Block's primary concern involves whether phenomenal consciousness (henceforth, P-consciousness) is independent of what he calls “cognitive accessibility.” He argues yes (e.g., that P-consciousness can “overflow” access), and contrasts this view with that of others (e.g., Baars, Dehaene, Dennett, Naccache, etc.) who argue instead that access (for them, a.k.a. being in the “global workspace”) is in part constitutive of P-consciousness – and hence that phenomenality without access cannot occur.

These fundamental questions can be usefully engaged from a framework informed by signal detection theory (SDT; e.g., Macmillan & Creelman Reference Macmillan and Creelman1991). To foreshadow, to show that P-consciousness is independent of access does not require demonstrating that there are P-conscious contents that cannot be accessed (Block's strategy). Rather, one need only show that sometimes P-conscious contents are not accessed, whereas sometimes they are. In this way, even if Block is wrong, one can still show that access is indeed an independent conscious process distinct from P-consciousness proper.

The easiest way to show this is to sketch our theory of consciousness, which derived from attempting to solve a perennial problem in unconscious perception research: How can consciousness be validly indexed? It seems clear that we must rely in some way upon subjects' reports – but what kind of report?

First, we could simply inquire about participants' subjective conscious experience; for example, “Can you see (hear, remember, etc.) X?” If they say no, they are not conscious of X; if yes, they are. Using such methods, ostensibly unconscious effects (including above-chance discrimination) are easily obtained. Despite “subjective threshold” methods' intuitive appeal, however, SDT shows that all reports involve not just (1) the perceptual process itself, but also (2) an additional, independent criterion-based decision process wherein the strength of available perceptual evidence is evaluated as to whether it is sufficient to reply in the affirmative. That is, contrary to many researchers' implicit assumptions, there is no such thing as an unmediated “subjective report” – ever. Consequently, the SDT analysis directly implies that if someone says “no,” this does not show that they have no relevant P-conscious experience, but perhaps merely that such experience, being very weak and hence low-confidence, is below criterion. With this in mind, subjective threshold effects – which are easily obtainable – are plausibly weakly P-conscious, not unconscious.

This “conscious but weak” interpretation is strongly supported by the under-appreciated fact that the same result (i.e., the same estimate of perceptual sensitivity, d′) is obtained in both the yes/no form of the detection task and in the so-called rating form, wherein participants rate their confidence that a target stimulus (vs. noise only) has appeared. This means that whenever d′>0, participants do in fact possess relevant subjective confidence – however weak – which enables above-chance stimulus discrimination.

Despite the above, some argue that subjective threshold stimuli really are phenomenally unconscious (henceforth, P-unconscious) after all. Perhaps the most compelling are various demonstrations of “exclusion failure” (e.g., Merikle et al. Reference Merikle, Joordens and Stolz1995), wherein participants respond with a previously presented word above baseline despite instructions not to, apparently suggesting that such stimuli are unconscious. Nonetheless, these effects too can be explained as weakly P-conscious if exclusion is a decision process such that sufficient confidence is required to exclude a candidate response (Snodgrass Reference Snodgrass2002; see also Haase & Fisk Reference Haase and Fisk2001).

Consistent with this analysis, Visser and Merikle (Reference Visser and Merikle1999) found that exclusion failure disappeared when participants were paid for successful exclusion – a classic criterion manipulation (SDT has long shown criterions can be shifted at will). This finding is crucial, because it shows that subjective threshold stimuli can be accessed, even though typically they are not. Here again, this is exactly what one would expect from (initially) below-criterion but nonetheless weakly conscious stimuli.

Perhaps instead, then, we should use the second form of report – namely, by asking participants to objectively discriminate the stimuli rather than deliver opinions about their phenomenal experience. By using SDT methods, we can independently estimate d' and the criterion (c). In this way, if discrimination responses are required even when confidence is low, criterion artifacts of the sort just mentioned are avoided, and if stimulus conditions are arranged such that d′=0 (i.e., the “objective threshold”), there are strong reasons to conclude that any other obtained effects are truly not P-conscious. Strikingly, under these conditions, clearly qualitatively different phenomena emerge (Snodgrass et al. Reference Snodgrass, Bernat and Shevrin2004; Snodgrass & Shevrin Reference Snodgrass and Shevrin2006) including both facilitation and striking below-chance inhibition – which never occurs under subjective conditions.

So are subjective methods simply invalid? Despite the above, we think not, because they do index a centrally important phenomenon: whether or not higher-order, controlled (vs. automatic) processing is applied to the relevant stimuli. In other words, subjective methods index our criterion for access to higher-order processes including reasoning, planning, and so forth. After all, strategic control processes will not be attempted if we don't think the relevant stimuli exist.

With all this in mind, we (Snodgrass Reference Snodgrass2002; Snodgrass & Shevrin Reference Snodgrass and Shevrin2006; Snodgrass et al. Reference Snodgrass, Bernat and Shevrin2004) propose that there are two, not one, kinds of consciousness: (1) first-order (phenomenal consciousness), and (2) second-order (reflective consciousness), which takes first-order (and other second-order) conscious states as its objects. In this model, while all reflectively conscious states are also phenomenally conscious, the converse is not the case – that is, there are phenomenal states that are not additionally reflectively conscious. Rather, because reflective consciousness is highly capacity-limited, only the most relevant phenomenally conscious stimuli – even if strong – are selected for additional, optional higher-order processing. At the same time, under usual circumstances essentially any phenomenally conscious content can be selected into reflective consciousness if desired, as the SDT analysis of typically subjectively unconscious stimuli demonstrates.

Putting it all together, stimuli can clearly be P-conscious without being accessed. This is most obvious with subjectively unconscious stimuli, which are typically not accessed, but nonetheless exhibit effects qualitatively identical to clearly P-conscious stimuli whenever further (and optional) reflectively conscious (i.e., strategic) processing is absent. Moreover, they can be accessed, given our willingness to do so (e.g., use more liberal criteria). In contrast, clearly (objectively) P-unconscious stimuli are intrinsically inaccessible, and display qualitatively different effects.