R2.2. Generic/specific phenomenology

The distinction between generic and specific phenomenology was crucial in the overflow argument, although I did not give it a name or draw enough attention to it (I did speak of experiences “as of … specific shapes.”) I will explain it here by example. In the Landman et al. experiment, the relevant generic phenomenology would be the phenomenal presentation that there is a circle of rectangles. The relevant specific phenomenology would be a phenomenal presentation that specifies for each of the rectangles (or anyway, most of them) whether they are horizontal or vertical. For the Sperling experiment, the relevant generic/specific difference would be that between a phenomenal presentation that there is an array of alphanumeric characters and a phenomenal presentation of specific shapes of all or most items in the array. I argued that in the Sperling, Landman et al., and Sligte et al. experiments, there was specific phenomenology involving all or almost all of the items, as well as generic phenomenology. This distinction figures – in those terms – in the commentaries by Burge, Grush, and Levine. Kouider et al. express it as the lower/higher distinction, Papineau as the scene/item distinction, Sergent & Rees as the scene gist/detail distinction, and in the target article as well as in the Byrne et al., Naccache & Dehaene, Jacob, Spener, and Van Gulick commentaries, the distinction is deployed without special terminology.

My argument was that before the cue, there is specific phenomenology for all or almost all items (and also generic phenomenology, though that does not figure in the argument). This specific phenomenology is what justifies the claim that the capacity of the phenomenal system is more than 4, whereas the capacity of the access system is 4 or less and thus that the two systems cannot completely coincide. A number of the critiques (most pointedly, Papineau's and Byrne et al.'s) challenge the premise that there are more than 4 items of specific phenomenology before the cue. It is important to recognize that the objectors have to agree that before the cue, there are specific (not just generic) visual representations of all or almost all of the 8 to 32 items in the Sperling, Landman et al., and Sligte et al. experiments. There have to be such specific representations given that any location can be cued with high accuracy of response. The locus of controversy is whether those specific representations are phenomenal.

Here is my evidence for claiming that there is specific phenomenology for all or almost all the items in the overflow experiment:

1. 1. As Burge notes, subjects (including myself) in overflow experiments often testify that their responses are based on specific phenomenology that was there all along. (Rogier Landman tells me that the extent to which subjects evince specific phenomenology may be correlated with how well they do in the experiments.) Shanahan & Baars say “It is worth noting that subjects in such experiments believe they are simply reporting the contents of their recent visual consciousness, even after the visual stimulus has disappeared from view.” In the Sligte et al. experiment, subjects have up to 4 seconds before the cue comes on, so their judgments about their phenomenology are not rushed.

2. 2. Subjects are attending to arrays in full view, in good viewing conditions, for half a second in the Landman et al., and some versions of the Sperling experiments, and one second in the Sligte et al. experiment, more than enough time for specific phenomenology. (Burge also makes this point.)

3. 3. If there is only generic phenomenology before the cue, and if the cue causes the generic phenomenology to be replaced by specific phenomenology, then there is a shift from generic to specific phenomenology. The fact that subjects report no such phenomenological shift might not be strong evidence against this view, but it is some evidence. The vast literature on this topic (including two Ph.D. theses I have read) contains no mention of such a thing as far as I know. I myself can testify that even looking for such a shift, one does not experience it. The point of the target article's example of the rectangle coming into view as if from a distance was to compare that phenomenal shift with the absence of such a shift in the overflow experiments. If there are some items of specific phenomenology before the cue, and different items of specific phenomenology after the cue, then one might expect that to be noticed as well.

4. 4. There is evidence mentioned in the target article that cortical persistence obtains at all levels of the visual system and therefore at the phenomenal level as well. In particular, there is evidence (mentioned in the target article) that the persistence exists at levels where depth and motion are represented. As Lamme and Landman & Sligte note, persisting representations obtain at a stage of visual processing past figure-ground segregation and feature binding, properties that “are more associated with conscious processes” (Landman & Sligte's commentary, para. 5). Hence, there is a neural case for phenomenal persistence.

5. 5. In the target article, I mentioned Di Lollo's paradigm using a 5 by 5 grid in which all but one of the squares is filled with a dot. Subjects see a partial grid with 12 of the dots filled in, then, after a delay, another partial grid with a different 12 dots filled in. The subjects' task is to report which square has a missing dot, something they can do easily if they have a visual impression as of the whole matrix of dots. Loftus and Irwin (Reference Loftus and Irwin1998) show that subjects' ability to perform the task correlates nearly perfectly with their phenomenological judgments of whether there appears to be a whole matrix rather than two partial matrices. When writing the target article, I did not know about the variant by Brockmole et al. (Reference Brockmole, Wang and Irwin2002) in which the appearance of the second partial grid was delayed by as long as 5 seconds and in which subjects were told that a good strategy was to “imagine the dots still being present after they disappeared” (p. 317). The subjects' memory capacity for the 12 dots in the first grid can be computed by the type of errors made. When the delay between the first and second partial grids is 100 msecs, the subjects' retention capacity falls from 12 to 4.1 of the 12 dots in the first partial grid. The striking result was that with delays over 100 msecs, subjects' capacity increased, asymptoting at a delay of about 1.5 seconds, at which time their capacity was 10 of 12 dots, and the capacity stayed that high for delays up to 4–5 seconds. Independent estimates of the time to generate a mental image (Kosslyn et al. Reference Kosslyn, Thompson and Ganis2006) are between 1 and 2 seconds, and the authors argue that the subjects were following instructions, generating a visual image of the first array, and integrating that visual image with the percept of the second array. This result constitutes converging evidence for high capacity specific phenomenology: Since the subjects could do the task well, the Loftus and Irwin result suggests they have a visual impression of the whole matrix; and in any case visual imagery is phenomenal. So the representations are phenomenal, and the capacity of 12 dots is substantially more than 4 items. The upshot is that there is a completely different paradigm in which the evidence favors high capacity specific phenomenology.

6. 6. Finally, there is the evidence mentioned by Kouider et al., which I discuss immediately below.

