Modelling biased human trust dynamics