In my opinion, the most important sources of moral disagreements are disagreements about what conditional or unconditional predictions—whether deterministic or probabilisitic [sic] predictions—to make about future empirical facts. For example, we cannot expect a moral consensus in our society as long as we strongly disagree on what the likely effects of the conventional anti-inflationary policies will be in somewhat unusual economic situations, for which no real historical precedents may exist; or on what the long-run effects of continuing inflation will be on our social institutions; or on how our children will be affected in the long run by an increased emphasis on 'creativity', and a decreased emphasis on 'intellectual excellence', in our schools; or on whether a greater pressure on certain despotic regimes will speed up or rather slow down their hoped-for liberalization and democratization, etc. No doubt, further progress in human psychology, in the social sciences, and, in some cases, in the natural sciences, will give clearer answers to some of these problems than now are available to us. But open problems of this type will always remain with us—partly because the very solution of some of these problems will give rise to new ones. By solving some of these problems a radically new historical situation is created, and even if we have learned how to make predictions in the old situation, this may not help us very much in making predictions in the new one. Keynesian economics has enabled us to make much better predictions about the effects of various economic policies in conditions of mass unemployment. By this means, it has also enabled us to eliminate these very conditions, and to create a completely novel economic situation of continuing high employment, in which Keynesian predictions may no longer work.
(John C. Harsanyi, "Nonlinear Social Welfare Functions: Do Welfare Economists Have a Special Exemption from Bayesian Rationality?" Theory and Decision 6 [August 1975]: 311-32, at 329-30 [endnotes omitted])