Kouider et al. suggest what they take to be an alternative to what I am suggesting but what I take to be a version of it. They hypothesize that Sperling-like paradigms result from “partial awareness: subjects have a transient and degraded access to fragments of all the letters in the grid.” Kouider et al. take comfort from a common observation that subjects in experiments involving masked stimuli often report seeing bits and pieces of stimuli. What are Kouider et al. saying about specific phenomenology? No specific phenomenology at all is not compatible with their view, since they say subjects are to some degree conscious of and have access to “fragments of all the letters in the grid.” And these commentators clearly don't envision full specific phenomenology, so they must envision partial specific phenomenology. Their evidence involves cases (Kouider & Dupoux Reference Kouider and Dupoux2004) of brief masked presentations of color words and pseudo-color words (e.g., “green” vs. “gener”) in which both have the same effect on subsequent identification of colored stimuli, facilitation in the case of congruent stimuli (e.g., “green” or “gener” followed by green) and inhibition in the case of incongruent stimuli (“green” or “gener” followed by red). Kovider et al. predict, plausibly, that in Sperling experiments that include some letter-like symbols which are not letters, subjects would treat false letters as similar to real letters. More generally, Kouider and Dupoux (Reference Kouider and Dupoux2004; Reference Kouider and Dupoux2007) give evidence that genuine semantic priming requires at least partial phenomenology and that totally unconscious stimuli can have only non-semantic effects. (But see Abrams & Grinspan Reference Abrams and Grinspan2007.) To the extent that I disagree with Kouider et. al., that disagreement is on just how degraded the specific phenomenology is. One of the experiments in Sligte et al. (Reference Sligte, Scholte and Lamme2008) used bars that differed in increments of 45 degrees, and subjects still showed high capacities – slightly under 8 for stimuli of 16 bars with the cues presented at 1,000 msecs after the stimulus offset. The specific phenomenology is good enough to make those distinctions between say 135 degrees and 180 degrees pretty well. In the Sperling experiment, the specific phenomenology was good enough for subjects to distinguish among the 26 letters of the alphabet. No doubt if there had been pseudo-letters, subjects would have made more errors, but distinguishing among 26 alternatives is still respectable. It is worth noting that in the experiment by Kouider and Dupoux, stimuli were presented at 29 msecs or 43 msecs and also masked, an intervention aimed at making them harder to see. Landman et al.'s stimuli were normally unmasked and presented for 500 msecs, Sligte et al.'s for 1,000 msecs (and Sperling's stimuli yield the same results with a 500 msec presentation – cf. also Burge); so the fragmentariness of the phenomenology in Kouider and Dupoux (Reference Kouider and Dupoux2004) could be predicted to be greater than in the overflow experiments. The upshot is that Kouider et al. have presented further evidence for high-capacity specific phenomenology, just what my overflow argument relies on.

Byrne et al. are not impressed with my appeal to subjects' testimony. They raise a plausible objection, that the generic/specific distinction is pretty abstruse, so how can the responses of subjects who don't know the distinction provide support for specific phenomenology? However, when subjects say (cf. Shanahan & Baars) that in reporting the letters in the cued row or in telling whether the cued rectangle changed orientation, they are simply reading their answers off of the visual impression that was in existence before the cue, they are evincing specific phenomenology regardless of whether or not they could state the distinction between specific and generic phenomenology.

Readers may be thinking that subjects may have seen the orientations/identities of a few of the items, but that subjects cannot be sure that they saw all or most of them. Recall that specific representations of all or almost all the items before the cue (though perhaps fragmentary representations) have to be postulated to explain the fact that subjects can report the items no matter which row is cued. So the options would appear to be either that there was no specific phenomenology before the cue, or that there was specific phenomenology involving all or most of the items, even if fragmentarily. As I just mentioned, subjects' testimony (and lack of surprise in what they can do) suggest the latter.

Papineau hypothesizes that the presence of generic phenomenology is my basic reason for postulating specific phenomenology. I should have drawn more attention to my reasons, as I have now done.

Van Gulick notes that the “movie screen of the mind” view would say that you cannot have generic phenomenology without specific phenomenology, implicitly suggesting that I am relying on the “movie screen of the mind” view, and on the fact of generic phenomenology, to argue for specific phenomenology. (Grush uses what would appear to be the movie screen view of pictorial representation, but the other way around from what Van Gulick ascribes to me: Grush says generic phenomenology cannot be an image because if it were an image, clear contents would have to present the details.) I reject the principle – applied by both Grush and Van Gulick – that pictorial representation has to specify the relevant details. I call this principle the “photographic fallacy” (Block Reference Block1983). More specifically, the photographic fallacy supposes that pictorial representations have to represent details of anything in view in the manner of a prototypical photograph. To see the fallacy, note that an impressionist painter might represent a hand in broad brush strokes that do not explicitly represent the number of fingers or whether one of them has a ring.

Van Gulick also argues that phenomenology in the overflow cases may be partial, weak, or somewhat indeterminate, but that access may also be limited in the same way, so there is no evidence for a discrepancy. With regard to the issue of whether degree of phenomenology is matched by degree of access, there is experimental evidence to the contrary mentioned by Sergent & Rees (e.g., Sergent & Dehaene Reference Sergent and Dehaene2004). Cognitive access appears to be more of a binary phenomenon than Van Gulick supposes.

Spener argues that the fine details of introspective grasp of specific phenomenology are thoroughly expectation-driven. In support, she mentions the disagreements in the philosophical literature about whether or not there is a determinate number of phenomenologically represented speckles on the speckled hen.

I think Spener exaggerates the problem. Phenomenological disagreement can be to some extent settled. The speckled hen case, in particular, has been illuminated by work on the different “grain” of vision as opposed to attention. Patrick Cavanagh and colleagues (Reference Cavanagh, He, Intriligator, Taddai-Ferretti and Musio1999) have shown that the grain of vision is about one-sixtieth of a degree at the fovea. For example, in order for a grating to be distinguishable from a gray field, the individual lines have to subtend more than one-sixtieth of a degree. However, for a subject to attend to visible lines and to be able to move attention from one to another, the lines must subtend at least five- to ten-sixtieths of a degree. How is this relevant to representations of speckles? Plausibly, the phenomenological disagreement about represented speckles stems from conflating seeing with attending. The speckles in the standard example in which they subtend more than one-sixtieth of a degree, but less than five- to ten-sixtieths of a degree, are visible but not attendable, so that one can see them (and thus phenomenally represent them) but not count them or do anything else that requires moving one's attention from one to another. If one is looking right at the speckles for sufficient time, the phenomenology really does determinately represent many of the individual speckles – specific phenomenology as opposed to generic phenomenology – but the speckles have an “elusive” quality because one cannot attend to them.

Grush argues for an illusion in which generic phenomenology presents an object as affording answers to certain queries, so when the answers to the queries are filled in, the subject does not notice. One might call this putative illusion the affordance illusion. (Grush gives it a less mnemonic name.) The affordance illusion is supposed to explain change “blindness” and inattentional “blindness” in terms of sparse generic representations. Of course the same idea, if it worked, could be used to undermine my reason (3) for specific phenomenology. The one argument for this illusion that Grush offers is that if there were an affordance illusion, it would serve to explain change “blindness” in terms of sparse representations. But as I noted in the target article, the Landman et al. and Sligte et al. experiments are themselves evidence against the “sparse representations” view of change “blindness.”

