For decades, the field of judgment and decision making (under the constant impulse coming from economics) has been struggling over how to overcome optimality as a default framework for decision making. With the benefit of this experience, we comment on Rahnev & Denison's (R&D) article, which may be traversing through similar hurdles in the field of perceptual decision making. Symptomatic of the fact that R&D (and the whole field of perceptual decision making) are in the early stages of a struggle with optimality is that they are inevitably led to call perceptual behavior that violates “optimality” suboptimal. By doing so, they compel themselves within the strictures of the optimality/suboptimality dichotomy, making their subsequent plea to overcome optimality – although repeatedly stated – methodologically unconvincing. R&D's plea for the construction of a standard observer model closely resembles well-known attempts to descriptively amend optimality frameworks in the face of evidence of optimality violations (e.g., Kahneman & Tversky Reference Kahneman and Tversky1979), maintaining, however, optimality as a normative benchmark, even if only for diagnostic purposes. On a more radical reading, R&D would reject normativity in perceptual decision making altogether, in favor of a purely descriptive account of both optimal and suboptimal behavior. We argue that both of these ways would methodologically result in a dead end. This would mainly be attributable to R&D's reliance on the hidden assumption that conflates optimality and normativity (i.e., the assumption that considers optimality as the sole possible normative benchmark for perceptual decision making). In this commentary, we advance the idea that “bounded rationality,” and particularly the notion of “satisficing” (Simon Reference Simon1955; Reference Simon1956), are able to altogether overcome optimality by (1) providing a benchmark that rejects optimality but salvages normativity in perceptual decision making and (2) proposing a more parsimonious and robust explanation of perceptual behavior.
With regard to the first point, R&D's contention that “[bounded rationality models] still place the greatest emphasis on the optimality of the decision rule” (sect. 5.1, para. 1), while mentioning Herbert Simon (Reference Simon1955) and Gigerenzer and Selten (Reference Gigerenzer and Selten2001) in support of their statement, is particularly striking. Simon (Reference Simon1956) was resolutely against any form of optimality, so much that he founded the notion of “bounded rationality” on a completely new criterion called “satisficing” (a neologism conflating “satisfy” and “suffice”). As Simon (Reference Simon1996) stated, “Many … have argued that the gap between satisfactory and best is of no great importance, hence the unrealism of the assumption that the actors optimize does not matter; others, including myself, believe that it does matter, and matters a great deal” (p. 29). Given the naturally limited availability of time, information, and computational capacity to make decisions in real-world environments, agents’ decision rules cannot optimize, but rather obey a criterion of satisfaction, operationally meaning that satisfaction is achieved when a certain threshold, or “aspiration level,” is reached. More recently, Gigerenzer et al. (Reference Gigerenzer and Todd1999), building upon Simon's framework, have emphasized its ecological traits, maintaining that a decision rule is rational “to the degree that it is adapted to the structure of an environment” (p. 13). In both Simon's and Gigerenzer's frameworks, “adaptation” is the keyword, as it provides an alternative normative framework to decision making (e.g., Hands Reference Hands2014). Notably, this adaptation framework rejects the very idea that adaptive criteria, such as satisficing, could be reduced to some form of optimization (typically, optimization under constraints) (Gigerenzer Reference Gigerenzer, Augier and March2004). R&D's misconception of Simon's and Gigerenzer's theories – particularly when they say that “influential theories [of bounded rationality] postulate that evolutionary pressures produced heuristic but useful, rather than normative, behavior” (sect. 5.3, para. 3) – has to be traced back to a failure to figure out that bounded rationality's and satisficing's normative content lies in adaptation. This is most unfortunate, as bounded rationality's normative core would greatly help perceptual decision making overcome optimality in any residual form, without abdicating to normativity.
