88

Concerns with ensuring unawareness aside, imaging data raise a further issue—namely, that differential cortical activation does not guarantee the presence of representations which can influence task performance (Williams, Dang, and Kanwisher 2007). Here, I suggest, we arrive at my more fundamental disagreement with Block. In my opening statement I cited Norman et al. (2014) as providing evidence of genuinely perceptual (constancy-involving) representation outside of consciousness. What I disputed was whether this constituted perception by the individual. Whether high-level feature representation occurs outside of consciousness is a separate issue. I thus agree with Block that whether Jiang et al.’s (2006) results reveal individual-level perception does not turn on whether the representations mediating their effect are high or low level. What matters is whether those representations are constancy-involving and individual level. Here I press this second issue. Let us grant that Jiang et al.’s results show unconscious attraction and repulsion of attention. Block’s case that the mediating representations constitute perception by the individual “is accomplished . . . by [their] relevance to personal-level gender preferences”. He adds: “I can’t imagine how attraction or repulsion keyed to one’s gender preferences could be automatic in any relevant sense of the term”. However, personal-level gender preferences correlate closely with many reflexive, autonomic responses. For example, Rieger and Savin-Williams (2012) examine the differential pupillary responses elicited by gendered erotic stimuli. Such responses are naturally thought of as automatic. Furthermore, it is doubtful that we must think of the representations mediating them as individual level. Perhaps Block’s talk of responses being “keyed” to preferences requires the direct involvement of preferences in mediating responses from occasion to occasion. However, gender preference data cannot evidence responses “keyed” in this sense, since gender preferences cannot be manipulated on a trial by trial basis. Block denies that “involvement in central agency” is a necessary condition of personal-level attribution. I did not propose such involvement as a necessary condition. My suggestion was only that when a representation is unavailable to central agency, we lack a positive ground for attribution. Nonetheless, Block and I agree that certain kinds of response may indicate personal-level attribution. However, Block claims that “attraction and repulsion of attention is an individual-level matter and is not unrelated to agency” (cf. Burge 2010: 372). On this we disagree. Where attentional responses are completely stimulus-driven reflexes, operating entirely outside of voluntary control (e.g., Schoeberl et al. 2015), and possibly mediated by subcortical pathways (e.g., Mulckhuyse and Theeuwes 2010), I am unpersuaded that we must think of them as exercises of individual-level agency. If they are not, we lack positive reason for thinking of the perceptual representations implicated by Jiang et al.’s data as constituting individual-level perception. 15031-0091-FullBook.indb 181 5/12/2016 9:23:17 AM182 • Ian Phillips and Ned Block Part 6: Block We agree that there are unconscious representations in CFS but disagree on whether they are personal or sub-personal. I have been arguing that they reflect personal-level values and understanding. Values: Phillips says attraction and repulsion in Jiang et al. (2006) may be involuntary stimulus-driven reflexes that are subcortically mediated. However, the only actual evidence he presents for this is that pupillary responses are affected by gendered erotic stimuli. He claims, “Such responses are naturally thought of as automatic”. This is outdated. A recent review on this topic says: “The pupillary light response has long been considered an elementary reflex. However, evidence now shows that it integrates information from such complex phenomena as attention, contextual processing, and imagery” (Binda and Murray 2015, 1). This is a review, not an opinion by a fellow-traveler. Further, though exogenous spatial attention such as orienting to a loud noise is stimulus-driven and reflex-like, I know of no evidence for any reflex-like feature-based attention. Moving from values to personal-level understanding—I quoted a review that concludes (p. 8): “In summary, neuroimaging studies investigating the processing of visual information during interocular suppression have shown repeatedly” weak high level activations (Sterzer et al. 2014). Thus, it is very likely that there is unconscious high-level representation. I think whether unconscious representation constitutes unconscious perception turns on both content and role. On content: low-level properties like edge and texture register in early vision but are not normally part of personal-level cognition unlike our awareness of faces and emotions—hence the focus on high-level activation. On role: in the article just cited, Sterzer et al. note that although there have been many studies showing behavioral effects of CFS, no studies as yet have measured behavioral effects simultaneously with neuroimaging evidence of high-level perception. However, I know of no case of a high-level brain activation that does not have