Jacob raises the issue of whether what I say in the target article is compatible with what I have said against representationism. I say yes, because the distinction between generic and specific phenomenology requires only that in the cases in question, phenomenology has representational content of the appropriate kinds; and that is compatible with the view (which I hold) that there is more to phenomenology than representational content.

R2.3. Unconscious representation before the cue

So far, I have been talking about the objection that before the cue, there is only generic phenomenology, and no specific phenomenology. However, there is a more radical view that says that until the cue appears, there is no phenomenological representation of the array at all, either generic or specific. Dehaene and his colleagues (2001; Reference Dehaene, Changeux, Naccache, Sackur and Sergant2006) were arguably committed to this more radical view by the refrigerator-light illusion hypothesis, as emphasized by Burge, but Naccache & Dehaene in their commentary advocate the less radical thesis. The only advocate of the more radical thesis among the commentators is O'Regan & Myin. My main response to the more radical thesis is that the arguments I have given for specific phenomenology before the cue constitute a case for some phenomenology before the cue.

O'Regan &Myin note, correctly, that I am assuming that consciousness is a natural kind and has some sort of neural signature. They claim this hypothesis is speculative and that even if true, it would not solve the “hard problem.” One of the points of my target article is that we can discover a lot about the neural basis of consciousness, short of solving the hard problem. On the issue of speculativeness: I am not assuming there is something physical in common to all possible cases of consciousness (e.g., including mammals, birds, octopi, conscious machines, and conscious extra-terrestrials, if there are any). My assumption is that there is a neural signature of consciousness in humans that is shared at least by other mammals with similar sensory systems. This is an assumption that I believe is shared by the field and looks promising so far. O'Regan & Myin note that I claim that phenomenal consciousness has effects on the basis of which we can find evidence about its nature. They interpret my earlier paper in this journal (Block Reference Block1995b) as arguing for the opposite view, epiphenomenalism. Although Block (Reference Block1995b) took epiphenomenalism seriously, the upshot was (and is) that we have reason to reject it.

R2.4. Hyperillusions

Naccache & Dehaene appear to agree with me (and disagree with Byrne et al.) that subjects think they have specific phenomenology of all or almost all the items in the overflow experiments; but Naccache & Dehaene think subjects are suffering from what I called a hyper-illusion. In ordinary illusions, appearance misleads a subject about reality, but in hyper-illusions, appearance of appearance misleads about appearance. (This may sound glib and meaningless – and perhaps it is – but I think it is fair to describe a putative phenomenon in a way that brings out its peculiarity.) Naccache & Dehaene argue that “we all have the illusion of seeing a world in full color although color-sensitive cones are absent in the periphery of our retina.” And Van Gulick makes the same claim, citing Dennett's (Reference Dennett1991) color marker demonstration: we hold a colored marker to the side of our visual field, bringing it gradually into the fovea, where we supposedly begin to see its color. However, the claim of illusion on the basis of retinal distribution of cones is analogous to (though not quite as obviously wrong as, for reasons I will get to) the idea that there is an illusion involved in seeing the world right-side-up, given that our retina contains upside-down images. What is on the retina is relevant to phenomenal experience only to the extent that it affects the parts of the brain that determine phenomenology. Assuming for the sake of the example that activity in V4 is the neural basis of color phenomenology, then the issue of whether the world genuinely appears colored in the periphery would be illuminated by knowing whether V4 has receptive fields in the periphery (that is, whether V4 contains groups of neurons that respond to and represent areas in the periphery), not by knowing what is on the retina.

The two putative illusions (the putative color illusion and the putative upside-down illusion) differ because there is a genuine issue: Does V4 have receptive fields in the periphery? And if it does, what is the purpose of those fields, given low color sensitivity in the retinal periphery? An obvious answer is: Because visual phenomenology depends on integrating information over time. Our visual representations comprising the whole visual field are not built up in an instant! So color representation of something now in the periphery can derive from its representation recently in more central vision. Striking evidence for integration over time in some aspects of perception is provided by evidence that multicolor representations are built up serially – the system processes one color at a time – although locations and shapes are processed in parallel (Huang & Pashler Reference Huang and Pashler2007; Huang et al. Reference Huang, Treisman and Pashler2007). There is another response to the question of why cortical color representations would represent color in the periphery: It turns out that hue discrimination at 50 degrees of eccentricity is as good as in the fovea (which subtends only 2 degrees – about twice the width of the thumbnail at arms length) if the size of the stimulus is magnified, and there is even some color sensitivity at 80 to 90 degrees (Mullen Reference Mullen1992).

Naccache & Dehaene mention another alleged hyper-illusion, the “moving window” experiment by McConkie and colleagues (Reference McConkie and Rayner1975; Reference McConkie and Zola1979), in which subjects suppose there is a full page of normal text even though the text outside the small moving window is degraded or changed. The point just made about representations integrated over time applies here too.

R2.5. Phenomenal memory

Jacob notes that my talk of phenomenal memory is misleading. What I meant was that there is neural persistence at all levels of the visual system, including the level that determines phenomenology. So instead of talking about phenomenal memory, I should have just talked about phenomenal persistence. Byrne et al. say I don't need high-capacity phenomenal persistence but only high-capacity phenomenology. The role of phenomenal persistence in my target article was in connection with the “looking again” dialectic that I presented. The argument for the high capacity of the phenomenal system can be undermined by the idea that the high capacity derives from the information in the environment or the high capacity of low-level vision, for example, the retina. Tracking the sources of phenomenal persistence allows one to rule out those alternatives. Recall that I argued that phenomenal persistence is at a level that involves binocularity (Engel 1970) and motion (Treisman Reference Treisman, Russell, Green, Dornic and Rabbitt1975), is not disturbed by a light mask that disturbs positive afterimages (Sligte et al. Reference Sligte, Scholte and Lamme2008), and is disturbed by a pattern mask that does not disturb retinal afterimages (Sligte et al. Reference Sligte, Scholte and Lamme2008). All these data provide reason for thinking that the phenomenal persistence is due to a persistence in the underlying basis of phenomenology that is not being driven by an earlier persistence. For these reasons, iconic memory is, as Landman & Sligte note, a window into phenomenal consciousness.

R3.1. Are reports privileged?

