It's easy to talk about "not preserving the option to behave badly" in the abstract, but actually doing it requires perfect foresight (predicting every way anyt

It's easy to talk about "not preserving the option to behave badly" in the abstract, but actually doing it requires perfect foresight (predicting every way anything could be used for evil) and often a lot of implementation difficulty (because you need to completely foreclose on the bad options without impeding good or neutral ones).