the potential to affect some kind of processing, if only on the temporal course of the processing (for example in priming). Phillips appeals to a claim by Hesselmann and Moors (2015) based on work by Randolph Blake and Bruno Breitmeyer on binocular rivalry that the default should “be not to expect much high-level unconscious processing during CFS” (emphasis added). This appeal is doubly flawed. First, the strong evidence for high-level CFS activations I referred to earlier is evidence for weak activations—as are the pupillary effects. If the default is to expect not “much” activation, that actually supports my position. Second, Phillips’ claim depends on a dubious inference from binocular rivalry to CFS. He says, “Yet given what else we know about flash suppression, and can reasonably extrapolate from studies of binocular rivalry (e.g. Zimba and Blake 1983; see Breitmeyer 2014 for a review) . . .”. However, Breitmeyer 15031-0091-FullBook.indb 182 5/12/2016 9:23:17 AMDebate on Unconscious Perception • 183 (2015) argues that binocular rivalry blocks off processing at the earliest stages of vision, whereas CFS operates at a mid-level. Even in the earlier (2014) Breitmeyer article, Breitmeyer places CFS above binocular rivalry in his hierarchy. If a stimulus (say a disk) is followed quickly by another stimulus (say a ring) which shares boundaries with the first stimulus (e.g., the disk sits just inside the ring), then conscious perception of the first stimulus can be reduced or eliminated. This “metacontrast masking” is strongest with the stimulus in one eye and the mask in the other especially when the stimulus and mask are presented nearly simultaneously, suggesting a combination of binocular suppression and metacontrast masking (Schiller and Smith 1968). CFS flickers at 10 hz suggesting that it also combines metacontrast masking with binocular suppression and that combined effect puts it higher on the hierarchy. A further item of evidence: Sklar et al. (2012) showed unconscious “semantic” priming in CFS. Sklar et al. presented three-digit subtraction problems to subjects under CFS (e.g., ’9–3–4’). Subjects then had to pronounce a single consciously presented digit that could be the result (e.g., ’2’). Results were faster than non-results. Subjects were asked to report the parity of the first digit in the subtraction problem, and those who got it right were excluded. And in a debriefing afterward, they excluded four subjects who said they had seen the primes. Phillips says excluding subjects who report more than the lowest visibility introduces a “notorious statistical artefact” (Shanks and Berry 2012). I don’t think the conditions for this artifact are met, but I don’t have the space for a discussion. There are four good reasons for thinking the effect was unconscious. First, Sklar et al. used Anthony Greenwald’s respected regression method that is designed to be used with a variety of visibilities (Greenwald, Klinger, and Schuh 1995). This method allows an extrapolation from higher visibilities to zero visibility. (See Kouider and Dehaene 2007, for further explanation.) Greenwald’s method showed a significant unconscious effect. Second, performance on the objective test was negatively correlated with the unconscious effect, suggesting that the effect is unrelated to conscious perception. Third, as Ran Hassin and Asael Sklar have emphasized in correspondence, the effect size for conscious priming is the same or at most twice the size of an unconscious effect. So in order for the effect in this study to be due to conscious perception, more than half the subjects would have had to be conscious of the stimuli—even after the elimination of all who scored above chance on the objective task. Fourth, the priming worked for subtraction but not addition. If subjects were indeed conscious of the stimuli, they should have been just as conscious of the addition as subtraction stimuli. (They speculate as to what the difference in unconscious processing of addition and subtraction and devise a procedure that shows effects for addition.) In sum, though there are plenty of loose ends in a rapidly moving field, there is a good case for personal-level unconscious representation. 15031-0091-FullBook.indb 183 5/12/2016 9:23:18 AM184 • Ian Phillips and Ned Block Part 7: Phillips I have been arguing that the existence of unconscious perception (construed as objective sensory representation by the individual) remains an open question. Where Block sees a few loose ends, I see unravelling tangled threads. Here I pull further on certain threads before offering some brief closing remarks. Block argues that unconscious representations in CFS are individual level since they reflect personal-level values and understanding. I cannot see how Jiang et al.’s data establish Block’s values claim (nor related claims about role). Block grants that “exogenous spatial attention” may be “stimulus-driven and reflex-like” but denies that the “feature-based attention” involved in Jiang et al. could be. But what does “feature-based” mean here? In Jiang et al., attention is not directed to gender as a feature but by gendered erotic stimuli to a spatial location. It is well-attested that fearful emotional stimuli can differentially draw reflexive spatial attention (e.g., Phelps, Ling, and