In a very stimulating target article, Ned Block places phenomenal consciousness at the core of his conceptual framework for consciousness (Block Reference Block1995b; Reference Block2001; Reference Block2005). In the past, we have proposed that access of information to a prefronto-parietal global neuronal workspace, capable of broadcasting that information to many distant cortical areas and therefore making it available for verbal or nonverbal report, is the essential ingredient of a theory of conscious perception (Dehaene & Naccache Reference Dehaene and Naccache2001). For Block, however, conscious reports fail to capture the whole conscious experience (and this incompleteness is not imputable solely to spurious factors such as reporting biases or errors). Block defends the existence of a phenomenal consciousness irreducible to conscious reportability. The key problem that we have with his thesis is its lack of empirical testability. Can we ever experimentally disentangle phenomenal consciousness from the cognitive accessibility inherent to conscious reports? In this commentary, we suggest that (1) Block's notion of phenomenal consciousness remains intractably entangled with the need to obtain subjective reports about it; and (2) many experimental paradigms suggest that the intuitive notion of a rich but non-reportable phenomenal world is, to a large extent illusory – in a sense that requires clarification.

1. Reports, conscious access, and phenomenology

If one wants to define phenomenal consciousness differently than conscious reportability, then one should resist the temptation to make use of subjects' reports to credit the existence of phenomenal consciousness. If the only support for the existence of phenomenal consciousness comes from conscious reports, then we find no reason to accept a major distinction between these two concepts. Yet Block's argument makes recurrent use of subjects' reports to advocate the existence of phenomenal consciousness. For instance, Block states:

In this case, aren't we also able to report it? Later, when discussing Sperling's seminal “iconic memory” experiments, Block writes:

and

In these sentences, Block relies heavily on subjective reports. He suggests that they provide face-value support for a rich phenomenal experience, supposedly going way beyond the few details that can actually be reported. Yet this seems to prove our point: Availability for report remains the sole criterion for conscious perception. We are still waiting for a clear, empirically usable definition of “phenomenality” or “qualia” that goes beyond what can be done based on subjects' reports alone.

2. What is a report?

It is, however, essential to better define what we mean by “report.” A report is not a “cut and paste” copy of a visual scene, but rather a conscious comment on an inner mental representation. This representation can originate from perceptual systems at multiple levels, but ultimately it results from their redescription by evaluative and interpretative systems. At any given moment, it provides only a partial and possibly biased description of the perceptual scene. In our view, the fact that, on any given trial, subjects cannot report the whole scene is therefore non-essential – what is essential is that any consciously visible item is accessible for report.

Block's comments on split-brain patients, locked-in patients and aphasic patients, also prompt us to remind him that conscious reports can be nonverbal or even become entirely covert, due to motor system impairments (Gazzaniga et al. Reference Gazzaniga, LeDoux and Wilson1977; Laureys et al. Reference Laureys, Pellas, Van Eeckhout, Ghorbel, Schnakers, Perrin, Berré, Faymonville, Pantke, Damas, Lamy, Moonen and Goldman2005). The absence of overt verbal report in patients is not diagnostic of the absence of conscious access and of internal form of “self-report” (e.g., internal speech). Using neurophysiological tools such as functional magnetic resonance imaging (fMRI) and event-related potentials (ERPs), indirect conscious reports can be obtained from these patients. They can and should be further complemented by looking for other psychological properties which are thought to be exclusive to conscious processing, such as active maintenance of mental representations (Owen et al. Reference Owen, Coleman, Boly, Davis, Laureys and Pickard2006) or dynamic regulation of executive control. In this broader definition, the reportability criterion is far from being a “behavioristic ladder” (see target article, sect. 4, para. 5) – rather, it provides an excellent test of whether or not an information is conscious accessible (Naccache 2006b).

3. Sperling's experiments

Block interprets Sperling's iconic memory experiments (Sperling Reference Sperling1960) and their recent extension by Landman et al. (Reference Landman, Spekreijse and Lamme2003) as a clear proof of a rich though non-reportable phenomenal world. We disagree, and propose an alternative interpretation of these experiments in the global neuronal workspace framework.

