You have some system that tried to balance bias, variance, and utility. Sometimes that best way to do that is to have some bias. The classic example of this is you may jump when you see something that looks like a predator, but then turns out not to be. The cost of being wrong (see nothing when predator, see predator when nothing) means it's optimal to sometimes see danger that isn't there.
You have some system that tried to balance bias, variance, and utility. Sometimes that best way to do that is to have some bias. The classic example of this is you may jump when you see something that looks like a predator, but then turns out not to be. The cost of being wrong (see nothing when predator, see predator when nothing) means it's optimal to sometimes see danger that isn't there.