[1]
Or at least we assume it does not. This is the basic intuition we start from in trying to explain consciousness. If not, one easily slides into pan-psychism. That is a viable option of course: it could be that the camera does see, yet cannot “tell” us. However, the arguments put forward in the remainder of this paper seem to suggest that the camera does not see.
[2]
Among other things, like whether they maintain body temperature or have hairy skin.
[3]
With reptiles and birds in between, laying amniotic eggs on the land.
[4]
Of course there is the occasional mammal that lays eggs (e.g., the platypus) or fish that give birth to live young (e.g., the hammerhead shark). Still, calling these mammals or fish depends on the relative weight of other defining features, such as their way of breathing, feeding, body temperature maintenance, etc.
[5]
In a somewhat more mathematical analogy one could take all properties of all animals in the world, and perform a cluster or factor analysis. A good taxonomy has clusters that are aligned along the primary factors. Traditional taxonomy seems to have operated in this way implicitly.
[6]
One could do so, of course, which would lead to the denouncement of consciousness as a scientific phenomenon altogether, much along the lines of eliminative materialism (e.g., Churchland 1981). Daniel Dennett, in his categorical denouncement of anything coming close to qualia or even the phenomenology of consciousness, seems to follow a similar agenda (1993). It is entirely possible indeed that consciousness is a figment of our imagination, one that will evaporate upon close scientific scrutiny. Something like that happened to ‘elan vital’—the unique property of living matter—once we learned about chemistry, biology, DNA, and natural selection. For now, let’s assume that consciousness exists, and is in need of an explanation. If not, I would rather not be spending my years in neuroscience.
[7]

One important caveat is that for introspection we have to resort to cognitive functions like attention, memory, and “internal report”. This may result in both a potential underestimation of what we actually see (see for example the iconic/fragile/working memory discussion), and to an overestimation of what we actually see (as in the illusion of peripheral colour vision). This has been dealt with extensively elsewhere (Lamme 2010a).

[8]
Note that a proper treatment of response bias is important in this case. “Shy” subjects may feel inclined to respond “not seen” on most trials, more liberal subjects may feel inclined to respond “seen” on most trials. Only treating the responses in terms of signal detection theory (Swets et al. 1978) can truly establish the absence of any sensation (because the number of false alarms—subjects saying “seen” on trials without a target—is taken into account). From that perspective, using only partially-effective masks is not a proper method, not even when only those trials are used in which subjects reported not seeing the target.
[9]
A potential problem with the CFS manipulation is that “time to breakthrough” is often used as a measure of relative awareness of stimuli. Time to breakthrough is more or less analogous to a “yes” response (or hit) in a masking paradigm, and hence can suffer from response bias. CFS studies where responses are more rigorously treated in terms of signal detection theory are scarce. See Stein et al. (2011) for a more elaborate discussion on this problem with the CFS paradigm.
[10]

Note, however, that categorization is typically far better for stimuli than patients—or normal subjects—are aware of.