With regard to the second point, satisficing could also provide a more parsimonious and robust explanatory framework for perceptual behavior with respect to optimality and suboptimality. To see how, we specifically consider the “diffusion model,” typically used to explain the speed/accuracy tradeoff in perceptual decisions. As it is commonly described, “the diffusion model assumes that decisions are made by a noisy process that accumulates information over time from a starting point toward one of two response criteria or boundaries.… When one of the boundaries is reached, a response is initiated” (Ratcliff & McKoon Reference Ratcliff and McKoon2008, p. 875). It is surprising how closely this model's mechanism resembles satisficing models of decision making introduced by Simon (Reference Simon1955). The descriptive consistency of these two classes of models can be used, we suggest, to parsimoniously explain the puzzling evidence that some subjects set decision thresholds optimally while others only suboptimally in the same task (Bogacz et al. Reference Bogacz, Hu, Holmes and Cohen2010). In principle, evidence of individual differences in threshold setting is consistent with the idea that individual variability falls within a common adaptive interval. In this adaptive interpretation, the main question addressed by diffusion models should not be whether thresholds are optimal or suboptimal, but whether they are adapted or not to a given perceptual task. In this line of argument, although ecology is mentioned as an explanation of optimality in the speed/accuracy tradeoff (e.g., Bogacz Reference Bogacz2007), ecological arguments seem to be missing in R&D's discussion of suboptimality. As certain perceptual tasks are distinctly oriented to accuracy and others distinctly oriented to speed, ecological arguments dictate whether speed and accuracy should be treated together or separately (Todd & Gigerenzer Reference Todd and Gigerenzer2003, p. 151). The usual explanation that first assumes the speed/accuracy tradeoff and then describes either speed-oriented or accuracy-oriented behavior as “limiting cases” can be unnecessarily complex from an ecological point of view. More generally, as analysis of optimality/suboptimality typically explains perceptual behavior in tasks that are relatively simple, it can say little regarding more complex tasks, such as those considered by Simon (Bogacz et al. Reference Bogacz, Hu, Holmes and Cohen2010, p. 888). In these latter cases, there are ecological reasons to argue that satisficing procedures may provide a robust explanation for a wider range of perceptual tasks (e.g., Martignon & Hoffrage Reference Martignon and Hoffrage2002).
For decades, the field of judgment and decision making (under the constant impulse coming from economics) has been struggling over how to overcome optimality as a default framework for decision making. With the benefit of this experience, we comment on Rahnev & Denison's (R&D) article, which may be traversing through similar hurdles in the field of perceptual decision making. Symptomatic of the fact that R&D (and the whole field of perceptual decision making) are in the early stages of a struggle with optimality is that they are inevitably led to call perceptual behavior that violates “optimality” suboptimal. By doing so, they compel themselves within the strictures of the optimality/suboptimality dichotomy, making their subsequent plea to overcome optimality – although repeatedly stated – methodologically unconvincing. R&D's plea for the construction of a standard observer model closely resembles well-known attempts to descriptively amend optimality frameworks in the face of evidence of optimality violations (e.g., Kahneman & Tversky Reference Kahneman and Tversky1979), maintaining, however, optimality as a normative benchmark, even if only for diagnostic purposes. On a more radical reading, R&D would reject normativity in perceptual decision making altogether, in favor of a purely descriptive account of both optimal and suboptimal behavior. We argue that both of these ways would methodologically result in a dead end. This would mainly be attributable to R&D's reliance on the hidden assumption that conflates optimality and normativity (i.e., the assumption that considers optimality as the sole possible normative benchmark for perceptual decision making). In this commentary, we advance the idea that “bounded rationality,” and particularly the notion of “satisficing” (Simon Reference Simon1955; Reference Simon1956), are able to altogether overcome optimality by (1) providing a benchmark that rejects optimality but salvages normativity in perceptual decision making and (2) proposing a more parsimonious and robust explanation of perceptual behavior.
