Theories are necessary for stimulating scientific inquiry. We have mountains of data, but weak theories. Therefore, the theoretical attempt by Jung & Haier (J&H) must be applauded. The search for converging evidence from a wide variety of empirical studies about the biological substrate for the psychological construct of intelligence is welcome.

However, I found some problems seeing the convergence J&H support. Only a very small number of discrete brain areas approach 50% of convergence across published studies employing the same neuroimaging strategy. First, structural studies nominate 32 brain areas, but only Brodmann areas (BAs) 10 and 39 to 40 50% of convergence. Second, PET studies nominate 22 brain areas, but only BAs 18 to 19 and 46 to 47 enjoy 50% of convergence. Third, fMRI studies summarily nominate 26 brain areas, but only BAs 6, 7, 9, 19, and 40 reach 50% of convergence. Furthermore, structural and functional results are not consistent. Of those brain areas approaching 50% of convergence, no one overlaps across neuroimaging research strategies. These empirical facts are not a good basis to support an integration theory of intelligence. Admittedly, this is not the authors' fault, but rather, a result derived from the heterogeneity of the available evidence.

My view is that (a) the proposed theory may or may not be correct, and (b) its likelihood cannot be supported (or rejected) by the collected data. I do think the theory is promising, but expressly designed empirical studies to appropriately test it are strongly required.

The P-FIT model postulates that, at a first stage, temporal and occipital brain areas process sensory information: BAs 18, 19, and 37 for visual material and BA 22 for acoustic processing. The second stage implicates integration and abstraction of this information by parietal BAs 7 and 39 to 40. Further, these parietal areas interact with the frontal lobes, which serve for problem evaluation. Frontal BAs 6, 9, 10, and 45 to 47 are highlighted by the P-FIT model. The anterior cingulate (BA 32) is then invoked for response selection and inhibition of alternative responses. Finally, white matter plays a critical role for a reliable circulation of information across these brain processing units.

Nevertheless, J&H do believe that not all these brain areas are really germane for human intelligence. Actually, they predict that only the discrete brain regions of the dorsolateral prefrontal cortex (BAs 9, 45 to 47) and the parietal cortex (BAs 7 and 40) will affect intelligent performance. However, examining the collected structural and functional evidence, it is difficult to conclude that their view is supported.

First, the strongest structural data points to frontal (BA 10) and parietal (BAs 39, 40) areas, relatively consistent with the authors' prediction. However, the PET data are consistent only with the authors' prediction regarding frontal BAs 46 and 47, not regarding the occipital BAs 18 and 19. The fMRI data are consistent with the authors' prediction with respect to frontal BA 9 and parietal BAs 7 and 40. But BAs 6 and 19 are not within the authors' framework. Therefore, only a minority of the identified brain areas overlaps with the central brain areas proposed by the P-FIT model. Structural and functional data identify a great number of discrete brain areas (see Figs. 2, 3, and 4 of the target article) and only a small number overlap with the P-FIT model.

Why is the evidence so heterogeneous? Several are the tentative alternatives, such as age, sex, and representativeness of the analyzed samples; but I think the key resides in the measurement of the construct of human intelligence.

Intelligence can be defined as a “general mental capacity … involving the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly, and learn from experience” (Gottfredson Reference Gottfredson1997, p. 13). However, this definition neglects the theoretically important distinction between “intelligence in general” and “general intelligence” (Colom et al. Reference Colom, Abad, Garcia and Juan-Espinosa2002; Reference Colom, Jung and Haier2006a; Reference Colom, Jung and Haier2006b; Jensen Reference Jensen1998). Typical psychometric intelligence scores, such as Full Scale IQ, measure “intelligence in general” comprising an array of cognitive abilities and skills in addition to general intelligence, or the g factor. We already know that (a) the g factor is the main component of the intelligence construct (Lubinski Reference Lubinski2004) and (b) not all intelligence tests measure the g factor of intelligence to the same degree (Jensen Reference Jensen1998).

Intelligence tests can be classified according to the degree to which they involve the g factor. The g-loading for test X relates to its average correlation with all the remaining tests in a comprehensive test battery: the higher its average correlation, the larger its g-loading. From a theoretical standpoint, a high g-loading for test X can only result from the fact that it shares a large amount of mental processes with the other tests in the battery (Arend et al. Reference Arend, Colom, Botella, Contreras, Rubio and Santacreu2003). Therefore, a test with a perfect g-loading should comprise most of the mental processes germane to the general factor of intelligence (g). Available measures of human intelligence confound g with other cognitive abilities and skills (Colom et al. Reference Colom, Abad, Garcia and Juan-Espinosa2002).

Neuroimaging studies on human intelligence must refine the way this construct is measured. Different measures will result in different structural and functional correlates. We do need to know the causes underlying these discrepancies, and J&H are sensitive to this central issue. Actually, we have shown that as the g-loading of a given intelligence measure increases, more widespread discrete brain areas become involved (Colom et al. Reference Colom, Jung and Haier2006a). Importantly, this increased recruitment is not related to the superficial characteristics of the measurements, like their verbal or nonverbal nature.

In conclusion, not every intelligence measure taps the intelligence constructs in the same way. Heterogeneity is an inevitable result derived from the lack of consensus regarding the question of the most appropriate avenue to get measures of the complex construct of human intelligence. If the g factor is its core component, but available measures tap this component to quite different degrees, then obtained data will be hardly comparable.