Although the "how" part seems to be "via our proprietary process", they are pretty specific about what concrete forms of bias they're trying to rule out. They l

Although the "how" part seems to be "via our proprietary process", they are pretty specific about what concrete forms of bias they're trying to rule out. They list that, with a few examples, here: https://www.theknifemedia.com/our-process/ https://www.theknifemedia.com/our-process/ The first stage ("the spin") is just identifying words and phrases which add spin with a positive or negative valance - this part could easily be automated with a dictionary search or by training a neural net. (There should probably be a Chrome plugin for it.) The other two steps are more tricky and would seem to require a human editor's eye. "the slant" involves noticing when one perspective is put in an article to the exclusion of other, similarly valid perspectives which have been left out - to do that you'd have to either be a subject matter expert or combine enough stories from different sources to identify the universe of possible takes. "the logic" involves noticing when conclusions don't follow from the facts presented; this also would appear to require somebody carefully reading and understanding the story. So my guess is that step 1 is fully-automated now (but being tweaked) and steps 2 and 3 need human editors with some algorithmic assistance (which they're trying to increase over time) - at least, that's how I'd have set it up...