Ever since early research on the building blocks of visual perception, basic stimulus properties such as line orientation and movement have been shown to be mapped in a graded fashion within early visual cortex (e.g., Hubel & Wiesel Reference Hubel and Wiesel1977). Relatedly, a hallmark of spatial coding within this region is retinotopic mapping, in which the external space is represented continuously within neural tissue. With the emergence of hemodynamic neuroimaging techniques, investigation of the nature of higher-level object recognition mechanisms was accelerated, revealing several subregions within ventral temporal cortex that are dedicated to processing discrete visual categories, such as faces (Kanwisher et al. Reference Kanwisher, McDermott and Chun1997; McCarthy et al. Reference McCarthy, Puce, Gore and Allison1997), places/scenes (Aguirre et al. Reference Aguirre, Zarahn and D'esposito1998; Epstein & Kanwisher Reference Epstein and Kanwisher1998; Maguire et al. Reference Maguire, Frith, Burgess, Donnett and O'Keefe1998), human body parts (e.g., Downing et al. Reference Downing, Jiang, Shuman and Kanwisher2001), and even letters/words (Fiez and Petersen Reference Fiez and Petersen1998; Puce et al. Reference Puce, Allison, Asgari, Gore and McCarthy1996). These regions, or networks of functionally specialized processing units, are typically referred to as domain-specific, as they are uniquely (albeit not exclusively) activated by specific stimulus categories (e.g., Kanwisher Reference Kanwisher2010; Yovel & Kanwisher Reference Yovel and Kanwisher2004) and are tied to rather well-defined anatomical structures (see Grill-Spector & Weiner Reference Grill-Spector and Weiner2014). The discovery of such functionally dedicated “islands,” or modules, within ventral visual cortex served as a window to the understanding of the underlying mechanisms of high-level visual perception, as clearly, we perceive the world as a collection of meaningful, individuated objects grounded within background sceneries, not of arbitrary orientations, colors, or motion directions. Visual categorization, or the ability to detect and recognize objects by representing them in a separable, discrete fashion, is thus one of the key functions of the visual system (Grill-Spector & Weiner Reference Grill-Spector and Weiner2014; Kanwisher Reference Kanwisher2010).

Yet more recent advances have shifted the pendulum back to acknowledging the importance of domain-general processes in visual recognition, by revealing various large-scale mapping principles that may in fact underlie categorical representation along the ventral visual cortex. Thus, for example, one such organizing principle is visual field eccentricity, according to which face- and word-selective regions reside within foveally biased regions (because of their requirement for high-resolution vision), whereas place- and scene-selective regions lie along peripherally biased areas within the visual cortex (because of their dependence on large-scale visual integration [e.g., Hasson et al. Reference Hasson, Levy, Behrmann, Hendler and Malach2002; Levy et al. Reference Levy, Hasson, Avidan, Hendler and Malach2001]). Other organizing principles are real-world object size (e.g., Konkle & Oliva Reference Konkle and Oliva2012), motion/mobility (Chao et al. Reference Chao, Haxby and Martin1999; Huth et al. Reference Huth, Nishimoto, Vu and Gallant2012), animacy (Connolly et al. Reference Connolly, Guntupalli, Gors, Hanke, Halchenko, Wu, Abdi and Haxby2012; Kriegeskorte et al. Reference Kriegeskorte, Mur, Ruff, Kiani, Bodurka, Esteky, Tanaka and Bandettini2008), and semantic meaning (Huth et al. Reference Huth, Nishimoto, Vu and Gallant2012). Importantly, a common feature of most of these organizing principles is their gradient nature, where categories are represented as locations in a continuous space mapped smoothly across the cortical surface. From a developmental perspective, it may well be the case that the domain-specific “modules” specialized to process faces or places have evolved from earlier gradient maps, which encode more basic visual properties such as eccentricity and motion.

Indeed, the extent to which the ability to process “unique” categories is innate, or acquired with experience and with cortical maturation, is still heavily disputed. From an evolutionary perspective, the existence of certain hardwired modules, dedicated to processing a specific type of content, can be easily justified for at least some of the aforementioned domains (e.g., face processing is necessary for infant bonding with caregiver, body-part processing is important to action understanding and social interaction). When discussing face perception, which is at the heart of scientific controversy, studies have demonstrated a genetic basis for face recognition abilities (Shakeshaft & Plomin Reference Shakeshaft and Plomin2015; Wilmer et al. Reference Wilmer, Germine, Chabris, Chatterjee, Williams, Loken, Nakayama and Duchaine2010; Zhu et al. Reference Zhu, Song, Hu, Li, Tian, Zhen, Dong, Kanwisher and Liu2010), along with prioritized processing of face-like configurations during early infancy (Goren et al. Reference Goren, Sarty and Wu1975; Johnson et al. Reference Johnson, Dziurawiec, Ellis and Morton1991). Both types of findings strongly support an innate account of face processing. Other developmental research, however, has revealed a rather complex, protracted profile of face-expertise development, according to which face recognition behavior and its underlying neural mechanisms develop gradually during childhood and do not reach full maturation until early adulthood (e.g., Cohen Kadosh et al. Reference Cohen Kadosh, Cohen Kadosh, Dick and Johnson2010; Gathers et al. Reference Gathers, Bhatt, Corbly, Farley and Joseph2004; Germine et al. Reference Germine, Duchaine and Nakayama2011; O'Hearn et al. Reference O'Hearn, Schroer, Minshew and Luna2010; Scherf et al. Reference Scherf, Behrmann, Humphreys and Luna2007). Furthermore, according to some models, despite their seemingly dissociated nature, face- and word-processing abilities emerge from shared, overlapping neural systems, both relying on high-acuity cortical regions (e.g., Hasson et al. Reference Hasson, Levy, Behrmann, Hendler and Malach2002). It is only with formal education and the acquirement of orthographic symbols, which are represented most dominantly in the left hemisphere, that face-recognition mechanisms are “forced” to become right-lateralized and, consequently, attain partial independence (Behrmann & Plaut Reference Behrmann and Plaut2013; Reference Behrmann and Plaut2014; Dundas et al. Reference Dundas, Plaut and Behrmann2013; Reference Dundas, Plaut and Behrmann2014). According to this view, the cortical mechanisms that support visual cognition at adulthood develop from a common, distributed neural network, rather than from a set of distinct, innate regions that each subserve a particular visual function.

As for the discussion regarding the innateness and independence of a potential “number sense” mechanism, although there are clearly differences in the nature of numerosity and visual categorization abilities, a possible common feature of the two types of processing is their emergence from more basic, continuous representations, which gradually develop into discrete representational entities. As Leibovitch et al. propose in their new theoretical framework, developments such as parsing stimuli one from another (visual individuation), the acquisition of language (counting), and the ability to abstract an exact number of objects from their variant physical properties (abstraction and inhibition), eventually allow one to represent magnitudes in a discrete, numerical fashion. Clearly, much research is still required to fully characterize the developmental profile of continuous and discrete magnitude processing and their interdependent nature (see, e.g., Gabay et al. Reference Gabay, Leibovich, Henik and Gronau2013; Reference Gabay, Kalanthroff, Henik and Gronau2016; Pinel et al. Reference Pinel, Piazza, Le Bihan and Dehaene2004; Walsh Reference Walsh2003), along with their interactions with other domain-general systems (such as cognitive control mechanisms [e.g., Leroux et al. Reference Leroux, Spiess, Zago, Rossi, Lubin, Turbelin, Mazoyer, Tzourio-Mazoyer, Houdé and Joliot2009]). The current model may serve as an overarching framework for such future study.