Do beliefs, desires, intentions, or other aspects of our cognitive lives influence how we perceptually experience the world? Put differently, is perceptual experience cognitively penetrated? The nomological possibility of such influences seems relatively uncontroversial. Recently, however, philosophers have concerned themselves with actual instances of such influence. Macpherson (2012) calls attention to color matching tasks to argue that beliefs relating to the characteristic colors of objects influence the perceptual experience of hue. Similarly, Stokes (2012) invokes New Look studies to argue that desire-like states influence perceived size. More subtly, Wu (2013) argues that intentions influence the experienced visual stability of objects during eye movements.
Needless to say, arguments in favor of actual instances of cognitive penetration have been challenged. There is a general schema for such debates. The proponent of cognitive penetration offers a particular example and claims that the standard ways of explaining away cases of cognitive penetration do not apply. The skeptic suggests that the proponent has overlooked some alternative explanation that effectively explains the example without entailing cognitive penetrability. Finally, a range of studies commonly cited by philosophers have been challenged on more basic methodological grounds (Firestone and Scholl, 2016). Such methodological worries provide additional reason for skepticism.
Visual attention has been traditionally excluded from the cognitive penetration debate. Recently, however, this exclusion has been criticized (Mole, 2015; Lupyan, 2017; Wu, forthcoming; Stokes, forthcoming). On one hand, I agree that such an exclusion is ill-motivated. On the other hand, it is not clear that the notion of cognitive penetration has remained constant across the history of the debate. I suggest that taking a closer look at attention’s role in top down processing sheds light on a valuable distinction between two forms of cognitive penetration. I claim that attention’s modulatory effects on perceptual experience can be understood in terms of a purely causal process that does not compromise the epistemic status of perceptual experience. This exemplifies a type of cognitive penetration―what I call “cognitive amplification”―that is distinct from the more familiar form of cognitive penetration discussed in the first paragraph of this paper―what I call “cognitive shifting.”
The structure of the paper is as follows. Section 2 reviews attempts to define the term ‘cognitive penetration.’ A theory-neutral approach which focuses on the consequences of cognitive penetration is endorsed as a precursor to a definition. Section 3 provides evidence supporting category-based attentional modulation of perceptual experience. Section 4 distinguishes cognitive amplification from cognitive shifting and shows that the attentional modulation described in section 3 is a type of cognitive amplification. Section 5 considers counterexamples to the generalization that cognitive penetration by attention is always an instance of cognitive amplification. The paper concludes by posing a series of open questions.
2. What is cognitive penetration?
Recent debates in philosophy and psychology concern cognition’s influence on perceptual experience. This is, to a degree, a distinct issue from whether there is a cognitive influence on perceptual processing. After all, there may be robust cognitive influences on perceptual processing that are not reflected in perceptual experience. One way of tightening the link between changes in perceptual processing and changes in perceptual experience is by considering only perceptual processing in neural regions that are a basis for perceptual experience.
What amounts to a case of cognition influencing perceptual experience? Concise definitions of cognitive penetration are often a source of controversy. However, certain cases are generally not counted as cases of cognitive penetration. The movement of one’s eyes, head, or body along with any changes in proximal stimulus are widely excluded as cases of cognitive penetration. Including these cases as instances of cognitive penetration would trivialize the debate, and one would be able to demonstrate cognitive penetration simply by intentionally closing one’s eyes.
Attention has, at least historically, been excluded from the debate as well. One motivation for such an exclusion stems from the notion that attention is a simple mechanism that operates pre-perceptually. Another motivation stems from the idea that attention is or is closely analogous to a spotlight mechanism that selects locations in one’s field of view. From this perspective, shifts in attention do not seem very different from instances of squinting or moving one’s eyes.
More recent discussions in the cognitive penetration literature have suggested that the exclusion of attention relies on faulty assumptions. Attention affects perceptual processing by enhancing spatial resolution and contrast sensitivity at attended locations (Carrasco, 2011). Likewise, the spotlight metaphor is unable to account for a range of non-spatial forms of attention, such as object- or feature-based attention (see Maunsell and Treue, 2006 and Scholl, 2001, respectively). In light of such evidence, an exclusion of visual attention from the outset of discussion seems unmotivated and potentially question begging.