With regard to the first point, R&D's contention that “[bounded rationality models] still place the greatest emphasis on the optimality of the decision rule” (sect. 5.1, para. 1), while mentioning Herbert Simon (Reference Simon1955) and Gigerenzer and Selten (Reference Gigerenzer and Selten2001) in support of their statement, is particularly striking. Simon (Reference Simon1956) was resolutely against any form of optimality, so much that he founded the notion of “bounded rationality” on a completely new criterion called “satisficing” (a neologism conflating “satisfy” and “suffice”). As Simon (Reference Simon1996) stated, “Many … have argued that the gap between satisfactory and best is of no great importance, hence the unrealism of the assumption that the actors optimize does not matter; others, including myself, believe that it does matter, and matters a great deal” (p. 29). Given the naturally limited availability of time, information, and computational capacity to make decisions in real-world environments, agents’ decision rules cannot optimize, but rather obey a criterion of satisfaction, operationally meaning that satisfaction is achieved when a certain threshold, or “aspiration level,” is reached. More recently, Gigerenzer et al. (Reference Gigerenzer and Todd1999), building upon Simon's framework, have emphasized its ecological traits, maintaining that a decision rule is rational “to the degree that it is adapted to the structure of an environment” (p. 13). In both Simon's and Gigerenzer's frameworks, “adaptation” is the keyword, as it provides an alternative normative framework to decision making (e.g., Hands Reference Hands2014). Notably, this adaptation framework rejects the very idea that adaptive criteria, such as satisficing, could be reduced to some form of optimization (typically, optimization under constraints) (Gigerenzer Reference Gigerenzer, Augier and March2004). R&D's misconception of Simon's and Gigerenzer's theories – particularly when they say that “influential theories [of bounded rationality] postulate that evolutionary pressures produced heuristic but useful, rather than normative, behavior” (sect. 5.3, para. 3) – has to be traced back to a failure to figure out that bounded rationality's and satisficing's normative content lies in adaptation. This is most unfortunate, as bounded rationality's normative core would greatly help perceptual decision making overcome optimality in any residual form, without abdicating to normativity.
With regard to the second point, satisficing could also provide a more parsimonious and robust explanatory framework for perceptual behavior with respect to optimality and suboptimality. To see how, we specifically consider the “diffusion model,” typically used to explain the speed/accuracy tradeoff in perceptual decisions. As it is commonly described, “the diffusion model assumes that decisions are made by a noisy process that accumulates information over time from a starting point toward one of two response criteria or boundaries.… When one of the boundaries is reached, a response is initiated” (Ratcliff & McKoon Reference Ratcliff and McKoon2008, p. 875). It is surprising how closely this model's mechanism resembles satisficing models of decision making introduced by Simon (Reference Simon1955). The descriptive consistency of these two classes of models can be used, we suggest, to parsimoniously explain the puzzling evidence that some subjects set decision thresholds optimally while others only suboptimally in the same task (Bogacz et al. Reference Bogacz, Hu, Holmes and Cohen2010). In principle, evidence of individual differences in threshold setting is consistent with the idea that individual variability falls within a common adaptive interval. In this adaptive interpretation, the main question addressed by diffusion models should not be whether thresholds are optimal or suboptimal, but whether they are adapted or not to a given perceptual task. In this line of argument, although ecology is mentioned as an explanation of optimality in the speed/accuracy tradeoff (e.g., Bogacz Reference Bogacz2007), ecological arguments seem to be missing in R&D's discussion of suboptimality. As certain perceptual tasks are distinctly oriented to accuracy and others distinctly oriented to speed, ecological arguments dictate whether speed and accuracy should be treated together or separately (Todd & Gigerenzer Reference Todd and Gigerenzer2003, p. 151). The usual explanation that first assumes the speed/accuracy tradeoff and then describes either speed-oriented or accuracy-oriented behavior as “limiting cases” can be unnecessarily complex from an ecological point of view. More generally, as analysis of optimality/suboptimality typically explains perceptual behavior in tasks that are relatively simple, it can say little regarding more complex tasks, such as those considered by Simon (Bogacz et al. Reference Bogacz, Hu, Holmes and Cohen2010, p. 888). In these latter cases, there are ecological reasons to argue that satisficing procedures may provide a robust explanation for a wider range of perceptual tasks (e.g., Martignon & Hoffrage Reference Martignon and Hoffrage2002).