Naccache & Dehaene pose a dilemma for me: “If one wants to define phenomenal consciousness differently from conscious reportability, then one should resist the temptation to make use of subjects' reports” as evidence for phenomenal consciousness. However, no one would think there is such a dilemma for, say, acidity or heat. An acid is a proton donor and heat is molecular kinetic energy. These are good scientific definitions, but no one thinks that these definitions preclude any kind of evidence. Naccache & Dehaene see inconsistency looming: How can I rely on reports in the Sperling experiment while at the same time claiming that subjects' reports that they don't see something can be wrong? Prinz maintains that reports are “authoritative.”

Vanhaudenhuyse et al. exhibit the ingenuity of researchers in finding ways that subjects who have no control of motor and verbal systems – and hence no capacity to make a motor or verbal report – can nonetheless report conscious states by, for example, thinking of milk rather than lemon. This point does not, however, challenge Dehaene & Naccache, since it does not show any daylight between consciousness and possibility of reporting it. My view is that reports are the starting point for building a theory of phenomenal consciousness, but can be rejected if the best explanatory theory requires it. The arguments for this view include the methodological points about inference to the best explanation, the sketch of an actual explanatory account on which reports can be wrong, and the points made forcefully by Snodgrass & Lepisto (and in Block Reference Block2005) involving signal detection theory.

It is obvious that reports fail to be authoritative in that we can have conclusive evidence against the truth of introspective reports. As Koch & Tsuchiya note, in Anton's syndrome, subjects are blind but think and report that they see. More generally, anosognosics deny their perceptual and motor disabilities, making all sorts of false reports about their own experience. Introspective reports do have a certain priority: we have no choice but to start with reports in investigating consciousness. I am in complete agreement with Koch & Tsuchiya when they say, “in the absence of compelling, empirical evidence to the contrary… if the subject denies any phenomenal experience, this should be accepted.” One very notable form of empirical evidence that can conflict with report is evidence about subjects' decision processes evaluated according to signal detection theory. As Snodgrass & Lepisto note, “contrary to many researchers' implicit assumptions, there is no such thing as an unmediated ‘subjective report’ – ever.”

In practice, Naccache & Dehaene use the methodology I advocate, not the one they themselves advocate. They concede that reports can be partial and biased and argue that “reports of a rich phenomenality cannot be taken at face value,” citing the moving-window paradigm and the claim that we have the illusion of seeing a world in full color. Further, both Naccache & Dehaene and Sergent &Rees suppose that reports have to be measured using high-tech machinery, leading to a concern about what they could mean by report. Is a report just (1) any evidence of consciousness? Or is it (2) evidence of consciousness from a subject's performing one voluntary action rather than another? Or is it (3) evidence of global broadcasting? If (1), these commentators' focus on report as the gold standard for evidence of consciousness is trivial, because it just means that evidence is the gold standard of evidence. If (3), their claim that the evidence for consciousness is always evidence for global broadcasting is trivial, amounting to the claim that evidence for global broadcasting is evidence for global broadcasting. If (2), the claim that reports are the gold standard is substantive but probably false. As Malach has shown in a previous study (see Hasson et al. Reference Hasson, Nir, Levy, Fuhrmann and Malach2004), you can get evidence about consciousness from people just watching a movie in a scanner and not making any voluntary response. In another such case mentioned by Sergent & Rees, Lumer and Rees (Reference Lumer and Rees1999) put subjects in a binocular rivalry experiment without requiring any voluntary response and found alternation between neural activations, providing evidence of the neural bases of different perceptual experiences without a voluntary response. An advantage of sticking to normal uses of words is that we don't have to guess what people mean.

R3.2. Panpsychic disaster

A number of commentators argue that once you give up the special authority of reports, you will have no way of avoiding attributing consciousness to lampposts. Papineau notes that I regard some states as uncontroversially unconscious and wonders, “what makes a state ‘uncontroversially unconscious’ if it is not that subjects tell us so.” He argues that once we allow that a state can be conscious even though normal subjects systematically deny it, there may be no uncontroversially unconscious states. Prinz says: “Block must either concede that reports are authoritative or deny that we can rule out the possibility of conscious states in V1, the LGN, and the retinae.” Lau &Persaud say that the methodology I rely on to support recurrent loops as necessary assumes that subjects are to be believed when they report that they don't see something. But given that I allow for phenomenology without access, Lau & Persaud wonder, how can I rule out that these subjects have phenomenology despite reporting that they don't? Sergent & Rees and Naccache & Dehaene would no doubt agree. But these critiques ignore the obvious failure of authoritativeness of reports just mentioned and the power of inference to the best explanation to sort the good from the bad attributions of consciousness.

Perhaps what Naccache & Dehaene and Sergent & Rees are supposing is that if biomarkers of consciousness conflict with actual verbal report, verbal report wins. But there is no evidence that this always wins. If items of evidence conflict, the right scientific procedure is to find an explanation of the conflict, doing further experiments if necessary to see which item of evidence is misleading.

R4.1. Awareness, access, and agency

Levine notes that the idea of a phenomenal consciousness that has nothing to do with any kind of access-relation to the subject doesn't really seem like any kind of consciousness at all. A major difference between my position in the target article and Block (Reference Block1995b) is that in the target article I acknowledge this point and accommodate it within a framework that avoids any constitutive connection between that access-relation (which I called Awareness in the target article and Levine calls awareness-access) and cognitive access, which I identify in the target article with broadcasting in the global workspace. Jacob wonders why I have changed my mind, noting that the kind of awareness I now say is a necessary part of conscious experience is not full-fledged self-awareness of the kind a rational thinking creature might sometimes have but that a mouse will presumably not have. In 1995, the only option I saw for explaining awareness-access in non-cognitive terms was as a kind of phenomenal property I called “me-ishness.” But now I see that awareness-access can be adequately understood in terms of “same-order” and deflationary theories, so there is no need for cognitive or other “higher-order” accounts.

Levine makes a plausible case that the resistance many feel to the idea that the machinery of phenomenal consciousness is separable from the machinery underlying report (i.e., broadcasting in the global workspace) stems from conflating broadcasting-access with awareness-access; and he mentions two interesting suggestions for ways in which the relation between awareness-access and the self might be further elucidated. One of them is that there is no reason why subjectivity cannot involve a fragmented subject. Yes, but the subject cannot be so fragmented that the experience is not for-the-subject. If G.K. has the face experience on the left that he denies, what keeps him from acknowledging it is his inability to attend to the left side of space. But that is compatible with the face experience on the left being part of the same visual field as whatever he is seeing on the right. (I use this term, “visual field”, as is standard in philosophy to mean the entirety of visual space, including left and right.) That is, the relations in the visual field would be normal and have normal awareness-access, but the subject would have poor broadcast-access to them. So we can make sense of the idea that experience of the face on the left is an experience for him despite some fragmentation of the self.

