Reasoning, rationality and adaptation

In the last two decades, evolutionary psychologists, such as Leda Cosmides and John Tooby, have provided both theoretical considerations and empirical data in order to show that the most common subjects’ responses in the classical reasoning experiments should be considered as normatively appropriate after having taken into account the evolutionary conditions in which the cognitive mechanisms providing those responses have evolved (e.g., Cosmides 1989; Cosmides & Tooby 1996; Fiddick et al. 2000). Their analysis is intended to show that supporters of the Irrationality thesis, such as researches in the heuristics and biases tradition, have not demonstrated that people are deeply irrational. As Samuels and his colleagues (2002: 244) put it, according to evolutionary psychologists,

“human reasoning is not subserved by ‘fast and dirty’ heuristics but by ‘elegant machines’

that were designed and refined by natural selection over millions of years”. As a result, in their view, people’s reasoning is the result of the processing performed by the cognitive modules that were evolved to solve the evolutionary problems our ancestors were frequently faced with, and so they draw normatively appropriate conclusions only when the information of a reasoning problem is given in the appropriate format and the reasoning problem itself resembles a problem from the environment of evolutionary adaptation. On this view, good reasoning and decision making should lead to successful outcomes but, at the same time, from an evolutionary perspective, most evolutionary psychologists define success as linked to inclusive fitness, that is, in terms of the number of genes one transmits to subsequent generations as a result of direct and indirect reproduction (Hamilton 1963),

arguing that behaviour is rational when it increases inclusive fitness. In other words, the evolutionary thesis about human rationality supported by evolutionary psychologists is based on the assumption that humans, we as well as our ancestors, tend to behave in ways that increase their inclusive fitness. Since evaluations of reasoning performances are made by looking at the contributions of their outcomes to inclusive fitness, human rationality is not assessed against short but rather long-term criteria. The evidence that humans are rational amounts to the fact that they have evolved in such a way as to be able to follow

“Darwinian” algorithms which enable them to handle recurrent adaptive problems confronted during the Pleistocene. As is pointed out by Samuels and his colleagues (2002:

256), evolutionary psychologists implicitly assume the distinction between the actual domain and the proper domain for a mental module, as proposed by Dan Sperber (1994).

Sperber has illustrated how a given mental module can be expected to work in terms of its scope of application. First of all, there is the proper domain of a mental module. The proper domain amounts to the class of situations that recall the specific adaptive problem which a mental module evolved to solve. That is, the proper domain for a mental module is “all the information that it is the module’s biological function to process” (Sperber 1994: 52).

However, many domain-specific cognitive mechanisms can be triggered in situations that are not actually within their proper domains. This is what Sperber calls the actual domain for a mental module. In Sperber’s words, the actual domain for a given mental module amounts to “all the information in the organism’s environment that may (once processed by perceptual modules, and possibly by other conceptual modules) satisfy the module’s input conditions” (Sperber 1994: 51-52). Evolutionary psychologists assume that, when evaluating reasoning performances, assessments should be always relativized to the proper domain of the cognitive mechanisms that have produced the performances. This means that

human rationality is defined by reference to the specialized cognitive mechanisms that have evolved for solving specific domains of reasoning (see also Samuels et al. 2002: 256).

Consider the implications that such an evolutionary view has for the interpretation of the responses in two well-known reasoning tasks which I have introduced in the first chapter (see Chapter 1, Section 3.1). First, Cosmides and Tooby claim to possess evidence for the existence of a domain-specific cognitive mechanism that activates in experiments based on the selection task (e.g. Cosmides 1989; Cosmides & Tooby 1992). That task has been studied using different kinds of conditional statements, some of which elicit the correct response in most subjects’ reasoning performances. In particular, subjects improve their performances when the task they are given has more realistic content, such as in the following case:

If a person is drinking beer [p], then that person must be over 19 years of age [not-q].

(Griggs & Cox 1982: 415)

In this modified version of the selection task, subjects are required to turn over the cards which might show whether the conditional rule cited above has been violated. When such a task is given to subjects, most of them make the right choice, that is, they select the p card, which has printed on its visible side “drinking beer”, and the not-q card, which has printed on its visible side “15 years old” (Griggs & Cox 1982: 414-417). In order to explain why only certain modified versions of the selection task elicit the correct response, Cosmides and Tooby have proposed the Cheater-Detection Hypothesis (Cosmides 1989; Cosmides &

Tooby 1992).³ According to their hypothesis, humans possess a module devoted to solving problems about social exchange, which they characterize as “adaptive cooperation between two or more individuals for mutual benefit […]” (Cosmides 1989: 193). In particular, a

3 More generally, the drinking age problem has been characterized as a deontic version of the selection task, as opposed to the standard indicative version of it (for a discussion on deontic versions of the selection task see, e.g., Manktelow & Over 1991).

recurrent adaptive problem our evolutionary ancestors would face with was to detect whether their conspecifics were trying to cheat them when involved in social exchanges. In Cosmides and Tooby’s view, we should have inherited from our evolutionary ancestors a module devoted to detecting people cheating on social contracts. This domain-specific module evaluates whether a social contract is upheld or violated, and in the selection task it prompts subjects to turn the cards that are most informative for determining this. Roughly speaking, a social exchange is characterized as an exchange between two individuals that is regulated by the following conditional rule “If you take the benefit A, then you must pay cost B”. A violation of such an arrangement, that is cheating, takes place when someone gets a benefit without paying the cost, which is a violation of a social agreement as opposed to a violation of logic. According to the Cheater-Detection Hypothesis, this domain specific module can be applied to the task concerning the drinking-age rule but not to the standard abstract version, as the drinking-age rule can be seen as a social contract and so violating it can be regarded as cheating. If the conditional statement involves a social contract subjects will react in a predetermined way, such as in the case of the drinking-age rule; the implicit reference to a social contract affects what goal they form. In particular, selection framed in social-contract terms where cheating can be seen as a risk will activate the cheater-detection module within its proper domain, which produces the correct response. Selection tasks that do not have this form, such as the standard abstract version, would not activate the module.⁴

Another analysis is concerned with the format in which statistical information is presented. Turning to probability judgments, Cosmides and Tooby have proposed the so-called Frequentist Hypothesis, according to which “some of our inductive reasoning mechanisms do embody aspects of a calculus of probability, but they are designed to take frequency information as input and produce frequencies as output” (Cosmides & Tooby

4 This hypothesis, however, remains highly controversial. For alternative explanations see, e.g., Cheng &

1996: 3). In other words, they hold that in the environment of evolutionary adaptation, statistical information was available to our evolutionary ancestors as frequencies rather than as probabilities, such as “four out of every ten times you go into this forest you meet a lion”

rather than “the probability of meeting a lion going into this forest is 0.4” (Cosmides &

Tooby 1996). This should explain why human subjects usually are not so good at reasoning with probabilities of single events, but make good performances when reasoning about frequencies. As a matter of fact, most empirical studies on probability judgments have demonstrated that a great majority of participants respond correctly when statistical information is presented as frequency (see Fiedler 1988; Hertwig & Gigerenzer 1999). In the “frequentist” version of the Linda Problem devised by Klaus Fiedler (1988) (for the standard version see Chapter 1, Section 3.1.2), for example, the description of Linda is followed by a question like this:

To how many out of 100 people who are like Linda do the following statements apply?

Linda is a bank teller.

Linda is a bank teller and active in the feminist movement.

While in the standard version of the task 85% of the subjects gave the wrong answer, only few subjects (about 13%) did the same in the “frequentist” version (Fiedler 1988; see also Cosmides & Tooby 1996: 17-21).