Amplifying this because nail->head. Also adding to the FAE angle, we have to consider what a potentially "bad" action looks like not just outside of context, b

Amplifying this because nail->head. Also adding to the FAE angle, we have to consider what a potentially "bad" action looks like not just outside of context, but in aggregate, because in these systems you're probably looking for outliers, unusual sentiment, not just key terms. This has a chilling effect on any extra-normal behavior. If it was implicit that context leading up to and afterwards were required to interpret the outlier, one would see it's probably not an outlier after all, but that requires a lot of people's time and potentially your involvement. Even if that context could be supplied with tons of surveillance and people/AI, I'd wager people would rather not be hauled in front of the internet police to explain every non-obvious, out-of-the-ordinary thing they directly or indirectly associated with.