My takeaway from that article was that there's not enough trust in the raw inputs and possibly a missing continuous weighting system of the sensory interpretations. (humans do this naturally, weighing their various sensory interpretations in real-time simultaneously)