Levine's hypothesis illuminates Rosenthal's critique, in which Rosenthal asks: If phenomenology necessarily involves awareness as I say it does, and given that awareness normally results in cognitive access, why would cognitive access fail in the G.K. case? Rosenthal is assimilating or at least supposing too tight a connection between awareness-access and broadcasting access. G.K. has awareness-access on the left if he has phenomenology on the left, but it is his lack of broadcasting-access that explains why he cannot report what is on the left. Rosenthal claims that higher-order thought theories of consciousness (known as HOT theories) explain why ordinary phenomenology always involves some awareness of it, whereas the global-workspace theories I favor as theories of access do not. But the global workspace theories are theories of cognitive access, not theories of awareness-access. Again, Levine's hypothesis is confirmed. The competition to higher-order theories as theories of awareness-access is offered by the same-order theory and the deflationary theory, not by the global workspace theory.

Rosenthal suggests that infant pain might have “weak” phenomenology and that weak phenomenology may be what I am supposing G.K.'s face experience has as well. I am not sure what “weak” phenomenology is supposed to be, but I guess it stands for the phenomenology you get when there is no global broadcasting and no higher-order thought. There is no reason to suppose that infant pain or G.K.'s face experience is any less intense than adult pain (cf. Gopnik and Izard et al.). Gopnik adds another layer of evidence to what I presented in the target article about infant phenomenology and Malach adds yet another layer relevant both to infant and adult phenomenology. Endogenous attention matures late, making it even less likely that infants are capable of higher-order thought. Inhibitory neurotransmitters, a major contributor to unconscious states, mature late, making it less likely that infants' perceptual states are unconscious. Younger infants are slower to habituate, another source of unconscious perceptual states (although this may be because younger infants are slower to encode stimuli rather than anything to do with the machinery of habituation). Evidence against both higher-order thought and unconscious states puts Rosenthal in a vise, since these states exhaust his options. Izard et al. also give suggestive evidence that emotion and action systems reveal phenomenal states that may not involve either global broadcasting or higher-order thought. The upshot is to increase the empirical squeeze on higher-order thought theories of consciousness, thereby increasing the attractiveness of same-order theories.

Rosenthal says standard same-order views are just as cognitive as HOT. Rather than argue about the texts he cites, let me just say that Caston's influential paper on Aristotle's same-order theory (Caston Reference Caston2002) emphasizes some of the advantages of the same-order view over higher order theories that I mentioned in the target article, namely, that there is no need to postulate that whenever there is a token conscious pain, there is also a token thought about it; and that there is no puzzle about why my thought about a teacup or your pain does not make the teacup or the pain conscious, whereas a higher-order thought about my own pain does. (See also Malach's Figure 2, part B, which depicts a way of thinking about the same-order account.)

Astonishingly, Rosenthal claims in regard to infant pain that, “because nonconscious pains have largely the same causal connections to behavior and to a strong desire for pain to cease cease, they are just about as bad.” It is well known that pain in infants has bad effects, but I have never heard of any experimental evidence to the effect that those bad effects do not depend on whether the pain is conscious or not, nor does Rosenthal cite any. Newborns who are circumcised without anesthesia or analgesia are more stressed by vaccination even 6 months afterward (Taddio et al. Reference Taddio, Goldbach, Ipp, Stevens and Koren1995; Qiu Reference Qiu2006). But this evidence does not approach the issue of whether the infant pain is conscious. I imagine that Rosenthal is taking the evidence that higher-order thought is unlikely in infants, and the evidence that pain in infants has bad effects, to come to the conclusion that unconscious pain in infants has bad effects. But this response suggests that Rosenthal is retreating from a substantive empirical claim to a trivial linguistic claim. Rosenthal can, if he likes, simply use the word “unconscious” to mean something on the order of unaccompanied by higher-order thought. But pains that are unconscious in that trivial sense cannot be supposed to be pains such that there is nothing it is like to have them. The danger of promoting such a triviality into a substantive thesis can be seen in Peter Carruthers' infamous claim (Carruthers Reference Carruthers1989; Reference Carruthers1992) that, given that pains in dogs, cats, sheep, cattle, pigs, and chickens are not available to be thought about, these pains are not felt and hence are not appropriate objects of sympathy or concern and are of no moral significance. (Carruthers [Reference Carruthers1999] backpedals, not on the ground that animal pains are appropriate objects of concern or sympathy, but on the ground that the frustration of animal desires are of moral significance.) Drug testers and slaughterhouses take notice! Given the evidence presented in the target article and in Gopnik and in Izard et al., I really don't see how advocates of this unattractive view can avoid applying it to human infants.

Snodgrass & Lepisto give a very plausible argument concerning Jacoby-style “exclusion” experiments (Debner & Jacoby Reference Debner and Jacoby1994) – that they involve phenomenally conscious stimuli or parts of stimuli which the subject has little confidence in having seen, and hence no higher-order belief in having seen it. (There was also an argument to this effect in Block Reference Block2001.) For example, the subject is instructed to complete the stem “rea_” with an ending that is not a word that might have just been flashed briefly. If the word is “reason” and the subject saw all or part of it, but has low confidence, the subject is more likely to complete “rea_” with “son” than if no word at all was flashed. As noted in Snodgrass (Reference Snodgrass2002), subjects who are penalized for errors do better at excluding, suggesting on a signal detection analysis that they really did see the word (or parts of it, as suggested in Kouider et al.) but had low confidence and no higher-order state that would lead to a report. This provides another piece of empirical evidence against the higher-order point of view. Perhaps the advocates of the higher-order point of view will take the degree of confidence to be an index of degree of consciousness, but the signal detection models show that consciously seeing the stimulus and confidence that one has seen it can vary independently, each being influenced by a variety of different factors. (For example, changing the catch rate can influence confidence without influencing perception. See Supèr et al. [Reference Supèr, Spekreijse and Lamme2001a] and Block [Reference Block2005].)

