Thought: right-leaning bias complexes are likely much more easily machine-learnable than left-leaning due to greater internal consistency/cohesion. Recommender systems converging on right-wing conspiracy content are a first indicator.
The singularity will be Trumpist 
That’s very well put! I do see this as more of an issue of model-ability of more basic “lizard brain” reinforcement loops (based on current state of AI) v “cohesion” which is too abstract a notion for me to grasp easily in this context.