Some Bayesian Experimental Design Theory for Risk Reduction in Extrapolation