Carrasco 2006, experiment 2). And an increasing body of work, including Jiang et al., indicates that “biological relevance, and not exclusively fear, produces an automatic spatial orienting toward the location of a stimulus” (Brosch et al. 2008: 362). Block draws attention to Breitmeyer (2015). What he does not draw attention to is that Breitmeyer there supports my suggestion that Jiang et al.’s effect may be subcortically mediated. Breitmeyer writes: “a suppressed erotically charged image presented in the left visual hemifield could, via retino-subcortical routes, activate the contralateral (right) pulvinar/amygdala, which, in turn, would activate their ipsilateral neocortex and thus bias attentive processing of stimuli in the left visual field” (243, fn. 6). Block is right that Breitmeyer does not think that CFS operates at the same level as binocular rivalry (BR). This does not mean that there is nothing that we can “reasonably extrapolate” from BR. After all, Breitmeyer holds that CFS partly relies on BR suppressive mechanisms. Moreover, whereas Block wishes to place CFS above metacontrast masking in the functional hierarchy, Breitmeyer places CFS “relatively low . . . in the functional hierarchy, somewhere between binocular-rivalry suppression and suppression by backward pattern or metacontrast masking” (2015: 243, fn. 7, my emphasis; cf. 2014: Fig. 5.4). Breitmeyer justifies his (avowedly speculative and tentative) placement of CFS above BR by appeal to Sklar et al. (2012), which Block also focuses on. This striking study reports the priming of responses to targets (e.g., ‘2’) by equations with those targets as answers (e.g., ‘9–3–4 =’). My earlier complaint that perceptual priming cannot directly reveal individual-level representation applies here. But this point aside, does Sklar et al. provide good evidence of sophisticated unconscious processing under CFS? As mentioned, a major issue here is the statistical artefact potentially introduced by the post hoc exclusion of subjects who performed above chance on either objective or subjective measures of awareness. Block denies that the conditions for this artefact 15031-0091-FullBook.indb 184 5/12/2016 9:23:18 AMDebate on Unconscious Perception • 185 are met (he does not say why), and finds it implausible that at least half of non-excluded subjects could have been conscious of the prime (something he suggests would be necessary to explain the relevant effect-size). I disagree. Sklar et al. excluded 60% of subjects. This suggests that the significant majority of their original group may have been conscious of the primes. This surely does raise serious concerns about truncation artefacts. It also appears consistent with half of the remaining subjects having some minimal awareness of the primes (cf. Hesselmann et al. 2015: §4.2). Block offers three further reasons for thinking that Sklar et al.’s effect was unconscious. First, their use of Greenwald’s respected regression method. However, Greenwald’s method is highly controversial given the large assumptions it requires, and great care is needed in its application (Dosher 1998, Merikle and Reingold 1998). Lacking space for a full discussion, let me note one salient point from Dosher which connects to Block’s second argument in favour of unconscious perception—namely, that performance on the objective task was negatively correlated with facilitation effects. This negative correlation indicates a non-linear relationship between direct and indirect measures. However, given such a relationship, facilitation may reduce to zero with or before the direct measure (indicating no unconscious perception), and yet the best-fitted linear regression misleadingly yield precisely the kind of non-zero intercept which Sklar et al. report as evidence of unconscious perception. At a minimum then, more sophisticated analysis is required for this method to be probative. Block finally argues that, in the relevant experiment, priming occurred for subtraction but not addition. But why think that this supports thinking of the effect as unconscious? Conscious or unconscious, the absence of an addition effect needs explaining. Sklar et al. suggest “that participants may have been less strategic in the [easier] addition equations”, providing evidence for this in relation to conscious arithmetic (Experiment 8). Thus, a strategic explanation is demonstratively available in relation to conscious perception. Where does this leave us? Throughout our exchange, Block has proposed various CFS studies as persuasive evidence for unconscious perception (e.g., Jiang et al. 2006; Mudrik et al. 2011; Sklar et al. 2012). However, as Block says, “everything depends on the details”. And, upon scrutiny, the proffered interpretation of these—and structurally similar—studies unravels. Arguably, breaking flash suppression studies only reveal differences in conscious detectability; attentional paradigms only reveal sub-individual-level perceptual representation; and the widespread practice of truncating data leads to the artefactual appearance of unconscious perception where none exists. Such unravelling is not unique to CFS. As briefly discussed in my opening remarks, neither traditional perceptual priming studies nor clinical conditions such as blindsight and neglect