The Adaptive Value of Decision-Making Speed, Frugality, and Accuracy

Gigerenzer et al. claim that, from an evolutionary point of view, there are (at least) three features of a decision-making mechanism that have important consequences for its adaptive value (see also Sober and Wilson Reference Sober and Wilson1998, chap. 10; Todd Reference Todd2001, 53–54). First, the faster an organism can make decisions, the fitter it is (Laland Reference Laland2001, 244; Carruthers Reference Carruthers2006). Second, the more valuable resources (like energy) the organism gets to save for other uses, the fitter it is. Third, the more accurate the organism is in adjusting its behavior to the state of the world, the fitter it is. Of course, since there are often interdependencies among the three factors—so that improvements in one dimension lead to losses in others—it frequently will be impossible to maximize all three factors simultaneously; instead, joint maximization will be all that is in the offing. What is clear, though, is that, from an evolutionary point of view, all three factors ought to be taken into account in evaluating a decision-making mechanism, as they all affect its adaptive value.

Given this, Gigerenzer et al. claim that RCT seems to make the mistake of overlooking exactly this point: the theory does not appear to pay any heed to factors other than accuracy.Footnote ⁷ However, as just noted, this may well be highly maladaptive: there may be many cases where sacrificing accuracy for speed or frugality may be greatly beneficial (Gigerenzer Reference Gigerenzer2001; Sadrieh et al. Reference Sadrieh, Güth, Hammerstein, Harnad, Hoffrage, Kuon, Munier, Todd, Warglien and Weber2001). By contrast, simple heuristics are tailored to allow for fast and frugal decision making: they place strong restrictions on the amount of information they consider and on how thoroughly that information is processed; in turn, this cuts down the time and energy it takes to compute the right decision (Sadrieh et al. Reference Sadrieh, Güth, Hammerstein, Harnad, Hoffrage, Kuon, Munier, Todd, Warglien and Weber2001; Todd Reference Todd2001).

This impression of adaptive advantage may be further strengthened when it is noted that many simple heuristics can be just as accurate—or even more so—than RCT in many situations (Gigerenzer Reference Gigerenzer2001; Sadrieh et al. Reference Sadrieh, Güth, Hammerstein, Harnad, Hoffrage, Kuon, Munier, Todd, Warglien and Weber2001). This is primarily due to the fact that, sometimes, only very few cues are needed to obtain good evidence about the state of reality, so it is unnecessary to acquire a lot of information when trying to make an accurate decision. This is important, as it suggests not only that simple heuristics may be faster and more frugal than RCT in getting the agent to act but also that the losses in accuracy that they entail may be quite small, or even zero (a sort of dominance argument in favor of simple heuristics).Footnote ⁸

However, there are several problems with this argument that make it ultimately unconvincing. In the main, these problems center on the fact that we lack the information necessary to assess whether the considerations Gigerenzer et al. put forward really speak in favor of simple heuristics, or against them. This becomes clear from noting that, in order to assess whether simple heuristics really come out on top by being able to compromise accuracy for decision-making speed and frugality, a rich understanding of the environment in which we evolved is necessary (as is also accepted by Gigerenzer et al.; see, e.g., Sadrieh et al. Reference Sadrieh, Güth, Hammerstein, Harnad, Hoffrage, Kuon, Munier, Todd, Warglien and Weber2001, 87; Todd Reference Todd2001, 67–68).

To see why this is so, note that Gigerenzer et al. seem to think that the fitness of a decision-making mechanism is determined by an equation of the following sort:

where S, F, and A are the values of decision-making speed, frugality, and accuracy of the decision-making mechanism in question, and a, b, and c are their relative adaptive importances. Given this, what we need to know in order to determine whether simple heuristics are more adaptive than an RCT-based decision-making mechanism is, first, the values of the parameters (a, b, c) and, second, the values of the variables (S, F, A) for the two decision-making mechanisms in the environments in which we evolved. However, as yet, we have absolutely no idea about what these values might be.

In particular, at this point, we are in no position to rule out values for (a, b, c) and (S, F, A) that entail that RCT was more—not less—adaptive than simple heuristics (Sadrieh et al. Reference Sadrieh, Güth, Hammerstein, Harnad, Hoffrage, Kuon, Munier, Todd, Warglien and Weber2001, 88). In the main, this is because we do not know the exact environmental conditions under which we evolved, which decision problems we were facing, when these problems became particularly pressing, how we went on to solve them, what the consequences were of the various solutions we tried out, and so on (for a case study of the complexities involved in this sort of project, see Mithen Reference Mithen1990; see also Todd Reference Todd2001, 67–68; Richardson Reference Richardson2007). In short, as it stands, we lack any of the information that is required to get Gigerenzer et al.'s argument off the ground.

Moreover, what we do know about the conditions under which we evolved seems to make the relative adaptiveness of RCT actually quite plausible. In particular, it is by now widely accepted that a key role in our cognitive evolution was being played by our social environment (Byrne and Whiten Reference Byrne and Whiten1988; Sterelny Reference Sterelny2003, 208–9). However, pace Gigerenzer and Selten (Reference Gigerenzer and Selten2001, 10), it seems that simple heuristics cannot handle these kinds of environments very well (Stanovich and West Reference Stanovich, West and Over2003; Buller Reference Buller2005, 158–60). This comes out easily from noting that when it comes to predicting the decisions of other organisms, there are many problems of strategic interaction to take into account. In turn, this need for strategic interaction quite plausibly entails that relying on simple heuristics is highly maladaptive: the resultant behavior is likely to be very rigid and, thus, easily exploited.Footnote ⁹

For these reasons, Gigerenzer et al.'s first evolutionary argument is unable to provide much in the way of support for a move away from RCT. As it turns out, however, their second evolutionary argument does not fare much better—and for similar reasons.

