Jones & Love (J&L) are right to criticize what they term “Bayesian Fundamentalism” as not empirically grounded, uninformed by psychological data, open to multiple rational accounts of a task or decision, and conducive to post hoc explanations. However, in advocating Bayesian Enlightenment as a solution, they appear to rule out a broader critique of rationalist approaches to human cognition. Specifically, Bayesian Fundamentalism is one example of the more general phenomenon of Rationalist Fundamentalism: the tendency to characterize a given judgment as rational and optimal in a post hoc manner, after the empirical data are already known. Few researchers would argue that human behavior is perfectly optimal and rational. However, a desire to see the human mind as operating rationally, and the use of post hoc justifications to reach this conclusion, suggest we should be skeptical of after-the-fact “rational” explanations.

Decades of empirical studies show people are strongly motivated to see themselves as rational and objective (for reviews, see Armor Reference Armor1999; Pronin et al. Reference Pronin, Gilovich and Ross2004; Pyszczynski & Greenberg Reference Pyszczynski, Greenberg and Berkowitz1987; Ross & Ward Reference Ross, Ward, Reed and Turiel1996). Decision makers engage in motivated reasoning and psychological rationalizations designed to preserve this “illusion of objectivity” (Armor Reference Armor1999; Pronin et al. Reference Pronin, Lin and Ross2002) – for instance, changing their definition of what an optimal judgment is after the fact (Dunning & Cohen Reference Dunning and Cohen1992; Epstein et al. Reference Epstein, Lipson, Holstein and Huh1992; Kunda Reference Kunda1987; Norton et al. Reference Norton, Vandello and Darley2004; Uhlmann & Cohen Reference Uhlmann and Cohen2005). Evidence that general psychological processes are not rational or optimal represents a threat to this cherished illusion. Fundamentalist resistance to evidence of human irrationality further stems from economics and related disciplines, in which optimality and the maximization of utility are widely perceived as necessary assumptions about human behavior.

A rationalist defense can involve constructing a post hoc Bayesian account of an empirical finding predicted a priori from theories grounded in psychological limitations and motives. Consider the phenomenon of biased assimilation, in which participants rate a scientific study that supports their political beliefs (e.g., about the deterrent effects of capital punishment) as methodologically superior to a study that refutes their beliefs (Lord et al. Reference Lord, Ross and Lepper1979). The cognitive-rationalist interpretation is that decision makers are simply making Bayesian inferences, taking into account subjective probabilities (e.g., their prior political beliefs) when evaluating new evidence. However, further findings contradict the claim that biased assimilation is merely the product of Bayesian inferences. For instance, individuals whose positive self-image is affirmed are less likely to exhibit biased assimilation (Cohen et al. Reference Cohen, Aronson and Steele2000; see also Dunning et al. Reference Dunning, Leuenberger and Sherman1995; Sherman & Cohen Reference Sherman and Cohen2002). This is consistent with the idea that biased information processing stems from a motivated desire to dismiss evidence that threatens valued beliefs and, by extension, the self (Sherman & Cohen Reference Sherman, Cohen and Zanna2006; Steele Reference Steele and Berkowitz1988). When a decision maker is feeling good about herself, there is less need to be biased. In addition, would-be parents who believe day care is bad for children, but plan to use day care themselves (and therefore desire to conclude that day care is just as good as home care), show biased assimilation in favor of day care (Bastardi et al. Reference Bastardi, Uhlmann and Ross2011). What decision makers desire to be true seems to trump what they believe to be factually true – the ostensive basis for any Bayesian inferences.

As J&L point out, one of the most problematic aspects of rational models is how little attention can be paid to whether the assumptions of the statistical model correspond to what is actually going on in people's heads as they engage in a task or make a decision. I once debated an economist who argued that micro-level psychological data on what goals people pursue in the dictator game are irrelevant: The material self-interest account must be true if people's offers correspond to the predictions of the statistical model. However, it is dangerous to assume that because a rational statistical model can mimic or reproduce a pattern of data, the underlying psychological process is a rational one. That a computer can mimic some of the outputs of human thought does not necessarily mean the mind functions in the same way as a computer.

The last defense of post hoc rationalism is to swap normative models of rationality entirely. In other words, researchers can speculate post-hoc as to what alternative goals decision-makers may have been pursuing, in order to preserve the view that participants were acting rationally. Never mind the goals to optimize material outcomes or achieve accuracy: Judgmental biases can be defined as “rational” because they preserve the decision maker's personal self-image, psychological well-adjustment, public reputation, cherished religious beliefs, desire to punish norm violators, existential goals, likelihood of survival in ancestral environments, or even the happiness of their marriage (Cosmides & Tooby Reference Cosmides, Tooby, Hirschfeld and Gelman1994; Hamilton Reference Hamilton1980; Krueger & Funder Reference Krueger and Funder2004; Lerner & Tetlock Reference Lerner and Tetlock1999; Tetlock Reference Tetlock2002; Tetlock et al. Reference Tetlock, Kristel, Elson, Green and Lerner2000; Reference Tetlock, Visser, Singh, Polifroni, Elson, Mazzocco and Rescober2007).

It has been argued that the heuristics-and-biases approach to cognition is itself biased, in the direction of attributions to irrationality (Krueger & Funder Reference Krueger and Funder2004). Despite its shortcomings, however, the heuristics-and-biases research program is at least based on a priori theoretical hypotheses. There are few cases of “post hoc irrationalism” in which robust empirical effects predicted a priori by Bayesian or otherwise rationalist models are redefined post hoc as due to motives such as the need for self-esteem or control.

Although Bayesian Enlightenment, as advocated by J&L, is a major improvement on Bayesian Fundamentalism, it is still subject to post hoc rationalism. An interface between Bayesian or otherwise rationalist models and data on psychological processes leaves plenty of room for the former to distort interpretations of the latter. A wealth of evidence indicates that human beings are subject to a powerful illusion of rationality and objectivity they are strongly motivated to maintain and which influences their perceptions of scientific data. Researchers are also human beings. It would be remarkable indeed if scientists were immune to the empirical phenomena we study.