What happens in the brain when a large array of letters is briefly flashed, then removed? Since no mask is present, there is no reason to doubt the subject's report that he or she is aware of the array. The conscious content presumably stems from both a dorsal map of approximate letter locations (perhaps in area lateral intraparietal [LIP]) and a ventral representation of the letters (perhaps in area V4; Dehaene et al. Reference Dehaene, Jobert, Naccache, Ciuciu, Poline, Le and Cohen2004). Both dorsal and ventral representations must be jointly accessed by prefrontal regions when subjects report that they see a spatially extended array of letters.

Sperling's experiments clearly demonstrate that all individual letters remain represented for ~1 second within a posterior, presumably retinotopic location-specific buffer (iconic memory). However, other studies indicate that when attention is distributed, temporal and prefrontal regions act as capacity-limited filters and cannot represent more than a few objects (Kastner et al. Reference Kastner, De Weerd, Desimone and Ungerleider1998). When subjects report seeing “all the letters,” we suggest that they distribute their attention globally over the array, and thus are only able to determine its approximate numerosity and “letterhood”; our model predicts that only this approximate content, not the detailed letter identities, accesses a fronto-parietal global neuronal workspace. When subjects report individual letters, they change to a focused attention mode that allows them to enhance one location at a time, to let its specific content access the global workspace, and, therefore, to name it. Our model predicts that in this case, a local subregion of letter-sensitive retinotopic cortex (e.g., V4) becomes temporarily synchronized with anterior cortices and transmits a reportable visual content to prefrontal cortex (for related evidence, see Haynes et al. Reference Haynes, Driver and Rees2005; Sergent Reference Sergent, Baillet and Dehaene2005).

The model we just sketched, although in need of further specification, seems capable of accounting for the subjects' reports within the global neuronal workspace without appealing to Block's hypothesis of non-reportable phenomenal states. The mismatch between the report of seeing the whole array and the reduced capacity to report the individual letters therefore fails to provide univocal support in favor of Block's phenomenality-without-report hypothesis (as noted above, it does not even speak to this issue, since it is a mismatch between two reports).

4. The illusion of seeing

Part of the confusion surrounding the Sperling paradigm may stem from our description of the report of seeing the whole array as an “illusion of seeing” (Dehaene et al. Reference Dehaene, Changeux, Naccache, Sackur and Sergant2006). The term “illusion” applies imperfectly to Sperling's paradigm because, after all, the subjects' introspection turns out to be veridical: they can see the whole array, and they can, when probed, report any of its letters (though not all of them at once). However, the term “illusion” emphasizes that, until information is accessed by the fronto-parietal workspace, it is illusory to think that subjects necessarily have veridical knowledge of it. Indeed, our model predicts that, if a single letter of the Sperling array was replaced by another letter or even a digit prior to the focused attention stage, subjects might not notice it and still maintain that they “see all the letters” – a clear illusion. This substitution paradigm would lead to many testable predictions. For instance, a subpart of area V4 should have veridical information about the symbol's identity, which could be decoded by fMRI (see Haynes & Rees Reference Haynes and Rees2006; Williams et al. Reference Williams, Dang and Kanwisher2007) – but this spot should be (temporarily) functionally disconnected from frontal decision areas, and its information should not be used in subject's reports.

Many other paradigms and neuropsychological syndromes (Naccache Reference Naccache2006a) that are not discussed by Block indicate that reports of a rich phenomenality cannot be taken at face value (though we agree with Block that they still have to be explained in all of their details). In the “moving window” paradigm, for instance, where a computerized display is changed in synchrony with eye movements, viewers claim that they see a normal page of text even when all parafoveal information is replaced by strings of X's (Rayner & Bertera Reference Rayner and Bertera1979). Similarly, we all have the illusion of seeing a world in full color although color-sensitive cones are absent in the periphery of our retina. Such illusions suggest to us that building a theory of consciousness based on intuitions of phenomenality without reportability may be building on sand.