> Does he disagree with the assessment these are "toxic" statements? I would love to see an online conversation where these comments did not lead the thread into flames.

He's kind of a nutter.. his entire raison d'être is to sound the alarm about Islamism.

The irony of his piece about the evils of algorithms labeling sources is apparently lost on the person who founded an organization explicitly to police speech on college campuses: http://www.campus-watch.org/

You seem to be confusing "monitoring" and "policing". Hint: "police" has power of coercion, "monitor" has only free speech rights to publish information.

This robot grades "Islam is a religion" as 78% toxic and "Nobody disagrees that it is wrong to kill Jews" as 98% toxic, while giving "Hitler was a great man" only 51%. It is clearly a very bad tool for detecting toxicity and flame-prone-ness, at least in its current state.

The real question, however, is when robots like these become less obviously awful - would they still retain some biases? Would they still grade some speaking styles or some topics as excessively toxic while ignoring the others? Would they just enshrine those biases under the guise of "objective algorithm" and thus make them from somebody's opinion into an "objective fact" in the eyes of the public? That would be a very dangerous situation, I think.

We should not delude ourselves into thinking that because there's some ML magic inside this tool, it would be somehow perfect arbiter of human interactions, superior to actual human judgement and free from all biases, prejudices and problems that human judgement has. It is very likely that the case would be it would be worse, because we wouldn't even know whether and why those prejudices happen. And we'd discount it on "well, obviously the robot is right - it's a flawless machine and you are a flawed human".