convincingly establish unconscious perception (see further Phillips forthcoming; and Phillips in press). Furthermore, whilst it has been convenient here to adopt a broadly Burgean conception of perception, 15031-0091-FullBook.indb 185 5/12/2016 9:23:18 AM186 • Ian Phillips and Ned Block that conception is hardly beyond dispute (e.g., Campbell 2011). And obviously answers to questions about unconscious perception turn crucially on our conception of perception. The upshot is that for all the resurgent field’s excitement about new techniques and findings, the current consensus in favour of unconscious perception remains significantly grounded in faith as opposed to fact. Part 8: Block Anna Karenina I endorse the “Anna Karenina” view of unconscious perception (Block 2011) according to which all conscious perceptions are alike, but each unconscious perception is unconscious in its own way. Successful conscious perception is a dance of oscillating feed-forward-and-back loops. Unsurprisingly, there are many substantially different methods of producing unconscious perception that interfere with the dance in different ways. Breitmeyer (2015) describes 24 substantially different ways in which unconscious visual processing can be produced, of which we have here discussed only a few. Given this variety of mechanisms, it is not surprising that Phillips’ criticisms of experimental paradigms have no real unity (other than the allegation of not-perception or not-unconscious). Here is a list of some of the experimental paradigms he discusses with a shorthand description of his criticisms: • Blindsight: failure of constancies and decision-theoretic criterion issues. • Unconscious color registration: representations are not available to central agency. • Breaking CFS (continuous flash suppression): control trials did not rule out CFS-specific effects, Breitmeyer hierarchy suggests high-level effects should not occur, post-hoc discarding of aware trials illegitimate, cortical activations may be epiphenomenal. • Gender-CFS: low-level confounds, reflexes. Personal Level Here are some areas of agreement between Phillips and me. First, we agree that there are unconscious representations that are involved in perception. (I say those representations often constitute perception, and he says not.) Second we agree that there are unconscious representations in perception that are objective. Phillips notes he “cited Norman et al. 2014 as providing evidence of genuinely perceptual (constancy-involving) representation outside of consciousness”. Phillips first complains that these color representations are nonetheless sub-personal because they “are not available to central coordinating agency”. Burge and I think central availability is not necessary for 15031-0091-FullBook.indb 186 5/12/2016 9:23:18 AMDebate on Unconscious Perception • 187 unconscious perception, and Phillips agrees, saying that the real point is that without such central availability we have no positive reason for ascribing the personal level. But when a sensory registration reflects personal-level understanding (Mudrik et al. 2011) or values (Jiang et al. 2006), that is a reason to think it is a personal-level perception. Value Phillips objects to Jiang et al. (2006) by alleging that it can be explained by unconscious perception of low-level features associated with gender. However, the connection to personal-level gender preferences is what is at issue, not high versus low level. Phillips also claims that the Jiang effect might be reflexive and subcortical, appealing to supposed reflexive effects of gender on pupil size. But as I noted, pupillary effects often reflect high-level processes. He also references Prinz’s view that the result is due to attraction of eye movements. But Prinz (2012) argues that unconscious recognition of the stimulus is what attracts the eye movement (p. 116). One caution about Jiang et al. (2006): this result is the only one I know of in which personal-level preferences are so strongly revealed in unconscious perception. Understanding I highlighted CFS as a method of producing unconscious perception because the episodes of unconscious perception last seconds (or even minutes) instead of milliseconds, and it can be experienced firsthand by anyone with a computer and a 10-cent pair of red/green glasses. When I started, I was ignorant of two issues concerning CFS, though I don’t see either of them as problems for the studies I cited (and neither apply to the Jiang et al. [2006] study just mentioned). One of the problems—the one emphasized by Phillips—seems to me a red herring. Many of the CFS experiments compared effects under CFS with comparable tasks without CFS to show there were no “CFS-specific” effects. This way of conceiving of the controls is a mistake. All should agree that unconscious processing underlies all conscious perception—though Phillips and I disagree about whether those underlying unconscious processes themselves constitute perception. So it would not be surprising if—without CFS—subjects recognize shaving with a fork faster than shaving with a razor. The real point of the controls should be to rule out low-level confounds and decision effects. On low-level confounds: Mudrik et al. use two batteries of measures to equate for low-level features. This gives