From a general standpoint, we develop automated solutions by following new product development[1]. More specifically, we develop features for users to provide different kinds of feedback, and write functions that combine that feedback with machine learning to perform actions when necessary.
Even more specifically, the entirety of forum interaction is an input/output of information in people's brains, computed along with emotion, heuristics, and whatever knowledge the brain has, and generates output. By collecting meta-information about the input and output (such as crowdsourced content flags like category of information, perceived intent of a comment, training data models on old content, etc) we can make functions that use the metadata to perform actions, such as auto-detaching threads, muting users, delaying reply buttons, providing feedback to commenters, and highlighting or shadowing content. A lot of these are already done (some automated, some not), but I think the big feature is when these are done; find the emotional pain points, provide functions that mitigate them based on metadata.