[11]
Or more precisely: as being either below or above 5, in this experiment.
[12]
It is unclear to what level invisible faces are processed exactly. Clearly, face/non-face categorization takes place for masked stimuli (see below), but whether face identity is also preserved depends on the exact experiment. Some find face-identity-specific priming and suppression of activation of the FFA and related face-selective-regions for backward masked stimuli (Kouider et al. 2009). However, this effect was only present for famous faces, not for unknown faces, showing that it may not be identity itself that is processed but “level of fame” or something similar. Others have made faces invisible using CFS, and found that face-specific adaptation only occurred for visible, and not for invisible faces (Moradi et al. 2005). The two studies are hard to compare, partly because of the different techniques used to make faces invisible (masking vs. CFS), but mostly because the latter used an adaptation effect as independent variable. It may be that unconscious categorization still occurred, yet did not result in learning (e.g., Meuwese et al. 2013; Meuwese et al. 2014).
[13]
Another illustration of the separation between feature selectivity and conscious experience is the observation that many neurons signal features of which we are not aware: V1 neurons signal the orientation of gratings that are of too finely spaced for us to perceive (He et al. 1996; Foster et al. 1985), respond to 3D disparity where we do not see depth (Cumming & Parker 1997), or signal invisible temporal frequencies (such as the flickering of light beyond the flicker-fusion frequency of about 15–25Hz, Maier et al. 1987).
[14]
Responses are modulated by such features, but typically this happens only after some delay (Sugase et al. 1999). The initial feedforward response is typically fully governed by a basic feature, like face vs non-face. Later on, responses are modulated by face identity or expression, and this is mediated by horizontal or recurrent interactions between neurons. We then enter the domain of feature integration, which is a hallmark of conscious recognition; see below.
[15]
It probably discounts the illumination by very much the same mechanisms as the illusory brightness shifts discussed above (via inhibitory lateral interactions). However, precise neural mechanisms may be different, as might be the cortical level at which neural responses reflect colour rather than wavelength.
[16]
It must be noted that in this experiment, the surrounds were not always fully invisible. In 86% of the trials, subjects reported not seeing the surrounds. Only these trials were used for the analysis. Within these trials, discrimination of the background (is the darker half left or right?) was at chance level, leading to the argument that indeed there was a full absence of awareness of the surround.
[17]
In this illusion, two surfaces of identical brightness are perceived as having different brightness, because there is a contrast edge between them.
[18]
I am not even going to dare mentioning their names here.
[19]
The fact that black-and-white photography works so well, has led us to believe that colour is a feature that is “painted” onto objects, as a sort of extra, independent of any other feature. We are now coming around from this view. For example, to compute the colour of an object, the object’s shape has to be taken into account, otherwise shadings would be misinterpreted. Object identity also influences colour perception: a brownish colour on a banana will be seen as more yellow than it would on a tomato.
[20]
This argument is, however, weakened by the fact that other long-range colour interactions remain in the blind hemi-field (Barbur et al. 2004), and by the finding that colour constancy mechanisms may depend on fairly early, monocular mechanisms (Barbur & Spang 2008). Moreover, it is reckoned that several colour constancy mechanisms exist, some of which are based on retinal adaptation mechanisms (Kamermans et al. 1998).
[21]
Obviously, these empirical issues about colour perception and consciousness have very direct consequences for many philosophical debates as well, given the many thought experiments that rely on colour perception and the whole notion of qualia.
[22]
Tononi similarly argues that consciousness always strives for “maxima of integrated information”, for which he uses the metaphor of the internet (2012). Like the brain, the internet is a highly interconnected structure where information travels from one part to other parts. In contrast to the brain, however, the internet is designed to transfer information from one specific part of the net (computer A) to a specified other part of the net (computer B), and it would in fact be rather counterproductive if this information were influenced by other information flowing from computers C to D or E to F. At another moment information may flow from A to C or D or F. The internet therefore does not strive for “maxima of integrated information”, whereas the brain typically does. Focussed attention, in such a view, would then be in fact a mechanism that counters this propensity towards maximally-integrated information, and which enables the brain to operate more strongly along the principles of the internet.
[23]
This is in fact such a strong intuition that it has led us to believe for a long time that consciousness must be some place in the brain “where it all comes together”. Descartes envisaged the pineal gland as such a place, and hence theories that lean towards such an explanation of consciousness are often said to suffer from the fallacy of the “Cartesian theatre”.
[24]
This manipulation is a combination of backward and forward masking, and also somewhat reminiscent of dichoptic masking, in that in subsequent displays images with the opposite orientation contrast are shown. See the two images of figure 9, but then not presented to the two eyes but in rapid alternation.
[25]
A similar setup was used in the curious case of alleged “blindsight in normal observers”. In one of the experiments in that paper, target figures were made invisible using the same manipulation of dichoptic presentation of orthogonally-oriented elements. It was claimed that despite their subjective invisibility, subjects were able to localize the targets above chance, just as blindsight patients do for unseen stimuli (Kolb & Braun 1995). The findings were not replicated, however (Robichaud & Stelmach 2003).
[26]
This paradigm has been shown to distinguish between seen and not-seen stimuli in monkeys with a V1 lesion in one hemi-field, and was used to differentiate between “conscious” visual responses and unconscious blindsight behaviour: without catch trials (i.e., when in forced choice mode), monkeys react to both stimuli in the intact and in the lesioned field, expressing blindsight capabilities. In catch trials, however, monkeys only respond to stimuli in the intact and not in the lesioned hemi-field, as if expressing conscious sensation instead of a mere reflex (Moore et al. 1995). Supèr et al. used the same paradigm in intact monkeys to assess conscious percepts of figure–ground stimuli.
[27]

A later investigation into neural activity preceding either seen or not-seen figure trials showed that not-seen trials are preceded by somewhat lower level of spontaneous activity, and also express less inter-neuronal synchrony (Supèr et al. 2003; Van der Togt et al. 2006).

