You might say that you target evil creatures because they're likely to commit offenses that are punishable under law by death, but then, you might say that certain crimes are punishable by death because they show that the perpetrators are Evil.

In a world in which you can cast "Detect Evil", but don't know which of these two is true, the word "Evil" attached to your "Detect Evil" spell may not have the semantic weight you think it does.

All you know is that you have a particular sensory action that you can perform, which returns a quantitative result when applied to a given target. We have chosen to call this quantitative value "Evil". To be clearer, let's call it their EQ (for "Evil Quotient").

You happen to know, experimentally, that beings with a high EQ tend to commit actions that decrease general utility in the population whose utility you care about. Now, you have an important question to ask yourself: is high EQ causative of that net decrease in general utility, or is it merely correlative?

You then have a further philosophical question: Should the utility of high-EQ individuals be weighed the same as the utility of other individuals when aggregating your global utility function? (This will depend on many things, one of which is the potential for "false positives", but another of which is the base assumption of whose utilities are worth considering).