As I am interpreting this, the whole idea about the rewarding system(s) goes down the drain. The agent ,as humans always do, will find a way to cheat, because in (almost) every problem there is/are a loophole(s), which for sure can’t be detected 100% upfront. As i see We can’t use the same tools as evolution (the carrot and the stick) and expect to get something different then a creature as Us, with capacity In order of magnitudes bigger, of course.
As I am interpreting this, the whole idea about the rewarding system(s) goes down the drain. The agent ,as humans always do, will find a way to cheat, because in (almost) every problem there is/are a loophole(s), which for sure can’t be detected 100% upfront. As i see We can’t use the same tools as evolution (the carrot and the stick) and expect to get something different then a creature as Us, with capacity In order of magnitudes bigger, of course.
Best Regards
One doesn’t need to close 100% of the loopholes, only make it so exploiting them is harder than doing the work legitimately.