Fiery Cushman brings a welcome synoptic vision to psychological processes and outcomes often considered in isolation from one another. By asking the evolutionary why? question, he brings attention to the need to understand how processes frequently associated with error could nonetheless have such a central and pervasive role in the human mind. This evolutionary perspective forces one to seek deep answers, since one must identify features of the mind that could plausibly be the result of natural selection, and so must look behind what plausibly are relatively recent appearances on the psychic landscape, such as rationalization's role in the enhancement of self-image or advertising of one's norm acceptance, which seem to require conceptual and linguistic capacities that probably appeared relatively late in evolutionary history. Moreover, the fact that rationalization as Cushman understands it – the post hoc construction of sense-making belief-desire explanations for one's own or others’ behavior – figures in such a variety of mental phenomena suggests that it is the result of core features of mental architecture. Further, evolutionary explanation requires that one be able to point to benefits sufficiently large and reliable to favor such an architecture, despite its costs.
Therefore, at the heart of Cushman's account are processes that must have long evolutionary histories: reinforcement learning, causal modeling including inverse causal inference, and sense-seeking agential explanation. And the central benefit he identifies is one that could be real and important even when the content of a rationalization is false: information flow from non-rational to rational processes that can serve to improve reasoning and yield more successful outcomes.
This strikes me as very plausible, and I'd like to suggest generalizing the picture. First, a large part of everyday rationalization is of emotion, not just action: trying to make sense of our own feelings and moods, and those of others. On influential views of emotion and its evolution, affective states serve to synthesize multiple dimensions of information relevant to deciding how to act, predicting how others will act, and guiding one's behavior appropriately in context (Nesse & Ellsworth Reference Nesse and Ellsworth2009). Yet because affective states bundle information together, making best use of them may require parsing, interpreting, and interrogating. If I feel uneasy in interacting with someone, why? Is it me or him or the interaction? What might this tell me about the situation, how I should react, how he's likely to react? Is the feeling appropriate, or am I over-reacting or prejudiced? And so on. Cushman's models of rational action (Fig. 1a in the target article) and rationalization (Fig. 1b in the target article) include beliefs and desires but not the agent's affective states as such, yet affect may be a primary source of the information needed to explain oneself and others, or to predict or evaluate what might happen next (cf. Thornton & Tamir Reference Thornton and Tamir2017). At the same time, seeking to bring an emotional reaction into a rationalizing frame can contribute to emotion regulation, focusing or changing felt affect, and dampening dysfunctional effects (Wilson & Gilbert Reference Wilson and Gilbert2003). On the evolutionary side, it is not hard to imagine that capacities to read and regulate one's own emotions or moods by trying to make sense of them, or to seek to make sense of the emotions or moods of others as a way of informing one's interactions with them, would be, even if not perfectly reliable, capacities that facilitated everything from episodic cooperation to long-term relationships, and from parent-child relations and language learning to effectiveness in competition and warfare.
Second, taking affect into account might suggest a shift away from thinking of rationalization as exchanging information between “non-rational” and “rational” processes. Cushman uses “rational” both to designate “reasoning” processes and to pick out cognitive and choice processes that use evidence to form representations of probability and reward information and guide choice by maximizing expected utility. What, then, about mental processes that reliably perform the latter functions but are not forms of controlled reasoning? A large body of research suggests that the affective system contains such information-gathering processes and representational structures and guides decision making via comparisons of expected value (modulo the special role of risk; see Stauffer et al. Reference Stauffer, Lak and Schultz2014). Yet affect makes no appearance in Cushman's schematic models of our “multiple adaptive processes” (Fig. 2a) or of the operation of rationalization (Fig. 2b). No doubt the schemata in Figure 2 are meant to simplify these processes, but rationalization can be thought of as important for information sharing in the mind between controlled, deliberative “reasoning” processes and other kinds of processes, without thinking that this is primarily a matter of putting a rational mind in touch with non-rational processes. Emotions themselves are readily spoken of as more or less rational, and even habit may draw upon expected-value representations and weighings involving the “rational” affective system, rather than constituting a distinct “non-rational” system in itself (cf. Dayan & Berridge Reference Dayan and Berridge2014; Smith & Graybiel Reference Smith and Graybiel2016). This kind of mental integration seems even more likely for norm compliance. Greater mental integration can help with a problem posed by Hume that Cushman rightly considers at the end: if “rational” is restricted to “reasoning,” then being “rational” won't always make us reasonable – it matters what our intrinsic values are, and affect is the currency of value. Neuroscientists have advocated abandoning the bifurcation of “cognitive” versus “affective” regions (Pessoa Reference Pessoa2008). If rationalization includes fathoming and using the evaluative information that might be present in affect, as Hume (Reference Hume, Selby-Bigge and Nidditch1738/1978) attempted in the Treatise, then we might find reasonableness in a less bifurcated picture of “rational” versus “non-rational” processes as well.