their study evidential weight. A second problem with CFS is that as I mentioned there are brief periods of partial awareness—of low-level properties such as color (Mudrik et al. 2013). However, since anomaly is not such a low-level property, this is not a substantial problem. Note that rationale given for the controls is to avoid CFS-specific effects on behavior. However, if we move from a behavioral 15031-0091-FullBook.indb 187 5/12/2016 9:23:18 AM188 • Ian Phillips and Ned Block experiment to brain imaging, there is no need for such controls. I quoted Sterzer et al. (2014) on unconscious representation in CFS: In summary, neuroimaging studies investigating the processing of visual information during interocular suppression have shown repeatedly that object- or category-specific neural activity in high-level visual areas of the ventral stream is strongly reduced, but can be retrieved when sufficiently sensitive methods of data analysis are used, such as multi-voxel pattern analysis of fMRI data. Note that this comes from a review by the team that Phillips relies on and does not use the methodology that he objects to. These studies don’t test behavioral effects of these activations in the same experiments, but any neural activation can affect the temporal course of responding (“priming”) in an appropriately chosen task. Philosophers may be thinking of the color of wires in a computer that do not affect its operation, but this kind of causal isolation does not happen in the brain. In sum, there is strong evidence that unconscious sensory registration often reflects person-level values and understanding. I turn now to a different paradigm. Sandwich-Masking In Draine and Greenwald (1998), subjects were presented with a “sandwich- masked prime”, in this case, a word preceded and succeeded by “masks”, noisy stimuli known to make the sandwiched item harder to see. Immediately after that they were given a speeded task: classify a word presented without masks—the “target”—as pleasant or unpleasant. Immediately after that they had to decide whether the prime was a word or a series of Xs and Gs. (This tests how visible the prime was.) Both the primes and the targets were chosen from negative words like “vomit”, “kill”, and “bomb”, and positive words like “honor”, “happy” and “kiss”. The result was that if the prime and target were in the same evaluative category, subjects were faster in classifying the target and made fewer errors. Values for unconscious perception were obtained by Greenwald’s regression technique (mentioned in my last segment) in which responses under various levels of visibility are extrapolated back to zero visibility. The classifications of the primes and targets in this study engage both personal-level values and cognition. The experiment just described was criticized by many, including the authors, because the same words were repeated as primes and targets, and it was found that even single consonants from the repeated words worked as primes—suggesting that the result was due to associations and that unconscious understanding of the evaluative was not required. Using accumulated wisdom of many years of inquiry using this sandwich-masked priming 15031-0091-FullBook.indb 188 5/12/2016 9:23:18 AMDebate on Unconscious Perception • 189 technique, Klauer et al. (2007) give a new and convincing version of the experiment showing unconscious priming of novel evaluative and gender-related stimuli similar to the ones just described. The degree of priming was the same regardless of visibility of the prime, strongly suggesting that the effect does not depend on conscious perception. And they got the same results even when the visible targets were smiley and grumpy faces and the primes were the evaluative words, again suggesting that unconscious evaluative categorization was involved. A similar congruency priming experiment was used with stimuli like the anomalous pictures illustrated earlier—for example, a person drinking from a football rather than from a bottle (Mudrik and Koch 2013). The primes were low in contrast, presented briefly (33 ms) and sandwich-masked. (In the Mudrik 2011 CFS experiment described earlier, similar pictures were presented at full contrast for 2.5 seconds.) Subjects were shown a sandwich-masked prime that could be anomalous or not, then a consciously presented target that could also be anomalous or not. Subjects had to press a button indicating whether the target was “weird”, then rate the prime visibility, then whether the prime was “weird”, being instructed to guess if they did not know. Results were reported only for subjects whose rating was “saw nothing”. (I don’t have the space to explain why this procedure is legitimate.) The result was that subjects were slower to judge that a consciously presented picture was congruent (i.e., not “weird”) if it was preceded by an incongruent prime than if it was preceded by a congruent prime. The authors suggest that the unconscious processing of an incongruent prime may have attracted attention, depriving the subsequent task of attention. Again we have unconscious sensory registration that engages personal-level cognition. In conclusion, there are many experimental paradigms that support personal-level unconscious visual perception. Criticisms form an ad hoc list.