Neuroconstructivism, Vol. 1: How the Brain Constructs Cognition (Mareschal et al. Reference Mareschal, Johnson, Sirois, Spratling, Thomas and Westermann2007) makes a persuasive argument that we should use computational modeling tools to attempt to understand the processes of change in those neural structures that underlie cognitive development. The authors argue that modeling can help researchers understand the constraints acting on experience (and thus neural development) at multiple levels, with particular emphasis placed on the constraints of the nervous system and the body. The consideration of these multiple levels of constraint represents an important step forward, as it provides a more formal conceptualization not just of the causal processes of change in cognitive development, but also of how cognitive processes in the mature adult are constrained by developmental trajectories (see also Westermann et al. Reference Westermann, Sirois, Shultz and Mareschal2006).

This approach pays homage to both cognitive neuroscience and to Piaget. Cognitive development is considered on a neural and functional neural systems level (and indeed is explained as arising from causal interactions between these systems). In homage to Piaget, the authors' approach asserts that children proactively construct their own representations by means of internally generated learning processes. However, when we consider the nature of our internal representations of the world – the contents of cognition – there is a key point of divergence with Piaget. For while Piaget (1952) was concerned with the question of how infants and children construct objective (or in Mareschal et al.'s terms, complete) representations of the world, the authors of Neuroconstructivism argue that partial representations form the substrate of cognition, both during early development and in mature adults. Mareschal et al. describe partial representations as “just sufficient to allow successful behavior,” (p. 16) and they justify this view of cognition on the basis that it is a by-product of learning to achieve a behavioural goal within a system possessing particular neural and embodiment constraints.

However, Mareschal and colleagues' formulation of constraints is somewhat limited. In particular, the authors neglect to consider the critical role of multiple interacting sensory modalities in development. While the authors present a number of models of learning that take different modalities as their inputs (visual, auditory, etc.), none of the models process inputs from more than one sensory modality at any given time. It should be remembered that the interaction of multiple sensory modalities was a key aspect of Piaget's constructionism. More specifically, Piaget argued that the initial steps toward objectivity in infants' representations of their world were achieved in part through the integration of separate, modality-specific schemas (Piaget Reference Piaget and Cook1953). In the following observation, Piaget notes what he termed the “organisation of reciprocal assimilation,” when, for the first time, his 4-month-old daughter watches herself grasping an object:

Piaget goes on to describe this as a new kind of schema within which infants can grasp what they see, and see what they grasp: a reciprocal relationship. For Lucienne, the object is no longer the thing of looking, nor the thing of grasping, but an entity existing in a more objective representation. Regardless of the reliability of Piaget's observations, his conception of a reciprocal schema shows us how two initially modality-specific (partial) representations of the environment (in this case, two unimodal schemas for acting on the rattle) can be enriched almost fortuitously by virtue of the spatial and temporal coincidence of the modalities. The resulting multisensory representation of the object, while still partial, is enriched and more objective since it goes beyond the immediate behavioural goal of either of the actions being performed.

If Mareschal et al.'s approach is to fully describe the nature of emergent cognitive abilities, then the multisensory constraints of the developing organism will certainly need further consideration. In particular, computational and robotic models of both modality-specific constraints (in neural machinery, and the interface of that machinery with the environment via the body) and also the constraints acting on the integration of these systems will help to further our understanding of how representations of the environment emerge.

Piaget's concept of a reciprocal schema prompts us to consider research demonstrating a gradual integration of action systems in early life, as this may point toward the development of mental representations or schemas that go beyond the simple satisfaction of an immediate goal. While much multisensory infancy research points toward the early unification of the senses enabling infants to detect a large array of amodal relations (Bahrick et al. Reference Bahrick, Lickliter and Flom2004; Gibson Reference Gibson1969; Lewkowicz Reference Lewkowicz2000; Phillips-Silver & Trainor Reference Phillips-Silver and Trainor2005), research addressing crossmodal spatial orienting responses in infants appears to indicate a more prolonged period of integration (e.g., Bremner et al. Reference Bremner, Mareschal, Lloyd-Fox and Spence2008; Neil et al. Reference Neil, Chee-Ruiter, Scheier, Lewkowicz and Shimojo2006; von Hofsten Reference von Hofsten2004). It is possible that the gradual integration of such multisensory action systems may, by going beyond the immediate behavioural goals of initially separate unimodal schemas, enrich the representations that human infants acquire in the first months of life, and indeed determine the nature of cognition in adulthood.