When a stranger collapsed onto the subway tracks at 137th Street and Broadway in New York City's Upper Manhattan, fellow subway-rider Wesley Autrey did not have time to assess the stranger's moral commitments, nor could he access the information needed to do so (see Buckley Reference Buckley2007). Instead, as the No. 1 train barreled toward the stranger, Autrey evaluated the scene and – if we believe the sentiments quoted in the epigram – acted solely on his own beliefs. He sprung from the platform and risked his life to save the stranger. This action, albeit exceptional in its potential costs and realized benefits, is but a single example of the anonymous altruism (Engel Reference Engel2011) and rapid-fire cooperation (Rand et al. Reference Rand, Peysakhovich, Kraft-Todd, Newman, Wurzbacher, Nowak and Greene2014) that has distinguished humans as “hypercooperators.” Also, it exemplifies a form of prosocial behavior that cannot be explained by moral externalization.

As Stanford explains in the target article, the effect of moral externalization on prosociality “is strictly limited by the accuracy and detail of the information agents are able to acquire about one another's distinctively moral commitments” (sect. 6, para. 1). According to this line of reasoning, Wesley Autrey will act altruistically only after gathering the information needed to compare his moral commitments with those of the guy lying on the tracks. By definition, this accurate and detailed information does not exist in the relatively anonymous environments in which humans display their unique form of prosociality (see, e.g., target article, sect. 4, paras. 1–3, for discussion of such environments). Accordingly, moral externalization seems to be an unlikely mechanism for the evolution of hypercooperation.

Furthermore, these information demands impose costs that other mechanisms for cooperation avoid, thus making moral externalization an unlikely mechanism for prosociality even in more pedestrian social encounters. To see this point, note that Stanford argues that moral externalization is a mechanism for correlated interaction (a.k.a. positive assortment). Correlated interaction ensures that the beneficiaries of prosocial behavior are those who share the same heritable attribute driving that behavior (e.g., genotype or strategy), not merely those with the same inclination to act prosocially in a given encounter (see, e.g., Hamilton Reference Hamilton1964a; Reference Hamilton1964b; Van Cleve & Akçay Reference Van Cleve and Akçay2014). After all, if only the inclination to cooperate is used to discriminate among social partners, then the population could drift to an easy-to-exploit cooperative strategy that opens the door to exploitation by defectors (see, e.g., Johnson & Smirnov Reference Johnson, Smirnov, Herwig and Hoffrage2013). Hence, for moral externalization to promote correlated interaction, agents must rely not only on truthful and meticulous information about the moral commitments that make others likely to cooperate in an encounter (target article, sect. 6, para. 1), but also on information about whether others’ moral commitments make them likely to cooperate because they share the same heritable attribute triggering cooperation.

Moral externalization thus requires extensive information to facilitate positive assortment. Other mechanisms need less information to do so. For instance, conditioning cooperation on material parity generates positive assortment via a comparison of two pieces of information: the value of one's wealth holdings and the approximate value of a social partner's holdings (Johnson & Smirnov Reference Johnson and Smirnov2012). This information is likely to be available without search, due to the benefits of signaling it, and it is likely to be accurate, due to the costs of generating the signal (see, e.g., Nelissen & Meijers Reference Nelissen and Meijers2011). On the contrary, cooperation via moral externalization requires the comparison of multi-dimensional information (do all of the attributes of your moral code match mine?), and that information can be readily faked, as Stanford's discussion about detecting moral commitments implies (sect. 6, para. 1). As a result, one must wonder: Why would natural selection favor a mechanism for prosociality that relies on intensive information search and complex cognition, when low-information and cognitively simple mechanisms for correlated interaction exist? I do not think one can answer this question in a way favorable to Stanford's thesis.

However, with these criticisms issued, Stanford deserves credit for identifying a mechanism that could enable prosociality in certain environments, though – per the discussion above – it likely would have had to evolve for another purpose. That is, were it a “spandrel” resulting from cognitive mechanisms that proved to be adaptive for other reasons, then moral externalization could be co-opted to facilitate prosocial behavior in information-rich environments. Arguing that it evolved because it produces correlated interaction that facilitates humans’ unique form of prosociality seems less plausible. The environments in which humans display their distinctive form of prosociality lack the information that moral externalization requires, and, in information-dense environments, less costly mechanisms can facilitate the correlated interaction needed to support cooperation.