For our purposes, a pre-theoretical understanding as to what counts as a case of cognitive penetration is more important than defining cognitive penetration. In fact, providing a definition or definitions at this point would be premature. The present aim is to distinguish two different varieties of cognitive penetration. This means that cognitive penetration is the explanandum, not the explanans. Simply giving two distinct definitions of cognitive penetration would not tell us why there ought to be two distinct definitions in the first place. Of course, two distinct definitions could be drawn from our discussion, but this would take place at the end of the inquiry, not the beginning.
For this reason, I recommend we shift our focus to the consequences of cognitive penetration. Stokes (2015) uses this approach to develop a definition of cognitive penetration. However, our aim can be more modest: we can use the consequences of cognitive penetration to guide our understanding of the phenomenon without the immediate aim of providing a definition. As Stokes notes, such an approach can help to provide a theory-neutral common ground for what counts as a case of cognitive penetration. These consequences are of concern to both the proponent and skeptic of cognitive penetration.
What are the consequences in question? First, cases of cognitive penetration have epistemic consequences (e-consequences) on observation, both in scientific and everyday contexts. If cognitive penetration occurs, then observation’s role in resolving theoretical or everyday disputes can be threatened. To take a toy example, suppose you and I disagree as to whether glass was dumped in the local bay thirty years ago. You claim that irresponsible sailors threw all their gin bottles into the bay while I deny this. Prior to any observation, our theories differ. In an attempt to resolve the dispute, we check if sea glass has drifted in from the bay. After a day of searching, you collect a handful of objects which you take to be sea glass. However, if cognitive penetration has occurred, then our perceptual experiences may differ. Thus, our observational beliefs―beliefs formed on the basis of perceptual experiences―may also differ. For instance, perhaps you experience the objects as having a frosty, kelly green surface, while I experience them as having a frostless, dull green surface. On the basis of your experience, you form the observational belief that you are holding sea glass. On the basis of my experience, I form the observational belief that you are holding plastic. In this situation, our observational beliefs are theory-laden and cannot adjudicate between the rival theories.
The e-consequences of cognitive penetration are not limited to disputes between individuals. As Siegel (2012) points out, cognitive penetration can result in epistemically pernicious effects for a single individual. If my antecedently unjustified belief that p influences my perceptual experience that p which comes to serve as further evidence for my belief that p, then I have, in a sense, used my unjustified belief that p to boost the epistemic status of my belief that p. In Siegel’s terms, “you’ll just be checking your beliefs against your beliefs” (ibid., p. 202). Note that the e-consequences of cognitive penetration need not be negative. Perceptual learning and perceptual expertise may amount to cases of epistemically beneficial cognitive penetration.
Second, cognitive penetration has, at least historically, had consequences for debates concerning cognitive architecture and modularity of the mind. Modular systems, such as color processing or face recognition systems, are associated with a range of properties detailed by Fodor (1983). For instance, modular systems are said to be domain specific, fast, automatic, mostly inaccessible by central systems, associated with fixed neural architectures, and so on. Nevertheless, Fodor takes informational encapsulation to be the essential characteristic of modular systems (ibid., p. 71). Informational encapsulation depends on the limits of information flow into a system. Informational encapsulation is a gradable and relational notion: a system can be more or less informationally encapsulated and can be informationally encapsulated from some but not other systems.
Cognitive penetrability is inversely related to the informational encapsulation of perceptual systems from cognitive systems. An instance of cognitive penetrability indicates that the information flow into some perceptual module is larger than previously anticipated. Therefore, widespread cognitive penetration would mark an important failure of informational encapsulation. For the Fodorian modularity theorist, this would entail the non-modularity of perceptual systems. For this reason, cognitive penetration has informational encapsulation consequences (ie-consequences).
On the basis of e- and ie-consequences, we can state a sufficient condition for cognitive penetration:
CONSEQUENTIALIST CONSTRAINT: If a tokening of the perceptual/cognitive relation of type A results in both e- and ie-consequences that are not due to movement or changes in proximal stimulus, then the tokening is an instance of cognitive penetration.
In sections 3 and 4, I suggest that taking a closer look at recent arguments in favor of cognitive penetration by attention highlights a distinction among two varieties of cognitive penetration. Extending the CONSEQUENTIALIST CONSTRAINT above, the argument relies on the following consequentialist premise:
SPLIT CONSEQUENCES: If perceptual/cognitive relations of type A and B: (a) both count as instances of cognitive penetration by the CONSEQUENTIALIST CONSTRAINT; (b) differ in their e-consequences; and (c) differ in their ie-consequences, then tokenings of A and B are instances of distinct types of cognitive penetration.
