The problem also goes in the other direction. That is platform rely too much on automation that human signal gets too faint. For example humans have a hard time flagging actual hate speech on these platforms as well. Another example: Every so often there is a front page post on HN where a company like Google will automatically shut down service for a customer (false positive). The customer has a hard time getting through and having this false positive corrected because their signal can’t reach through the layers of automation.