ABSTRACT

Some evolutionary psychologists have argued that people do not have a natural logical ability to use content-independent inference rules (Cosmides & Tooby, 1992, 1994; Gigerenzer, 2000; Gigerenzer, Todd, & the ABC Research Group, 1999; Tooby & Cosmides, 1992). According to them, the mind is like a Swiss army knife: it contains many different contentdependent modules for processing domain speci®c information about adaptive problems, but no general-purpose ``blade''. (See Cosmides & Tooby, 1994, on the knife analogy, and Over, 2003, for problems with it.) This view of the mind has been called the massive modularity hypothesis (Samuels, 1998; Sperber, 1994). The argument for this is that ``content free, general-purpose systems could not evolve, could not manage their own reproduction, and would be grossly inef®cient and easily outcompeted'' by domain speci®c mechanisms (Tooby & Cosmides, 1992, p. 112). Cosmides and Tooby (1994) criticise the dual architecture theory of Fodor (1983), according to which the mind contains both speci®c modules and a contentindependent capacity for logical inference (see also Fodor, 2000). Others have carried massive modularity to the still further extreme of implying that the content-independent rules of logic and probability theory are surplus ``baggage'' for solving adaptive problems. Todd and Gigerenzer (1999, p. 365) claim that ``thought processes that forgo the baggage of logic can solve real-world adaptive problems quickly and well'' (also note this theme in Gigerenzer, 2000). The mind is described as an ``adaptive toolbox'' that only implements content-dependent ``fast and frugal'' heuristics (Gigerenzer et al., 1999). But even more than that, logic apparently should never be used as the norm to assess human rationality. (See Gigerenzer, 2006; this extreme view is implied in many of the papers of Gigerenzer, 2000.)

One in¯uential argument that evolutionary psychologists have used in order to try to support massive modularity concerns probability and frequency. They have argued that people do not follow content-independent rules to solve certain word problems about frequencies (Brase, Cosmides & Tooby, 1998; Cosmides & Tooby, 1996; Gigerenzer & Hoffrage, 1995, 1999). Instead, people rely on an adaptive domain speci®c ability to think about the frequencies of real-world events. I shall argue, in reply, that these

frequency word problems are in fact solved by following contentindependent rules (see also Over, 2003, and Sloman & Over, 2003). To solve these, people cannot ``forgo the baggage of logic'', but must follow elementary logical rules, which are the normative and descriptive basis of the solution. I will begin, however, with more general points about logical inference and what psychological experiments have shown about it.