The plausibility of SPLIT CONSEQUENCES cannot be evaluated without further explanation of the differences mentioned in (b) and (c). We will return to SPLIT CONSEQUENCES in section 4.
3. Cognitive penetration by attention
Recently, Mole (2015), Lupyan (2017), Wu (forthcoming), and Stokes (forthcoming) have suggested that instances of top down attentional modulation amount to cases of cognitive penetration. In the spirit of these recent suggestions, I provide evidence for four claims in support of cognitively influenced, attentional modulation:
- There are non-spatial forms of attention.
- Attention is directed on a cognitive basis.
- Attention affects perceptual processing, not just pre- or post-perceptual processing.
- The changes in perceptual processing in (3) are reflected in perceptual experience.
Together, (1)-(4) entail:
ATTENTIONAL MODULATION: Non-spatial forms of cognitively directed attention affect perceptual experience by affecting perceptual processing itself.
Two brief preliminary remarks are in order. First, (1)-(4) only entail ATTENTIONAL MODULATION if the form of attention is kept constant across all four claims. For instance, if (1) is established by evidence in favor of the existence of feature- or object-based attention and (2)-(4) are established by evidence relating to spatial attention, then ATTENTIONAL MODULATION is not established. Second, ATTENTIONAL MODULATION provides evidence against the rudimentary view of attention discussed earlier according to which attention is a simple mechanism analogous to a spatial spotlight that operates pre-perceptually. Accepting ATTENTIONAL MODULATION challenges anything akin to a spatial spotlight model of attention and provides evidence that attention’s modulatory role in perception is not indirect. An acceptance of this rudimentary view motivates dismissing attention in debates concerning cognitive penetration, whereas a rejection of the view makes any such dismissal difficult (see Mole, 2015).
I now briefly review evidence for (1)-(4) by focusing on a series of studies relating to category-based attention by Gary Lupyan and colleagues. Lupyan and Spivey (2010) show that category-based attentional cues facilitate visual detection. Subjects are told to fixate on a cross at the center of the screen and click as soon as they detect a visual probe. After either an informative or uninformative auditory cue, a stimulus array appears. On some trials, a probe appears next to one of the stimuli whereas on other trials no probe appears. During these latter trials, subjects are instructed not to click. There is no evidence suggesting that these auditory cues result in a speed-accuracy trade-off. The central result of this experiment is that informative auditory cues decrease reaction times for the detection of probes located near stimuli of the same semantic category as the cue. For instance, if the valid auditory cue “attend to the two” is played prior to the display of an array of ‘2’ and ‘5’ symbols, a subject will detect a probe located next to a ‘2’ symbol significantly more quickly than she would have had she heard the uninformative auditory cue “attend to the category.”
Since the auditory cues provide no information concerning the spatial location of the probe, attention in these experiments is not an instance of selection or enhancement by location. Insofar as spatial attention is understood as involving the selection and enhancement of perceptual information on a spatial basis, this form of category-based attention is non-spatial, thereby supporting (1). Of course, this is not to suggest that such category-based attention operates independently from spatial attention. Indeed, category-based attention, along with feature- and object-based attention, can influence spatial attention in target detection tasks (Kravitz and Behrmann, 2011).
In addition, the informative auditory cues, such as “attend to chair” and “attend to two,” direct attention on the basis of linguistic representations. These cues would provide no attentional benefit without some possession of the corresponding concepts ‘chair’ and ‘two,’ respectively. These auditory cues are also, at least in some sense, stimulus independent since they do not involve the presentation of a particular object. For instance, the auditory cue “chair” does not involve the presentation of a particular chair. This means that an explanation of the cuing benefit in terms of perceptual priming is implausible. Therefore, given that such linguistic representations fall under the domain of cognition, this study also supports (2).
Furthermore, evidence in favor of (3) and (4) is given by Luypan and Ward (2013) who demonstrate that informative auditory cues presented prior to stimulus onset can also help to bring otherwise invisible, masked stimuli into awareness. Lupyan and Ward demonstrate that valid auditory cues (e.g. auditory cue = “pumpkin,” masked stimulus = pumpkin) result in increased detection sensitivity to masked stimuli relative to both invalid cue conditions (e.g. auditory cue = “sailboat,” masked stimulus = pumpkin) and no cue conditions. In fact, invalid cue conditions show decreased likelihood of stimulus detection relative to no cue conditions. In short, the likelihood of masked stimulus detection varied depending on the validity of the auditory cue.
