4 An alternative to localization

Luckily, we can do without PFC, at least for the purposes of explaining conscious perception, while still maintaining many of the other, more compelling aspects of the GNW model. One central component that the GNW model shares with several other proposals about the neural correlates of consciousness (for example Melloni & Singer 2010; Tononi 2004; von der Malsburg 1997) is the concept of global integration of information. In light of the modular organization of the brain, a mechanism is required that brings information together such that an integrated, coherent percept can be formed. One attractive neural mechanism that can account for this requirement is neuronal synchrony (Bosman et al. 2012; Bressler et al. 1993; Salazar 2012). As has been discussed in greater detail elsewhere (Melloni & Singer 2010; Melloni this collection; Singer this collection), areas can be brought into direct contact with each other by synchronizing their neuronal activity, for example by phase alignment of neuronal oscillations, thus binding them into a functionally coherent assembly that forms a distributed representation of perceptual content. This self-organizing process can flexibly create and dissolve assemblies on top of a fixed anatomical architecture and does so without the need for anatomical convergence or broadcasting bottlenecks.

For example, imagine that a subject is confronted with two superimposed, transparent surfaces of moving dots (Figure 1). The dots on one surface are green and move to the left, and the dots on the other surface are red and move to the right. The two colors of dots are represented in a brain area coding for color, while the two motion directions are represented in an area coding for motion. For the subject to become conscious of the two surfaces, the neurons coding for green in the color area would need to synchronize their activity with the neurons coding for motion to the left in the motion area, and the neurons coding for red would need to synchronize with the neurons coding for motion to the right (Figure 1a). If the dots change direction, a new state of synchronization needs to be established, this time linking neurons coding for green with neurons coding for motion to the right, and neurons coding for red with neurons coding for motion to the left (Figure 1b). Hence, while the contents of the subject’s experience are determined by the specific neuronal assemblies being active, conscious perception would be an emergent property of the state of synchronization. Recent tracing and modelling work in the macaque brain suggests that the kind of direct connectivity required to flexibly instantiate numerous, high-dimensional combinations of features is indeed afforded by high-density, reciprocal connections between brain areas (Markov et al. 2013).

Figure 1: Neuronal synchrony binds distributed neurons into coherent assemblies, giving rise to conscious experience. Consider an experiment in which the subject is confronted with two superimposed, transparent surfaces of moving dots, as shown in the first column. (a) The dots on one surface are green and move to the left, and the dots on the other surface are red and move to the right. The two colors of dots are represented in a brain area coding for color, while the two motion directions are represented in an area coding for motion. If the neurons coding for green in the color area synchronize with the neurons coding for motion to the left in the motion area, and the neurons coding for red synchronize with the neurons coding for motion to the right, the two surfaces are consciously perceived. (b) A change in experience does not require a change in activity levels within areas, but a change of which neurons are synchronized. The opposite percept of (a) arises if neurons coding for green are synchronized with neurons coding for motion to the right, and if neurons coding for red are synchronized with neurons coding for motion to the left. Such content-specific synchronization between neurons has for example been observed in working-memory tasks in monkeys (Salazar et al. 2012). (c) Even when activity is synchronized within the color or motion area, respectively, a coherent conscious percept does not arise unless the areas are synchronized with each other.

Theoretical considerations and empirical evidence further suggest that the critical feature differentiating conscious from unconscious processing is the spatial scale at which information is exchanged: while the integration of information in local modules, even in higher sensory areas (Sterzer et al. 2008), does not give rise to conscious experience by itself, large-scale integration over long distances does (Del Cul et al. 2007; Melloni et al. 2007). In the example of the transparent surfaces, this implies that even when activity is synchronized within the color or motion area, respectively, a coherent conscious percept cannot arise unless the areas are synchronized with each other (Figure 1c). Taken together, functional connectivity between distributed brain areas (i.e., connectivity that does not imply that one drives or controls the other) is an attractive alternative to localization in PFC as a candidate for the neural correlate of consciousness.

Coming back to the MVPA technique, this proposal makes a clear prediction that could be tested using decoding algorithms: specifically, one would predict that the large-scale connectivity patterns between brain regions for different percepts should differ, even if only slightly, for different conscious contents, and hence that conscious content should be decodable from them. This may well be the case in light of the fact that, at a much coarser scale, the neural correlates of auditory and visual awareness involve different brain networks (Eriksson et al. 2007), and that higher decoding accuracy for a subject’s percept can be achieved if the joint activity patterns of several areas are considered instead of only singular patterns (Pessoa & Padmala 2007). The MVPA technique could in principle also be applied to other neuroimaging techniques that afford higher time resolution, such as electro- or magnetoencephalography or electrocorticography, thus potentially resolving the problem that arises because the representational carriers of perceptual content are highly dynamic and thus require a time-resolved analysis.