You asked

what safeguards can we build into algorithmic design to prevent 'blindness' or 'perversion'?

Reading with you…