As I point out in my contribution to this volume, cognitive integration should be understood as a thesis about the enculturation of human cognition. It is a thesis about how phylogenetically earlier forms of cognition are built upon by more recent cultural innovations (e. g., systems of symbolic representation). This results in a multi-layered system with heterogeneous components, dynamically interwoven into a co-operative of processes and states an integrated cognitive system (henceforth ICS). The co-ordination dynamics of the system are, at least in part, understood in terms of the physical dynamics of brain–body–niche interactions in real-time; however, they are also to be understood in terms of NPP that govern and determine those interactions (over time). NPP operate at both social/population levels and individual, even sub-personal, levels. They originate as patterns of activity spread out over a population of agents; consequently they should be understood primarily as public systems of activity and/or representation that are susceptible to innovative alteration, expansion, and even contraction over time. They are transmitted horizontally across generational groups and vertically from one generation to the [3]next. At the individual level they are acquired, most often by learning and training, and they manifest themselves as changes in the ways in which individuals think, but also the ways that they act (intentionally) and the ways in which they interact with other members of their social group(s) and the local environment. NPP, therefore, operate at different levels (groups and individuals) and over different time-scales (intergenerationally and in the here-and-now).
Given this, it is clear that What? Now[4] processes that reduce prediction errors on their own could not drive the innovation of NPP; nor could they determine the properties of NPP on their own. Less obviously, I would argue, they do not drive the acquisition of NPP, because scaffolded learning requires both a physically and temporally-structured learning environment and the capacity for functional changes to cortical circuitry to be driven by the structured learning environment. The mechanism of acquisition includes both neural and environmental processes working in concert and over long periods of ontogenetic time. What? Now processes may help us to understand the here-and-now processes by which we enact NPP; they may even tell us something about the neural mechanisms for learning and plasticity; but we should be wary of making prediction and error minimization the driving factors behind the why and how of enculturation.
Fabry’s commentary focuses on the neural level, functioning in real-time, where the primary aim is to give a mechanistic account of how cognitive capacities can be transformed by learning and training in rich socio-cultural niches. Rather than looking at the origin of ICS in cultural inheritance, phenotypic plasticity, and learning driven plasticity, Fabry argues that a version of the PP framework can provide the neural mechanisms by which ICS are (partly) constructed. My contribution to this volume focused primarily on the origin of ICS in the recent cultural evolution of NPP and then explored how mathematical practices could be learnt and how this process of learning could drive functional changes to circuitry in the brain. Consequently, the CI/ENC framework pursues the phylogenetic and ontogenetic basis of the larger brain–body–niche nexus. What, though, of the neural mechanisms of transformation?
I don’t agree with Fabry’s starting premise that CI/ENC lacks a mechanism of transformation: the mechanism of transformation is learning-driven plasticity (LDP) with neural redeployment in a scaffolded learning environment. The fundamental plasticity of the brain explains the nature of neural transformations and why the brain is open to scaffolded learning driven by the environment. (E)PP does not have the resources to explain redeployment (this is a theme I take up in the next section). Why would it, since PP is not a framework for explaining redeployment. It might be the case that PP fits with a certain conception of scaffolded learning⎯such as path-dependent learning, but I have yet to see a thorough working-through of the details and it's not clear to me that all scaffolded learning should be reduced to a predictive form of path-dependent learning.
Fabry claims that a dynamical systems approach to integration “does not spell out the mutual influence that neuronal and extra-cranial bodily components have over each other” (2015, p. 3). The EPP approach is supposed to fill in the details here. However, I suspect that this judgement is made a little too quickly, because the dynamical systems description of brain–body–niche interactions is in one sense a higher-level description of those interactions. The dynamical interactions are described as being part of a larger system comprising brain, body, and niche. We can zoom in and focus upon the dynamics of brain or body, but we shouldn’t confuse the dynamics of the brain for the dynamics of the overall system. I have highlighted and outlined the neural dynamics required for enculturation in a number of places. For example, in the account of body schema dynamics and in the case of NPP for symbolic cognition, I have outlined the case for dual component transformations (e. g., Menary 2007, pp. 78–83; 2010; 2013 and 2014). Lets take these two cases in order.
In a now famous series of studies, Maravita & Iriki (2004) studied the bimodal interparietal neurons in trained Japanese macaque brains. These neurons respond both to tactile stimulation on the hand (tactile receptive field) and visual stimuli in the same vicinity as the tactile receptive field (the visual receptive field). The visual receptive field was centred on the hand following it through space. When macaques where trained to use a rake to pull food towards them on a table, the observation that struck Maravita and Iriki was that when the macaques used the rake the receptive fields of the bimodal neurons extended along the axis of the rake, including its head. Iriki’s interpretation of this is that “either the rake was being assimilated into the image of the hand or, alternatively, the image of the hand was extending to incorporate the tool” (Iriki & Sakura 2008, p. 2230). The extension of the body schema (receptive field) to include the tool happened only during active holding; it reduced to just the hand during inactivity. The interesting result of these experiments is that the existing body schema has the latent capacity to extend to incorporate the tool. LDP can be cashed out in terms of functional changes as the result of scaffolded learning even in the case of macaques, let alone the notoriously plastic brains of humans.
Functional changes can be cashed out in terms of neural redeployment and cortical connectivity. Returning to the case of mathematical cognition, inherited systems for numerosity are evolutionary endowments; we can be reasonably sure of this because they are constant across individuals and cultures and they are shared with other species. The numerosity systems are “quick and dirty”; they are approximate and continuous, not discrete and digital. By contrast, discrete mathematical operations exhibit cultural and individual variation; there is a big difference between Roman numerals and Arabic numerals. They are subject to verbal instruction (they actually depend on language); one must learn to count, whereas one does not learn to subitise. Mathematics depends on cultural norms of reasoning (mathematical norms). The ability to perform exact mathematical calculations depends on the public system of representation and its governing norms. We learn the interpretative practices and manipulative practices as a part of a pattern of practices within a mathematics community, and these practices transform what we can do. They are constitutive of our exact calculative abilities. Mathematical practices get under our skins by transforming the way that our existing neural circuitry functions.
The relationship between the evolutionarily earlier system and the recent development of public mathematical systems, norms, and symbols comes down to the redeployment of the cortical territories that are dedicated to evolutionarily older functions by novel cultural artefacts (e. g., representations, tools). The transformation results in new connections between the frontal lobe for number-word recognition and association, the temporal lobe for the visual recognition of number form, and the parietal lobe for the approximate recognition of magnitudes across both left and right hemispheres (Dehaene 1997).
The deeply transformative power of our learning histories in the cognitive niche relates to the development of our capacities for understanding symbolic representations and for physically manipulating inscriptions in public space. In learning to understand symbols, the first transformation involves our sensorimotor abilities for creating and manipulating inscriptions (the transformation of the body schema). This is something we learn to do on the page and in the context of a learning environment, in public space, before we do it in our heads. Our capacities to think have been transformed, but in this instance they are capacities to manipulate inscriptions in public space.
It looks like PP can provide models of some of the fundamental processing principles at work at the sub-personal neural level, but it is not obvious that it would replace LDP and neural redeployment in the mechanism of transformation. However, Fabry may be right and PP may add another string to the bow of our understanding of how the brain exhibits the plasticity required for cognitive transformation. In that case it provides extra explanatory depth to the account of enculturation, but only as part of a much broader explanatory framework.