'For over a century, most statisticians have known that, in parametric inference problems, Bayes's theorem provides the best achievable informational summary of the data if certain prior information was available. For nearly that long, some statisticians have sought to extract sufficient information from the same data to provide a reasonable substitute for the needed prior information, in effect allowing the analyst to reach near perfection without making unreasonable prior assumptions. Koenker and Gu summarize and advance this program with a welcome display of what may be termed these days 'Natural Intelligence.' Stephen M. Stigler, University of Chicago.