The interesting thing is that literally all of the inferences we are "worried" about computers making were social consensuses not too long ago, with ample first-order evidence to support them even in the absence of fancy machine learning algorithms.