@newscientist um... how about not training on data randomly extracted from a known-toxic media?
Then again, if the data is to be vetted, who gets to do the vetting? Clergy? Lawyers? Politicians? Academics? Mom?
Better idea: instead of building sociopathic superstition machines, let's build reasoning that recognizes probable consequences of utterances before uttering?
Mind you, the Age of Enlightenment didn't have a great track record at this either...





