So why do Gestalt grouping and segregation bear such a close relation to consciousness? From a neural perspective, they differ from most other functions in that they depend on interactions between neurons at rather large distances. For example, for a neuron to “know” whether it sits on the figure or the background of the stimulus in figure 5, information has to travel over a distance of about 20 millimetres in the visual cortex.[33] Moreover, the modulations of neural activity that accompany this “knowing” depend on the incremental push-pull interactions between horizontal and feedback connections (Lamme & Roelfsema 2000; Roelfsema et al. 2002; Roelfsema 2006). These require quite extensive processing steps, given that the contextual Gestalt effects typically manifest themselves at long latency.
Intuitively, seeing an illusion like the Kanizsa triangle, or the contextual shifts in brightness or colour perception discussed above, also seems to depend on “long range” interactions: information travels over large distances in the visual field. But distance travelled over the visual field does not always equal distance travelled in the brain. These phenomena may depend on fairly hardwired and feedforward mechanisms, and their neural correlates typically have relatively short temporal latencies (Von der Heydt et al. 1984). Seemingly, these phenomena tap into mechanisms that have high ecological relevance to the visual system, and are hence solved in a few processing steps, using dedicated feedforward mechanisms. The same holds for all categorization responses in the brain, regardless of their apparent complexity: the progression from low-level to high-level feature detection (including categorization of faces or other complex stimuli) proceeds in a feedforward “sweep” that lasts 100 ms or less (Lamme & Roelfsema 2000).
What emerges is the nagging feeling that consciousness has nothing to do with the seeming complexity or “high-levelness” of a visual function. Whether a visual function depends on consciousness may simply be related to the amount of space that has to be travelled in the brain, how many processing steps have to be taken in between, and hence how much time it takes to complete. This converges onto a thesis that we may call:
The STERP-property of phenomenal representations =Df conscious representations depend on the spatio-temporally extended neural processing mediated by recurrent interactions.
What that extent is remains to be specified, but has been studied directly by Faivre & Koch (2014), who measured the effects of stimuli made invisible using CFS on the perception of subsequent visible stimuli. Both for apparent motion and for biological motion walkers, it was found that unconscious motion integration only occurred for relatively short (100 ms) and not for longer (400, 800, 1200ms) temporal intervals. Meng et al. (2007) observed that neural signals representing the spatial filling-in of a grating over a gap in the visual field depended on conscious experience of the grating.[34] This suggests that for visual information to literally “bridge a distance” across the visual field, consciousness is required.
The importance of the spatial and temporal extent of neural processing in consciousness also emerges from an entirely different field: that of disorders of consciousness. It is generally believed that there is a gradual decrease of consciousness from the healthy awake state towards minimally conscious, vegetative state and coma. These states also show a gradual decrease in the extent of neural interactions, in both space (Casali et al. 2013) and time (Bekinschtein et al. 2009). Particularly striking is the finding that the presence or absence of consciousness (in this case: the difference between minimally conscious and vegetative state patients) could be classified by simply looking at the amount of “shared symbolic information” in the EEG[35] at various distances in the brain. Shared symbolic information at distances of 10 cm and beyond signalled the presence of consciousness, and moreover was indicative of the prognosis of vegetative state patients (whether they would eventually awaken or not). Strikingly, this measure hardly depended on the location of the interactions (King et al. 2013). In other words, whenever and wherever neurons share information at distances of 10 cm or more, there is consciousness.[36]
Both distance and time are continuous. Arguing that consciousness is related to the temporal or spatial extent of neural processing therefore almost automatically seems to imply that the transition from unconscious to conscious processing is gradual rather than discrete. This is not necessarily so, however. Recurrent processing is mediated by highly non-linear interactions, and in such interactions, rather discrete phase transitions are possible (Steyn-Ross et al. 1999; Del Cul et al. 2007; Hwang et al. 2012). It could thus very well be that there is a discrete transition from a phase where information integration is rather limited to a phase that is characterized by extensive information integration, and that this transition depends on the temporal or spatial extent of recurrent interactions.[37]
Whether the transition from unconscious to conscious processing is discrete or continuous has been argued on different grounds, such as on the distribution of behavioral responses (“seen” versus “not seen”) in relation to manipulations of stimulus variables (Sergent & Dehaene 2004; Overgaard et al. 2006). In signal detection theory, the strength of perceptual information is considered to be continuous, while the decision criterion imposes a discrete boundary between what is reported as “seen” or “not seen”. In its classic form, however, signal detection theory is agnostic about whether consciousness is pre- or post-decisional. Recently, many attempts have been made to incorporate consciousness into the framework of signal detection theory, and in many of these models consciousness is considered post- rather than pre-decisional (Maniscalco & Lau 2012; King & Dehaene 2014)—thus the boundary between the conscious and unconscious is taken to be discrete. Based on neurophysiological findings in the monkey visual cortex, a signal-detection model was devised in which consciousness was considered pre-decisional. In this model, the distribution of sensory information was considered bi-modal, reflecting either a conscious or an unconscious state. The model could explain both the behavioral and neurophysiological findings in the monkey visual cortex, obtained using a variety of stimulus strengths and decision criteria (Supèr et al. 2001). Note that also in this pre-decisional model the conscious–unconscious divide is discrete (or at least bi-modal), rather than gradual.