The target article succeeds in convincing the reader that prediction mechanisms are pervasive in the central nervous system (CNS), and that prediction is involved even in the early stages of sensory processing. Indeed, these results are not surprising for informed readers, as discussions on this topic have been prevalent for decades (Butz et al. Reference Butz, Sigaud, Pezzulo and Baldassarre2007). Nonetheless, after the neat survey by Nijhawan, it is likely – and indeed, highly desirable – that most of the controversies about the predictive character of visual processing will cease.

At the same time, we feel that the wealth of psychophysical results discussed by the author – which may constitute a strong basis for predicting behavioral performances in this kind of tasks – fails to further significantly our understanding of the phenomenon of “perception” in the brain: Without a suitable proposal about how anticipation is implemented in the CNS, we can hardly assume we have won one more point in this direction. One must acknowledge that the target article argues forcefully for the existence of predictive capabilities in visual processing. However, what we need now in order to explain the behaviors observed in psychophysics experiments is to shift from the level of psychophysical analysis to the level of neuroscientific modeling, and analyze the biological mechanisms responsible for these behaviors. “Understanding visual prediction” (target article, Abstract) requires identifying parts of the target nervous system, studying their behavior, their interconnections, and their functional role with respect to the visual processing capacity under investigation (Craver and Darden Reference Craver, Darden, Machamer, Grush and McLaughlin2001; Cummins Reference Cummins1983; Glennan Reference Glennan2005). In the lack of sufficiently precise theoretical models, we can hardly evaluate the biological plausibility of some of the author's claims, especially the highly speculative claim, discussed in the final sections of the article, that “the goal of any type of compensation is to produce coordination between the time-varying neural representations that exist in parts of the CNS separated by significant delays” (sect. 9.1, para. 1). Similarly, in the lack of theoretical models, we can hardly appreciate the relationship between the hypotheses proposed by the author and other models available in the literature: Notably, what is the distinctive character of the “internal model for visual prediction (IMVP)” hypothesis discussed in the target article (see sect. 7.2) with respect to the internal-model-based hypothesis discussed by Wolpert et al. (Reference Wolpert, Ghahramani and Jordan1995; quoted in the article) which involves sensory anticipation as well?

This is not to deny that conjectures on the involvement of some neural structures in anticipation-based capacities may play important roles in the discovery of anticipation mechanisms. As has been pointed out (Machamer et al. Reference Machamer, Darden and Craver2000), the formulation of mechanistic models of biological behaviors often involves the formulation of mechanism sketches, which provide partial or underspecified accounts of the target mechanism. Often one starts with the isolation of some parts of the target system which are supposed to play key functional roles with respect to the capacity under investigation, and proceeds by identifying, in a stepwise fashion, other functional components (and their material instantiation) which are needed to obtain a “working” mechanism. One must acknowledge that Nijhawan, albeit being chiefly concerned with attesting the “existence” of visual prediction, does occasionally formulate sketches of the neural mechanisms that could produce the behaviors under investigation.

However, we surmise that some of the author's sketches are based on too simplistic representations of the architecture of the visual cortex, which undermines the possibility of extending the proposed hypothesis to the study of other predictive phenomena in the CNS. In fact, the article focuses on psychophysical experiments in which only simple stimuli are involved. If we consider more complex classes of stimuli, then the strictly retinotopic organization of the visual cortex – which plays a key role in the model – is no longer sustainable. Instead, according to a wealth of physiological data reported in Serre et al. (Reference Serre, Kouh, Cadieu, Knoblich, Kreiman and Poggio2005, and references therein), the main feature of the cortical architecture is a hierarchical organization, which aims (in a series of stages) to increase the invariance to object transformations and to tune to more specific features. It is acknowledged that one of the main functions of the ventral stream pathway is to achieve an exquisite trade-off between selectivity and invariance, which results in an improved capability to accomplish readily many recognition tasks. As a possible side effect of such a trade-off, it is very likely that strict information about the position of the objects becomes less significant (in the higher levels of the cortical hierarchy). If we assume that predictive phenomena also play a role in this context – and we do – then it is not clear how extrapolation can be of use. In our opinion, different phenomena should be considered, such as, for example, the effect of the temporal and spatial integration of the neural signal from the lower layers of the hierarchy to the higher ones. That could have induced a kind of “characteristic time” for each layer insofar as different classes of objects would be affected by different amounts of delays, and consequently, distinct anticipation mechanisms.

In conclusion, we feel the author argues convincingly that predictive mechanisms must be implemented in the CNS at various levels. This claim is consistent with a wealth of literature on sensory processing and sensory-motor coordination. However, as far as scientific understanding of these phenomena is concerned, empirically testable theoretical models of visual prediction, at a neurophysiological level, are needed. In our opinion, the author's suggestions in this direction are to be evaluated on the basis of more realistic experimental methodologies and by taking into account the hierarchical character of the human visual cortex.