Burkart et al.'s premise is that cognitive abilities can be supported either by the evolution of “primary modules” (sect. 1.2.3, para. 2; domain-specific adaptations to specific environmental challenges), or by the expansion of domain-general intelligence (G). If this premise were true, then the current empirical research, based largely on principal component analyses (PCAs), would be more consistent with the idea that a large portion of cognition in several species is explained by G rather than by collections of primary modules. Reviewing this empirical literature, the authors admit the results are somewhat ambiguous. Nevertheless, they predict we will find stronger evidence for the evolution of G in the future, because the data seem largely inconsistent with the primary modular perspective.

Here, we argue that the strict dichotomy of primary module versus G is misleading: There are occasional evolutionary discontinuities in neurobiological architectures that support a range of cognitive abilities, which are neither domain general nor modular adaptations for specific environmental challenges.

Our target example is the arcuate fasciculus (AF), which is a neural fiber tract enabling a direct connection between temporal cortex (including auditory cortex) and inferior frontal gyrus (involved in cognitive control) (Catani et al. Reference Catani, Jones and Ffytche2005). This tract, exceptionally well developed in humans in comparison with other primates (Rilling et al. Reference Rilling, Glasser, Preuss, Ma, Zhao, Hu and Behrens2008), is a neurobiological evolutionary discontinuity. By neurally binding the regions responsible for auditory processing and cognitive control, this new architectural feature greatly enhanced (1) the working memory for verbal information (vWM) – quite poor in nonhuman primates (Plakke et al. Reference Plakke, Hwang and Romanski2015; Scott et al. Reference Scott, Mishkin and Yin2012); and (2) the capacity to process sequences (Dehaene et al. Reference Dehaene, Meyniel, Wacongne, Wang and Pallier2015).

This peculiar connectivity pattern seems to be a crucial prerequisite for the evolution of multiple abilities relying on hierarchical sequential structure (e.g., language, music, and complex action) (Fadiga et al. Reference Fadiga, Craighero and D'Ausilio2009; Fitch & Martins Reference Fitch and Martins2014). However, improvements in vWM and sequence processing do not necessary permeate other (nonsequential) cognitive domains, thus not allowing any interpretation in terms of modules or G. For instance: (1) Some nonhuman primates (e.g., chimpanzees) seem to show spatial WM superior to that of humans (Inoue & Matsuzawa Reference Inoue and Matsuzawa2007), and (2) although the capacity to represent social hierarchies seems to be within the range of nonhuman primate cognition (Seyfarth & Cheney Reference Seyfarth and Cheney2014), and the ability to process spatial hierarchies is conserved among nonhuman mammals (Geva-Sagiv et al. Reference Geva-Sagiv, Las, Yovel and Ulanovsky2015), the capacity to process sequential structures nonetheless remains limited in these clades.

Another source of evidence for this specialization comes from neuroimaging. Although the processing of sequential hierarchies activates the inferior frontal gyrus (a region strongly connected with the AF) (Fadiga et al. Reference Fadiga, Craighero and D'Ausilio2009; Fitch & Martins Reference Fitch and Martins2014), the same is not true for nonsequential hierarchies in the visual, spatial, and social domains (Aminoff et al. Reference Aminoff, Gronau and Bar2007; Kumaran et al. Reference Kumaran, Melo and Düzel2012; Martins et al. Reference Martins, Fischmeister, Puig-Waldmüller, Oh, Geißler, Robinson, Fitch and Beisteiner2014). Instead, the latter group of hierarchies seems to be represented by a domain-general episodic memory system.

This cognitive mosaic argues against a simple gradual expansion of G. When performing a PCA, including individuals of different primate species, the emergence of the human AF (and enhanced vWM) would be more easily classified as multidomain or multipurpose cognitive ability, but neither as domain-specific (because it increases the capacity within a range of domains) nor as domain general (because these improvements are specific to sequential but not to nonsequential domains).

In sum, we suspect that the research program advanced by Burkart et al. is designed to distinguish only between modules and G, leaving aside other possible interpretations that would fit better with the available data (e.g., Anderson Reference Anderson2016; Karmiloff-Smith Reference Karmiloff-Smith2015). In our opinion, a third way between modules and G will give a more suitable account for clade-specific discontinuities (grounded on neurobiological architectural changes), which would fit better the statistical models. These discontinuities offer a great opportunity to capture capacities that are neither gradual expansions of G nor specific modular adaptations to specific environmental problems. Therefore, they are required to overcome intrinsic limitations of current models, theoretically improving them and achieving a more realistic account of the evolution of cognition across different species.