[28]
Yet this transition may be the “Holy Grail” for those willing to understand qualia—or at least for those believing in “soft qualia”, i.e., phenomenal properties that are not entirely detached from visual functioning, and having some sort of neural substrate (Block 1996, 2005, 2007).
[29]
In my reading, the predictive coding models are sometimes rather vague about exactly which signal mediates conscious experience. It is often seen to be a combination of the matching process and the posterior, e.g., Seth et al. (2011).
[30]
But note that this is in fact nothing more than a semantic priming experiment. The results primarily show that if a semantic category has been activated, this category will then break earlier from CFS.
[31]
Of course one could argue that in the case of a face on a blank background there also is figure–ground segregation. This type of segregation clearly does not depend on consciousness. This touches on the debate on whether categorization is possible without segregation (Wagemans et al. 2012).
[32]
A promising theory of consciousness holds that conscious representations and states are characterized by the integration of information, or more precisely, on the formation of complexes of integrated information (Tononi 2004, 2008, 2012). That integrated information characterizes consciousness is, however, mainly derived from a set of axioms and introspective or intuitive thought experiments, most of which have already been discussed in the previous text or footnotes (Tononi 2012). What this review of experimental findings however shows is that the “integration of information” comes in many guises, not all of which are equally strongly related to consciousness. A somewhat more precise definition of “integrated information” may be guided by these experimental findings.
[33]
The figure is 4 degrees of visual angle wide. Neurons in human V1 with receptive fields at that distance are about 20 mm apart, given a cortical magnification factor of 0.2 degrees per millimetre at 2.0 degrees eccentricity (Duncan & Boynton 2003).
[34]
It did not depend on attending the grating, however, which is of relevance to the discussion on the relation between attention and consciousness. See below (Lamme 2003, 2004, 2006, 2010a, 2010b).
[35]
At each electrode, EEG signals were first transformed into symbolic shapes (e.g., up-down-up) for various temporal intervals. Then it was determined to what extent these EEG “symbols” covaried between electrode pairs of various distances, after the exclusion of covariance that was caused by simple volume conduction.
[36]
Which made me wonder whether any piece of cortex of 10 cm or larger that is held on life support in a petri-dish might have consciousness.
[37]
It could even be that the mere fact that information exchange extends over a particular time and space is critical for that exchange to be accompanied by a conscious sensation. When the same amount of information would be exchanged much slower — as in plants — or much faster — as in a supercomputer — or over a smaller or larger space (as in a microchip or over the internet) no conscious sensation ensues.
[38]
Note that “meaning” in this context refers to the meaning information has to the organism, shaped by and in accordance with its evolutionary history and ontogenesis (like colour has the “meaning” of the edibility of fruit). It does not refer to “meaning” in any linguistic sense.
[39]
That may explain why two seemingly similar phenomena like the brightness and colour shifts of figure 2, and the arrival at colour constancy in figure 3 are depend on consciousness in different ways. Colour constancy requires the computation of the full distribution of wavelengths over the entire image, which takes more time than the computations required to compute brightness of adjacent patches.
[40]
Maybe that is the reason why the transition from unconscious to conscious processing also marks the transition between veridical and inferred representations (e.g., from wavelength to colour). Dedicated modules can do their thing in isolation, and therefore have no need to compromise towards a non-veridical representation of the outside world. When modules interact, the necessity may arise to compromise veridical representations to achieve global coherence into the combined super-positioned representation that cannot be represented otherwise.
[41]
Even something as simple as a white background will give the black line another visual “meaning” than a yellow background, a green background, or a textured background. The same point has been formulated by Giulio Tononi (2004, 2008, 2012): a conscious representation is conscious because it differentiates from the endless other potential representations that could have been. In this case: the oriented black line on the white background is one of the endless possible configurations of lines on backgrounds, and only by integrating the information of line and background is it known which of these configurations is actually present.
[42]
Similar ideas exist in the context of motor learning: a task that first requires extensive conscious practice will gradually become more and more automatic, up to the point where it can be executed fully unconsciously.