On The Use Of Surrogate Models For Bayesian Inverse Problems