerousburden; without full knowledge of the environment itself, a designer may forge a suboptimal learning target that poorly balances the amount ofinformation an agent must acquire to identify the target against the target's associated performance shortfall.