Massive Modularity

The second evolutionary argument to which Gigerenzer et al. appeal in order to support simple heuristics rests on the idea that the latter fit much better with recent advances in evolutionary psychology than RCT does—in particular, they connect much more easily to the ‘massive modularity thesis’ (MMT; Cosmides and Tooby Reference Cosmides, Tooby, Cosmides, Tooby and Barkow1992; Todd Reference Todd2001; Samuels et al. Reference Samuels, Stich, Faucher, Niiniluoto, Sintonen and Wolenski2004). This thesis states that the mind consists of a collection of special-purpose modules that are adapted to deal with particular environmental problems (Cosmides and Tooby Reference Cosmides, Tooby, Cosmides, Tooby and Barkow1992; Pinker Reference Pinker1997; Carruthers Reference Carruthers2006). Two specific reasons are commonly cited for why this kind of massively modular mind is likely to have evolved; one focuses on the benefits of this massively modular structure itself, and the other focuses on the nature of a cognitive adaptation in general and its implications for what our minds are like.

The first reason centers on the (alleged) fact that having a massively modular mind is itself adaptive: it is said that only this kind of mind allows for easy, quick, and biologically cheap alterations in our cognitive traits (Samuels Reference Samuels1998; Carruthers Reference Carruthers2006, 12–29). The idea here is that it is highly adaptive to be able to change parts of our minds without having to redesign the whole: since any change in some part of an organism is likely to be costly, the fewer changes one has to make, the better.

The second reason centers on the (alleged) fact that all that natural selection can design are specific solutions to the specific adaptive problems that were faced (not general solutions to general problems). Hence, it is argued that our minds ought to be expected to consist of a large number of specialized adaptations, and not one domain-general reasoning mechanism (Cosmides and Tooby Reference Cosmides, Tooby, Cosmides, Tooby and Barkow1992).

In this way, Gigerenzer et al. can claim that there is further evolutionary support for the theory of simple heuristics—namely, via the evolutionary support for the MMT (see, e.g., Hammerstein Reference Hammerstein2001; Todd Reference Todd2001). Once again, though, there are major problems with this argument. To see this, consider the two aspects of this argument in turn.Footnote ¹⁰

First, upon closer consideration, the more specific version of that argument turns out to be fallacious: there is no reason—either theoretical or empirical—to think that independent alterability is generally adaptive: in particular, for traits that are functionally integrated, evolutionary integration may be highly adaptive (Schulz Reference Schulz2008). For example, if successful locomotion requires the integration of visual and auditory information, then it may be highly adaptive to ensure that changes in one's visual processing are correlated with changes in one's auditory processing. This is important, as it means that any reason for thinking that it was adaptive for our decision-making mechanism to be organized in a massively modular way must rest on the specifics of that mechanism. However, we again lack any of the required empirical information to assess this situation: we do not know if it was adaptive to be able to change each of our decision-making abilities individually or whether they are so highly integrated that a domain-general approach was more beneficial. For this reason, until the key premises of the first version of the massive modularity-based argument become further substantiated, this version cannot be seen to be very compelling.

Alas, the second, more general version of that argument turns out to be not much better. The major problem here is that it is not clear that it is true that all specific adaptive problems require specific solutions (Sterelny Reference Sterelny2003, chap. 10; Buller Reference Buller2005, 144–46). In particular, it is not clear why a novel solution has to be found for every novel problem—sometimes old solutions may be put to use also in different and novel circumstances (Dawkins Reference Dawkins1986). Importantly, whether specific solutions were needed in the case of our decision-making abilities is highly questionable: we lack exact knowledge about what the adaptive problems were that we had to face (see also Sadrieh et al. Reference Sadrieh, Güth, Hammerstein, Harnad, Hoffrage, Kuon, Munier, Todd, Warglien and Weber2001, 88), and what we do know suggests that, in many cases, our adaptive problems were instances of a general problem—namely, that of dealing with a complex social environment. Without more information on this score, therefore, we are not in a position to determine if our adaptive problems were numerous and variegated enough to warrant the evolution of specific solutions or if they were merely ‘variations on a theme’ that could be handled well by a nonspecialized decision mechanism (see also Buller Reference Buller2005, 145–47).

On the whole, it thus seems clear that Gigerenzer et al.'s evolutionary arguments cannot (yet) play a major role in the debate surrounding the plausibility of RCT. However, before this conclusion can be fully accepted, it is useful to consider an objection that might come to mind at this point. This objection notes that Gigerenzer et al. may be using their evolutionary arguments merely to suggest novel hypotheses about how our decision-making mechanism might work (which are then further investigated using traditional social psychological methods). If this is the case, then the above, evidence-focused criticisms are beside the point: the evolutionary considerations here are used only heuristically, not evidentially.Footnote ¹¹

However, it is not clear that Gigerenzer et al.'s arguments, in fact, can live up to this objection. As noted earlier, much of Gigerenzer et al.'s work is an extension of the work of Simon (Reference Simon1957), which questioned RCT in essentially the same way—without, however, appealing to evolutionary theory. This is further supported by the fact that many of the particular heuristics Gigerenzer et al. are putting forward—like ‘Take the First’—do not seem to be directly inspired by rigorous evolutionary thinking at all but by casual observations of current everyday life. For this reason, it is not clear that much that is novel has in fact been suggested by Gigerenzer et al.'s evolutionary perspective—and, hence, appeals to a heuristic reading of their arguments ring somewhat hollow.