Clark & Kiverstein argue for a view opposed to Rosenthal and to Levine and me, that for a representation to be phenomenal, it must be “available for use in the planning and selection of deliberate, stored-knowledge–exploiting, and goal-reflecting and goal-responsive, actions.” Clark & Kiverstein makes a cogent case, but it doesn't involve any empirical evidence. Rather, it is an a priori conceptual appeal. “To count as a conscious experience, an informational state must” be available for “rational choices and considered actions.” One and the same condition of the skin, intrinsically described, will count as sunburn if it is caused by the sun, but not so if it is caused by a fire. Similarly, according to Clark & Kiversteins's view, one and the same neural-informational state, intrinsically described, can count as a conscious experience if it is properly connected to rational choices and considered actions, and not so if it isn't. But our ordinary concept of consciousness is too vague to support this kind of claim. Further, one reasonable precisification of our ordinary concept of consciousness will make it the concept of a natural kind. The key feature of that natural kind-property is the way it feels to have it. My own view is that there is an ordinary concept of phenomenal consciousness. Gopnik gives evidence that infant consciousness may “not be accessible for goal-directed planning nor be the subject of the endogenous attention that accompanies such planning.” I don't give this as evidence against Clark & Kiverstein's view since conceptual claims are not subject to evidence. Instead, if Gopnik is right, Clark & Kiverstein's conceptual claim would dictate that our concept of consciousness does not apply to infants. But the absurdity of this step strongly indicates that our concept of consciousness – or at least one of our concepts of consciousness – is the concept of a phenomenal feel that has no conceptually necessary relation to agency of the sort Clark & Kiverstein describe.

Malach presents exciting evidence that intense experience such as watching an engrossing movie (a spaghetti western?) activates an “extrinsic” sensory system in the same ways in different people; but it activates an “intrinsic” system based in frontal areas (Hasson et al. Reference Hasson, Nir, Levy, Fuhrmann and Malach2004) less strongly and not in the same ways in different people. Malach's intrinsic system overlaps considerably with a “default” system that is active when the subject is “doing nothing” and is inactive when the subject is performing intense goal-directed tasks (Raichle et al. Reference Raichle, MacLeod, Snyder, Powers, Gusnard and Shulman2001). In one study, Malach and colleagues (Goldberg et al. Reference Goldberg, Harel and Malach2006) used the same stimuli (pictures and audio clips) in an introspective task and a difficult rapid categorization task – but with different instructions. In the introspective task, subjects were asked to categorize their emotional reaction as positive, negative, or neutral. In the categorization task, the stimuli were presented at triple the rate and the subjects were asked to categorize the stimuli, for example as animals or not. Subjects also rated their own self-awareness. Their self-awareness was high in the introspective task and low in the rapid categorization task. Part of the intrinsic system – see Malach's Figure 1 – was activated both in the introspective task and other “self”-relevant tasks, including judging the application of emotionally neutral words to oneself. This area is suppressed in the rapid categorization task. The upshot is that the intrinsic system is less active when the subject is doing a demanding perceptual task, or watching Clint Eastwood, and that a part of it that is especially tuned to self-relevant tasks is especially suppressed in difficult perceptual tasks. This study presents a challenge to views such as Lycan's, Harman's, Rosenthal's, and Van Gulick's that emphasize the importance of the self in phenomenal experience because it provides evidence that the self is lost in intense perceptual activity.

R4.2. Fractionation of access

As I have already mentioned in commenting on Clark & Kiverstein and Rosenthal, the evidence presented in Gopnik and Izard et al. is highly relevant to the issues of higher-order thought and agency that connect with consciousness. However, I want to make a few more remarks about developmental evidence.

1. 1. The main methodological point made by Gopnik and Izard et al. is well taken. Adult capacities fractionate in infants. By studying that fractionation and the process by which the distinct capacities and abilities are integrated, we can learn a lot about adult consciousness and cognition.

2. 2. Much of the behavior that seems at first glance to be a reflection of cognitive access in infants may bypass global broadcasting altogether. One example from adults mentioned by Izard et al. serves to illustrate the point: the example concerns dorsal visual system guidance of action as exemplified in Goodale and Milner's famous patient D.F. who can post a letter in a variable angled slot about as well as normal persons and avoid obstacles, but who does not have the globally broadcast visual contents that are created by the ventral visual system. D.F. knows about the angle of the slot only by noticing her own actions (Milner & Goodale Reference Milner and Goodale2008). The data presented by Gopnik and Izard et al. might make one wonder about the extent to which infants even have global broadcasting. But young infants – at least by 10 months of age – do show working memory systems that function in much the way those of adults do, albeit with a slightly lower capacity (Feigenson & Carey Reference Feigenson and Carey2003; Reference Feigenson and Carey2005; Feigenson et al. Reference Feigenson, Carey and Hauser2002).

3. 3. I agree with Gopnik's cluster of points having to do with the preponderance of exogenous (stimulated from outside) over endogenous (internally directed) attention in infants. However, I am not sure that this fact suggests that infants have a lantern of consciousness instead of a spotlight. Why can't exogenous attention focus the spotlight?

4. 4. Although I agree in general with Gopnik's point that infant cognition is not geared towards goal directed planning in the early months, infants are certainly capable of goal-directed action by age 5 months, and there is evidence of some capacity for goal-directed action (and perception of it) at age 3 months (Sommerville et al. Reference Sommerville, Woodward and Needham2005).

R4.3. Kinds of access

Representationism is the view that the phenomenology of an experience is the experience's representational (intentional) content. Harman, Jacob, Lycan, and Tye are advocates of this view; I am an opponent. However, I am willing to allow that every phenomenological state has representational content, even that the phenomenology consists – in part – in its having that representational content. I think that phenomenology goes beyond representational content, but this opposition played no role in the target article and will play no role in this reply. With this bone of contention cleared away, Harman's argument can be streamlined.

Harman argues that clearing up some unclarities renders the claim that phenomenology does not require cognitive access non-scientifically true or non-scientifically false. In my weight-lifting this morning, I experienced “flow.” Among my visual experiences was seeing the barbell. Is the cognitive access in question with regard to the experience of seeing the barbell or with regard to the barbell itself? If the cognitive access in question is with regard to the experience of seeing the barbell, then we know from nonscientific observations of flow (Csikszentmihalyi Reference Csikszentmihalyi1975) that cognitive access to the experience itself destroys the experience, or at least its flow aspect. However, cognitive access to the barbell itself is guaranteed by the experience being an experience of the barbell. In neither case, Harman concludes, is there any need for science.

