Confirmation bias is the essential engine of AI training. The weight given to an outcome that deems it “most likely” doesn’t come from reason, but confirmation bias.
AIs with confirmation bias are also notoriously opaque - decisions are made quickly and confidently, but never justified. The closest you might get to an explanation is a vague indication that some input resembles past inputs. This is of course how prejudice and intolerance work: the only explanation is “everyone just knows this is true.”