Although time delay poses a difficult issue for visual motion perception in particular, there is nevertheless the benefit to allow large amounts of computation to take place unconsciously. Presumably, evolution must have set optimal delay lengths that are short enough to avoid too much delay, but long enough to allow sufficient amounts of computation. If visual delay is within such optimal lengths, compensation might be unnecessary. The key issue, then, is to determine the magnitude of delay inherent in the visual system. As mentioned by Nijhawan, the perceptual lag of a moving object due to neural delay can be calculated as v×t _max, where v is object velocity and t _max is the maximum cumulative delay in the vision pathway before conscious percepts emerge. The magnitude of delay is therefore determined by t _max. This boils down to this question: To what extent can the visual process proceed in the visual hierarchy without awareness? The farther the unconscious visual process proceeds, the larger is the perceptual lag that occurs.

Because humans are not aware of the delay compensation process, it is tempting to speculate that the brain areas subserving it might be unrelated to awareness. Indeed, neural activity at the level of photoreceptors is not sufficient to yield visual awareness, and some researchers further argue that later stages after the retina, such as primary visual cortex (V1), are not involved in consciousness. This logic is problematic, however, and can be illustrated more clearly when we take a step back to examine a more general problem in visual perception – what Hermann von Helmholtz called unconscious inference (i.e., before percepts emerge, huge amounts of computational processes take place without consciousness). A key notion here is that unconscious inference does not imply that the neural substrates are unrelated to consciousness. For example, in the binding problem, a distributed hierarchical network in the brain processes different features of an object. The neural substrates that correctly select and integrate separate neural representations of features to form coherent object representations are believed to comprise both parietal and frontal cortex (Treisman Reference Treisman1996). Can we thus infer that activity in parietal and frontal cortex is unrelated to consciousness? Apparently this is not the case (Lin & He, in press).

Yet, the notion that observers are not directly aware of neural activity in the optic nerve, the lateral geniculate nucleus (LGN), and even V1 has been popular among many researchers (e.g., sect. 3 of the target article), especially after the appearance of the influential paper by Crick and Koch in 1995. Although this idea (Crick & Koch Reference Crick and Koch1995) was initially based on behavioral studies that show equal adaptation strengths for both visible and invisible conditions (e.g., He et al. Reference He, Cavanagh and Intriligator1996), recent evidence demonstrates, instead, that V1 is directly involved in visual awareness (for a review, see Lin & He, in press). Monkey single-unit recordings, human electroencephalogram (EEG) recordings, and brain imaging converge to reveal robust awareness-dependent modulations in V1; neural events in V1 are actually attenuated in response to suppressed (vs. non-suppressed) visual stimuli. Convergent psychophysical evidence also comes from binocular rivalry, the phenomenon that when each of the two eyes views dissimilar images, the two images rival for visibility, with one temporarily dominating perception for several seconds before being replaced in dominance by the other. With adaptation, a recent study shows that rivalry suppression weakens the strengths of aftereffect for low-contrast adaptors (Blake et al. Reference Blake, Tadin, Sobel, Raissian and Chong2006), which is believed to transpire in V1. It appears that although unconscious inference per se is beyond consciousness, neural activity in V1 is actually correlated with consciousness.

Where, then, does consciousness enter into the information processing sequence? How can one distinguish conscious processing from unconscious processing in the brain? It is important to know which neural substrates of unconscious inference are conscious and which are unconscious, as measured by correlations between neural activity and conscious experiences. This appears extremely difficult, given the unclear mechanisms of how the brain computes unconscious inference. In addition, it is almost impossible to isolate the inference processes (e.g., feature binding), since we only know that there are such processes after we are conscious of the objects. Moreover, although many studies focus on brain activation differences between conscious and unconscious experiences, it is an oversimplification to classify certain areas as conscious or unconscious based on such comparison. For example, a recent study shows that although V1 activity is closely correlated with conscious experience, even a considerable difference in cortical activation in many visual cortical areas does not necessarily lead to different conscious experiences (Jiang et al. Reference Jiang, Zhou and He2007). It therefore appears that no single area, or even network, is sufficient for consciousness. A more dynamic approach to distinguish consciousness from unconsciousness is to go beyond isolated activity of certain brain areas, and, instead, focus on how different brain areas interact with each other. Lamme and Roelfsema (Reference Lamme and Roelfsema2000) suggest that when information first enters the early visual cortex in an feedforward sweep, it is unconscious; when such information is attentively grouped through recurrent processing (by means of horizontal connections and feedforward/feedback projections), it gives rise to consciousness. This might provide important clues as to what give rise to consciousness.

In sum, although the neural correlates of consciousness are still elusive, it is now well documented that V1 is directly related to awareness (Lin & He, in press). This has important implications for the magnitude of time delay by constraining t _max (the maximum cumulative delay in the vision pathway before conscious percepts emerge) to less than 72 msec on average in V1 neurons (Lamme & Roelfsema Reference Lamme and Roelfsema2000). Such a short delay might be within optimal delay lengths to allow sufficient computations to take place, and might even make it unnecessary to compensate for the delay.