Recall that my target article agress with the idea of access as global broadcasting accepted by Shanahan & Baars and Naccache & Dehaene. What is globally broadcast is a representation with both representational and phenomenal content. What makes the access cognitive is that the consuming systems in terms of which global broadcasting is defined include the cognitive machinery of reasoning, planning, problem solving, categorization, memory, and the like. Harman assumes that what makes access cognitive is just that the accessed state is about something. But a frog might have a perceptual state that is about a fly without engaging any cognitive machinery. Let's go back to the first step of Harman's argument, that access has to be to the experience or to the barbell. In the framework I was using, access is neither to the experience nor the barbell but to the content of the state, since that is what is broadcast. So Harman's dilemma leaves out the crucial case. It may seem as if the issue between me and Harman is verbal – what is meant by cognitive and access. However, I chose the global broadcasting framework because it seems the best empirical framework for thinking about access and accessibility. Within that framework, the issue of phenomenology without access becomes an empirical issue. So, contrary to Harman, the issue is doubly scientific. Scientific considerations go into choosing the terms of the debate, and then, once they are chosen, scientific considerations determine the answer.

Balog, like Harman, is concerned with the question “Access to what?” She mentions, in effect, three options: global broadcasting of either (1) the nonconceptual contents of perception; or (2) the conceptual contents of perception; or (3) the introspective conceptual contents that are concerned with the phenomenal character of the experience itself. Option (3) involves phenomenal concepts, that is, concepts whose tokening itself involves phenomenology. (See papers in Alter & Walter [Reference Alter and Walter2006].) Balog argues that my interpretation of what subjects in the overflow experiments say “crucially relies on the subjects' introspective report of the phenomenality of their entire visual experience, including those aspects of the experience whose content is not access conscious” (in the sense of option 2; Balog's emphasis), and this motivates her to suggest a sense of access (as in option 3) in which there is no gap between phenomenality and access.

Balog says my view of access involves the second option described earlier – that I define it in terms of conceptual contents. There is some justification for that in my 1995 paper (Block Reference Block1995b, see especially Note 11), which was written before I had adopted the global broadcasting model of access. My model of access in 1995 was dominated by inference and other reasoning, and it is often taken as part of the definition of nonconceptual content that it can play no role in inference. However, once I had adopted the global broadcasting model (Block Reference Block and Chalmers2002) the issue of whether the broadcast contents were conceptual or nonconceptual no longer looms large. I don't see why broadcast contents cannot be a mix.

Balog is responding to a problem that comes up repeatedly in the commentaries (especially Naccache & Dehaene's and Sergent & Rees's): How can I use reports to demonstrate inaccessible (including non-reportable) phenomenology? My reply is that in some cases, one infers unreportable or unreported phenomenal states via theoretical considerations that are themselves based on reports. In the overflow experiments, the situation is this: Generic contents are reportable, and one of those generic contents is that the subject has had specific phenomenology. That generic content need not deploy any phenomenal concepts in the sense of a concept whose tokening itself involves phenomenology. So there is no need to appeal to phenomenal concepts.

Tye notes that in one sense of the term accessible, phenomenology does not overflow accessibility. The sense he has in mind is that for each letter that the subject sees, the subject can demonstrate it with, for example, “that letter.” He thinks that this point deprives my claim that phenomenology overflows accessibility from its “bite.” Recall that I distinguished between wide and narrow accessibility. In my terms, what Tye has done is to provide another wide sense of accessibility – demonstration as opposed to identification, as in Sperling, or comparison, as in Landman et al. and Sligte et al. Demonstration, identification, and comparison are all different and have different cognitive demands. Demonstration is in a way more demanding (see Koch & Tsuchiya) because, whereas identification can take place without attention or with limited attention, demonstration requires attention. As I noted in the discussion of Spener, the grain of vision is finer than the grain of attention – you can see more than you can attend to. But regardless of whether Tye's version of wide accessibility is wide enough for his purposes, my point is about something different. The overflow point in the target article was this: The Sperling, Landman et al., and Sligte et al. experiments show that the phenomenal system is at least in part distinct from the global workspace because the capacity of the phenomenal system is greater than that of the global workspace. This point can be defended without being concerned with different senses of “accessibility,” so I do not agree that providing another wide sense of accessibility affects the bite.

R4.4. Phenomenology and working memory

Lau & Persaud suppose that the apparent difference in capacity between phenomenology and working memory is just a consequence of deterioration of information, as in the “Telephone” or “Chinese whisper game” in which information is whispered from one person to another, becoming ever more garbled. However, what decreases in the Chinese whisper game is quality or amount of information, not capacity of an informational system. Lau & Persaud note that there are cases in which forced-choice responses overestimate phenomenology (blindsight) or yield inconsistent measures of phenomenology. They conclude that forced-choice reports are not ideal for measuring phenomenology. My replies to this are as follows:

1. 1. Forced choice measures in the Landman et al. and Sligte et al. experiments are measures of working memory, not phenomenology. Subjects' ability to make the comparisons depends on the representations being present in the global workspace.

2. 2. The Sperling version of the partial-report superiority effect used free recall, not forced choice as in Landman et al. and Sligte et al., and got similar results. It is the use of different methodologies with the same results that avoids difficulties of the sort that Lau & Persaud are mentioning.

Shanahan & Baars say their aim is to defend GWT (Global Workspace theory) from my challenge. More specifically, they say I use an outdated and imprecise concept of access and perpetuate a common misunderstanding of GWT that conflates the workspace with working memory. My main disagreement with them is simple: they say what is broadcast in the global workspace is all there is to consciousness, whereas I argue that it is an empirical issue whether there is more to consciousness and that the tentative answer is, Yes, there is. Shanahan & Baars' remarks do not seem to speak to this main issue. On the “misunderstanding” of working memory: the issue here is purely verbal. I use the term working memory to mean what Baars and Franklin call active working memory (Baars & Franklin Reference Baars and Franklin2003). One of the first things that strikes a reader of the “working memory” literature is that the term working memory is used differently by different theorists. Cowan (in press) notes “Working memory has been conceived and defined in three different, slightly discrepant ways.” It would take too long to explain all three ways, but what I can say briefly is that a pretty standard model (Cowan Reference Cowan, Izawa and Ohta2005) includes three ingredients: (i) an activated part of long-term memory (often called “short-term memory”), (ii) attentional processes that operate to broadcast perceptual and short-term memory contents, and (iii) what is broadcast in the global workspace. I tend to use short-term memory for the activated part of long-term memory, and working-memory for what is broadcast in the global workspace; but judging from their commentary, Shanahan & Baars prefer to use both short-term memory and working memory for what I call short-term memory. The bottleneck of working memory on any definition I have seen, including that of Shanahan & Baars, is the capacity of the global workspace, so if that bottleneck is what is of interest, my use of working memory is apt. Whether or not I am attacking Global Workspace theory depends on whether Shanahan & Baars' controversial ideas about consciousness are part of it. Baars gets credit for many of the original ideas about the global workspace, but now that the ideas have been widely adopted, he no longer gets to dictate the terminology. My overflow argument could be put entirely in terms of global workspace terminology, excluding all talk of memory. Here it is: The capacity of phenomenology is greater than the capacity of the global workspace, so phenomenology must contain machinery not part of the machinery of the global workspace.

