One of Piaget's better-known tasks is the class-inclusion problem, in which participants are shown, for example, five daisies and three tulips and asked, “Are there more flowers or more daisies?” Although the task's intended normative response is flowers, many (usually younger) participants say that there are more daisies – arguably because they understand flowers to mean flowers-that-are-not-daisies (for a recent review, see Politzer Reference Politzer, Noveck and Sperber2004).
The target article's reference to “intelligence and reasoning” harks back to Piaget's task, because the oft-used conjunction leaves the impression that the two represent a single area of study, when in fact the domain Intelligence is very large (the target article covers chess playing, Go, IQ tests, etc.) and perhaps includes reasoning. We argue that the presentation of Intelligence in this way is infelicitous, much like the option flowers in the class inclusion problem. More specifically, we ask the two following questions: First, are there advantages in studying a large area of cognitive performance over investigating one subpart (i.e., reasoning) alone? Second, do approaches that rely on individual differences (and primarily correlations among subtests) provide conclusions and insights that have greater validity or greater predictability than those drawn from investigations of a subset (i.e., reasoning)? It is our view that the answer to both questions is negative.
For better or worse, investigations into reasoning take a structural approach. The field breaks down reasoning into its component parts, both conceptually and empirically. Thus, reasoning researchers make the distinction between deductive and inductive reasoning (where the former concerns valid conclusions and the latter, conclusions that are more or less probable). Once in the deductive domain, which will remain our example, one then aims to determine the role played by factors such as logical validity, semantic content, development, as well as perceptual or belief biases that affect participants' responses. Generally speaking, reasoning researchers make the assumption that findings are universal. For example, it is generally accepted that Modus Tollens (if p then q; not-q//Therefore, not-p) is more difficult to carry out than Modus Ponens (if p then q; p//Therefore q). The literature on the neuroimaging of reasoning, which is covering the same ground as its cognitive forebears, also aims to depict the way the above factors play out, but with respect to the brain mechanisms or structures that are shown to be responsible for these universal effects. Neuroimaging has not only provided some specific findings that are inaccessible to classic cognitive paradigms, but has informed theory making as well.
For example, Prado and Noveck (Reference Prado and Noveck2006; Reference Prado and Noveck2007) have demonstrated how participants are more prone to errors and are slowed down when features mentioned in a rule mismatch those in a test item (e.g., note how a test item depicting a P-in-a-circle verifies the rule If there is not an H then there is not a square while providing two mismatches). The neuroimaging experiment (Prado & Noveck Reference Prado and Noveck2007) revealed that an increase in mismatching leads to greater activity in the medial prefrontal cortex (PFC) and the right mid-dorsolateral PFC. This indicates that mismatching, rather than negation-interpretation, is likely critical to correct performance. Interestingly, this restricted network is basically the same as the one reported when prior beliefs interfere in evaluating logically valid conclusions (see Goel et al. Reference Goel, Buchel, Frith and Dolan2000; Goel & Dolan Reference Goel and Dolan2003). This is also in line with growing evidence showing that the right lateral PFC is specifically involved in inhibiting a prepotent response (see Aron et al. Reference Aron, Robbins and Poldrack2004) and that its non-activation in children (8–12 years) is linked to less effective attentional control when compared with adults (Bunge et al. Reference Bunge, Dudukovic, Thomason, Vaidya and Gabrieli2002a).
By making distinctions between an attentional control system described above and the parietal-frontal system that is implicated in fundamental logical inferences such as Modus Ponens (Noveck et al. Reference Noveck, Goel and Smith2004), one gets an informed account of the way reasoning is distributed in the frontal and parietal lobes. This dual-system approach also describes some other novel findings. For example, it can explain (a) why solutions to insight tasks, which arguably benefit from having less attentional control, are more accessible to those who have lesions in the frontal cortex (Reverberi et al. Reference Reverberi, Toraldo, D'Agostini and Skrap2005), and (b) why right lateral PFC activity is predictive of successful logical performance (Goel & Dolan Reference Goel and Dolan2003). Providing explanations for such non-intuitive findings is the hallmark of scientific advances. Most importantly, this indicates that correct performance on higher-level tasks has little to do with the better use of normative rules; it has more to do with avoiding biases while using such rules.
In contrast, studies that investigate cerebral correlates of Intelligence point to a large number of regions, but they hardly describe the role played by each. The upshot is that differences among individuals are linked to an entire system that fails to distinguish between its functionally distinguishable parts (e.g., rule access and perceptual integration). In fact, when intelligence research does aim to isolate factors, it largely confirms what is found through more structural approaches (e.g., Gray et al. Reference Gray, Chabris and Braver2003). The worry is that infelicitous analyses of brain function could lead to infelicitous theoretical claims (e.g., about evolution).
Although reasoning research has benefited from descriptions of individual differences (e.g., see Jackson & Griggs Reference Jackson and Griggs1988; Stanovich & West Reference Stanovich and West2000), such descriptions do not amount to a central strategy in investigations of reasoning (at least as far as WAIS-based tests are concerned). There are two viable reasons for this. First, quality of education and economic background are critical for better performance on standardized tests of intelligence (see Georgas et al. Reference Georgas, Weiss, Van de Vijver and Saklofske2003; Shuttleworth-Edwards et al. Reference Shuttleworth-Edwards, Kemp, Rust, Muirhead, Hartman and Radloff2004). This makes measures of intelligence unstable across populations. Second, as Georgas et al. (Reference Georgas, Weiss, Van de Vijver and Saklofske2003) report, of the 11 subtests that account for performance on the WISC-III, the two subtests that imply reasoning (and not necessarily deductive reasoning) – Mazes and Picture Completion – are among the least predictive of variance across the 12 large populations (and 15,999 people) studied. In other words, reasoning per se has a limited impact on standardized tests of intelligence.
