This reminds me of the false-positive vs false-negative dilemma faced in medical testing. Either you optimize for low cost and convenience, or for catching true-positives or true-negatives. For HIV testing for instance, if someone does test HIV-positive falsely, their changed safer behavior in the short-term wouldn't harm anyone, and follow-up tests could catch that they're actually negative. But if we falsely say someone does not have cancer, when they do, it could grow a lot in the short-term before symptoms arise and another test is given. They may also use a lot of unnecessary care trying to diagnose the issue after that false-negative cancer test.
What's the right solution? It's case by case, down to a mixture of morality and expertise to decide.
Seems these tech algorithms often generate a lot of false-positives wrongfully, or that's what's posted online afterwards. It'd be interesting to dig into the numbers for various platforms, see if they're falsely negative for spam accounts and bad actors. We wouldn't hear posts on HN about spam bots that cut into FB's bottom line, would we?