R4.5. Report and color

Kentridge describes an intriguing case: Cerebral achromatopsics can recognize borders without being able to distinguish the colors on either side of the borders. Of course, guitar strings that are less than a JND (just-noticeable-difference) apart in pitch create interference “beats” that allow one to tell that they are different. And color patches that are less than a JND apart can be discriminated by normal subjects if they share a border. As I understand it, Kentridge et al. (Reference Kentridge, Heywood and Cowey2004) rule out such explanations, so phenomenal presentation of colors that the subject cannot report is a live option in the case that Kentridge describes. This case is as much a challenge for my views as an opportunity, since there are issues regarding how one can tell whether Kentridge's patient M.S. can see colors that he cannot report, and whether he has some kind of unconscious appreciation of color differences without phenomenal presentations of the colors themselves.

R5.1. Phenomenology and attention

Lycan asks me to explain how there could be a form of awareness that is intrinsic to consciousness but does not necessarily involve attention or cognitive access. He notes (and I agree) that any form of awareness (worthy of the name) is representational and that there can be representation without attention. However, he does not see how cases of representation without attention could fail to be sub-personal. According to Lycan, representing by the whole person has to involve attending, or at least he does not see how it could not involve attending. The claim that representing by the whole person has to involve attention is in part empirical and I believe the evidence goes against it.

One kind of evidence derives from unconscious perception, more specifically perception that is unconscious because the person is not paying attention. To pick one of many examples: In visuospatial extinction, as mentioned in the target article, subjects often claim not to be able to see something on the left if there is a stimulus that competes for attention on the right. But as first noted by Volpe et al. (Reference Volpe, LeDoux and Gazzaniga1979) (and nailed down by Verfaellie et al. Reference Verfaellie, Milberg, McGlinchey-Berroth, Grande and D'Esposito1995), the subject's claim not to see the object on the left is often combined with the ability to make comparisons between the thing on the left and the thing on the right. The thing on the left is certainly seen and, it is usually supposed, unconsciously so. Unconscious seeing is still seeing, and seeing by the person, not by a part of the person. This is seeing and therefore representing without, as far as we can tell, attention (whether or not the seeing on the left is unconscious). Therefore, the principle that Lycan can see no alternative to no personal-level representation without attention, is empirically suspect.

In the discussion of Spener, I mentioned another type of evidence against the claim that there is no personal level representation without attention, namely, that one can see more than one can attend to. Seeing the speckles on the speckled hen is personal and phenomenal, but there is no attention to the individual speckles; the individual speckles are conscious but unattended.

Lycan (Reference Lycan1996; Reference Lycan and Gennaro2004) follows Locke (Reference Locke1689/1975) and Armstrong (Reference Armstrong1968) in holding that what makes a state conscious is that it is the object of an “inner sense,” which Lycan seems to think is to be explained in terms of attention. The point I just made is not directly relevant to that view, but rather to supporting it in terms of personal-level representation requiring attention. However, as noted by Koch & Tsuchiya (and Koch & Tsuchiya Reference Koch and Tsuchiya2007), conscious gist perception of peripheral non-targets persists for certain kinds of stimuli even when attention is maximally drained off by a demanding central task. So there is direct evidence against the attentional version of the inner sense thesis as well.

In cases of inattentional blindness, subjects say they do not see the stimulus – and this contrasts with the Sperling and the Landman et al. cases in which subjects clearly do see the stimulus. Relying on this contrast, Prinz and Grush are convinced (against Naccache & Dehaene) that phenomenology outstrips actual broadcasting in the global workspace, but, like Snodgrass & Lepisto, they think I go too far in supposing that perhaps phenomenology outstrips even potential broadcasting, as in the G.K. case. Prinz, like Lycan, argues that the flaw in my position is to allow phenomenology without attention. Prinz's account is based on the idea that attention is necessary for phenomenology and for accessibility in the sense of potential for broadcast in the global workspace (Prinz Reference Prinz2000; Reference Prinz, Brook and Akins2005). On both issues, I think the evidence is piling up against Prinz's view. First, on the claim that attention is necessary for accessibility, that is, potential broadcast: Dehaene and his colleagues (Dehaene Reference Dehaene, Changeux, Naccache, Sackur and Sergant2006; Kouider et al. Reference Kouider, Dehaene, Jobert and Le Bihan2007) have convincingly shown that in conditions in which attention is maximally drained away by other tasks, representations that are as unattended as can be ensured by such conditions are nonetheless very strongly activated. They call this category “preconscious” as opposed to “unconscious” to indicate the strength of the activations, their strong influence, and their potential to be globally broadcast. Tse et al. (Reference Tse, Martinez-Conde, Schlegel and Macknik2005) have obtained similar results. These representations are accessible in that with a shift of attention, they will be accessed. Regardless of whether or not attention is necessary for actual access, it is not necessary for potential access, contrary to Prinz.

On the more important issue of the relation between attention and phenomenology, as Koch & Tsuchiya note, the evidence points toward the conclusion that neither is necessary for the other (Koch & Tsuchiya Reference Koch and Tsuchiya2007). On attention without phenomenology, subjects' attention is drawn by nudes of the opposite sex (for heterosexuals) in conditions in which the nudes are invisible according to experimental standards (Jiang et al. Reference Jiang, Costello, Fang, Huang and He2006). (Whether the invisible pictures attract attention to one side of the visual field or the other is judged by performance on a subsequent task to which distribution of attention is crucial.) Kentridge et al. (Reference Kentridge, Heywood and Weiskrantz1999) have shown the effects of attention in blindsight. On phenomenology without attention, again in a “dual task” paradigm, subjects are able to see and report certain kinds of scene gists (e.g., the presence of a face in the periphery) despite maximal siphoning off of attention by a very demanding task. Work by Alvarez and Oliva (in press) suggests that without attention (or much attention) conscious visual representations represent much more abstract properties of stimuli, for example, center of mass rather than shape. As Koch & Tsuchiya (Reference Koch and Tsuchiya2007) point out, it is difficult to make absolutely sure that there is no attention devoted to a certain stimulus; but given this limitation, this evidence points away from both of Prinz's claims. And the same point about the speckled hen case I made in discussing Spener and Lycan applies also to Prinz.