Fiery Cushman brings a welcome synoptic vision to psychological processes and outcomes often considered in isolation from one another. By asking the evolutionary why? question, he brings attention to the need to understand how processes frequently associated with error could nonetheless have such a central and pervasive role in the human mind. This evolutionary perspective forces one to seek deep answers, since one must identify features of the mind that could plausibly be the result of natural selection, and so must look behind what plausibly are relatively recent appearances on the psychic landscape, such as rationalization's role in the enhancement of self-image or advertising of one's norm acceptance, which seem to require conceptual and linguistic capacities that probably appeared relatively late in evolutionary history. Moreover, the fact that rationalization as Cushman understands it – the post hoc construction of sense-making belief-desire explanations for one's own or others’ behavior – figures in such a variety of mental phenomena suggests that it is the result of core features of mental architecture. Further, evolutionary explanation requires that one be able to point to benefits sufficiently large and reliable to favor such an architecture, despite its costs.
Therefore, at the heart of Cushman's account are processes that must have long evolutionary histories: reinforcement learning, causal modeling including inverse causal inference, and sense-seeking agential explanation. And the central benefit he identifies is one that could be real and important even when the content of a rationalization is false: information flow from non-rational to rational processes that can serve to improve reasoning and yield more successful outcomes.
This strikes me as very plausible, and I'd like to suggest generalizing the picture. First, a large part of everyday rationalization is of emotion, not just action: trying to make sense of our own feelings and moods, and those of others. On influential views of emotion and its evolution, affective states serve to synthesize multiple dimensions of information relevant to deciding how to act, predicting how others will act, and guiding one's behavior appropriately in context (Nesse & Ellsworth Reference Nesse and Ellsworth2009). Yet because affective states bundle information together, making best use of them may require parsing, interpreting, and interrogating. If I feel uneasy in interacting with someone, why? Is it me or him or the interaction? What might this tell me about the situation, how I should react, how he's likely to react? Is the feeling appropriate, or am I over-reacting or prejudiced? And so on. Cushman's models of rational action (Fig. 1a in the target article) and rationalization (Fig. 1b in the target article) include beliefs and desires but not the agent's affective states as such, yet affect may be a primary source of the information needed to explain oneself and others, or to predict or evaluate what might happen next (cf. Thornton & Tamir Reference Thornton and Tamir2017). At the same time, seeking to bring an emotional reaction into a rationalizing frame can contribute to emotion regulation, focusing or changing felt affect, and dampening dysfunctional effects (Wilson & Gilbert Reference Wilson and Gilbert2003). On the evolutionary side, it is not hard to imagine that capacities to read and regulate one's own emotions or moods by trying to make sense of them, or to seek to make sense of the emotions or moods of others as a way of informing one's interactions with them, would be, even if not perfectly reliable, capacities that facilitated everything from episodic cooperation to long-term relationships, and from parent-child relations and language learning to effectiveness in competition and warfare.
Second, taking affect into account might suggest a shift away from thinking of rationalization as exchanging information between “non-rational” and “rational” processes. Cushman uses “rational” both to designate “reasoning” processes and to pick out cognitive and choice processes that use evidence to form representations of probability and reward information and guide choice by maximizing expected utility. What, then, about mental processes that reliably perform the latter functions but are not forms of controlled reasoning? A large body of research suggests that the affective system contains such information-gathering processes and representational structures and guides decision making via comparisons of expected value (modulo the special role of risk; see Stauffer et al. Reference Stauffer, Lak and Schultz2014). Yet affect makes no appearance in Cushman's schematic models of our “multiple adaptive processes” (Fig. 2a) or of the operation of rationalization (Fig. 2b). No doubt the schemata in Figure 2 are meant to simplify these processes, but rationalization can be thought of as important for information sharing in the mind between controlled, deliberative “reasoning” processes and other kinds of processes, without thinking that this is primarily a matter of putting a rational mind in touch with non-rational processes. Emotions themselves are readily spoken of as more or less rational, and even habit may draw upon expected-value representations and weighings involving the “rational” affective system, rather than constituting a distinct “non-rational” system in itself (cf. Dayan & Berridge Reference Dayan and Berridge2014; Smith & Graybiel Reference Smith and Graybiel2016). This kind of mental integration seems even more likely for norm compliance. Greater mental integration can help with a problem posed by Hume that Cushman rightly considers at the end: if “rational” is restricted to “reasoning,” then being “rational” won't always make us reasonable – it matters what our intrinsic values are, and affect is the currency of value. Neuroscientists have advocated abandoning the bifurcation of “cognitive” versus “affective” regions (Pessoa Reference Pessoa2008). If rationalization includes fathoming and using the evaluative information that might be present in affect, as Hume (Reference Hume, Selby-Bigge and Nidditch1738/1978) attempted in the Treatise, then we might find reasonableness in a less bifurcated picture of “rational” versus “non-rational” processes as well.