OTOH, some particularly clever trolls are of course intended to provoke division of an intellectually unproductive sort. Not sure if machine learning could identify that...though features computed from the commenter's past history could be informative.
Then you can enable or disable aspects of the site for each user-category. e.g. disable comments from trolls, disable downvoting for people who downvote when disagree as this stimulates groupthink, etc...
A set of users can serve as a training set for each category. It is possible to let the 'genuine'-category classify comments from users from the other categories in order to enhance the classifiers.
Actually this is an interesting machine learning problem, I still need to pick a project for my course in machine learning so I can research this if I can get my hands on the datasets.
The goal could be something like Google's Beer Goggles, but for comments: "it looks like you're writing a flame, are you sure you want to do that?"
Another thought: there's a big difference between a comment at -1 with 10 upvotes and 12 downvotes vs. one that has occasioned 2 downvotes. Might be interesting to see the effect of showing both the number of up & down votes rather than just the sum. The idea is that genuinely trollish comments at -1 will be differentiated from minority points of view.