If the system can be set up so that the status quo is one possible result, the risk of it getting worse is bounded by the difference between the optimization criterion and whatever you think of as better. If you are reasonably sure they are well-aligned, you can just try it out to see if things get better.