> The problem is when we interpret every event through a single filter, or give one filter primacy in all interpretations of events that filter now colors everything we see.

Anyone noticing the irony of this seeming to be pretty much what the paper author seems to be blaming the machine learning systems for ?