Importantly, Lupyan and Ward’s study involves masking stimuli through the use of continuous flash suppression (CFS), a procedure by which a static stimulus presented to one eye is temporarily rendered invisible by a rapidly changing stimulus presented to the other eye. Indeed, it is the use of CFS that provides evidence in favor of both (3) and (4). With regard to (3), CFS heavily attenuates the amplitude of the N400 event-related potential (ERP) component, an electrophysiological measure of semantic processing, suggesting that stimuli masked using CFS receive little or no semantic processing (Kang et al., 2011). The main difficulty in establishing (3) is the possibility that category-based attention only affects post-perceptual processing, not perceptual processing itself. If a post-perceptual account were correct, then the auditory cues would alter post-perceptual judgments or memories related to semantically relevant stimuli. However, under such a post-perceptual account, little or no cuing benefits would be conferred to semantically unprocessed stimuli, such as those stimuli masked using CFS. Nonetheless, the results of Lupyan and Ward’s experiment indicate exactly the opposite of such a post-perceptual account prediction: valid auditory cues both increase sensitivity to and decrease detection times for semantically unprocessed stimuli.
With regard to (4), Lupyan and Ward’s results provide prima facie evidence that category-based attention results in changes in perceptual experience, and this interpretation is in line with the explicit self-reports of subjects in the experiment. However, one may challenge this interpretation on the grounds that Luypan and Ward’s experiment involves an instance of phenomenal overflow. In simple terms, phenomenal overflow involves phenomenal consciousness outstripping cognitive access (see Block, 2007). On what I call the phenomenal overflow interpretation, the valid auditory cues aid in the access of masked visual stimuli; nonetheless, such stimuli are present in phenomenal consciousness with or without the aid of the cues. According to this interpretation, the effects of category-based attention are reflected in cognitive access but not in phenomenal consciousness. If the phenomenal overflow interpretation were correct, then (4) would not met, at least assuming the term ‘perceptual experience’ in (4) is taken to refer to just the phenomenal aspects of perceptual experience.
One may resist the phenomenal overflow interpretation on the grounds that it relies on a distinction between phenomenal consciousness and cognitive access. Such a distinction is controversial and has been denied by proponents of the thesis that consciousness cannot be split from access (Cohen and Dennett, 2011) or availability (Prinz, 2012). However, due to the experiment’s use of CFS, the phenomenal overflow interpretation can be resisted even if one grants the phenomenal/access distinction. This is because CFS also appears to disrupt low-level visual processing of stimuli. Evidence for this comes from experiments demonstrating that CFS masking results in heavily diminished visual afterimages (Tsuchiya and Koch, 2005), where the presence of such afterimages is commonly taken to reflect low-level perceptual processing. Indeed, Lupyan and Ward demonstrate that the CFS technique used in their experiment results in a significant and reliable reduction in afterimage effects. Moreover, if CFS masking results in low-level processing disruptions, then it is plausible that a subject’s phenomenal consciousness of the masked stimulus is usually absent or heavily degraded without the help of category-based attention.
Suppose, for the sake of contradiction, that the phenomenal overflow interpretation is correct. That is, suppose that phenomenal consciousness remains static across valid/invalid/no cueing conditions. If the above considerations are correct, then since a subject’s phenomenal consciousness of the masked stimulus is usually absent or heavily degraded without the help of category-based attention, the subject’s phenomenal consciousness of the masked stimulus is usually absent or heavily degraded with the help of category-based attention. However, this does not accord with the positive self-reports of subjects in Lupyan and Ward’s experiment who usually claim that they are aware of the masked stimuli in valid cueing conditions. Thus, the phenomenal overflow interpretation is false.
One might be tempted to dismiss the self-reports of subjects in Luypan and Ward’s experiment. It may be the case that subjects’ phenomenal experiences in the experiment are similar to those of blindsight patients. Typically, blindsight patients can accurately detect visual stimuli in one half of their visual field even though there is no evidence that they are phenomenally conscious of such stimuli. Nonetheless, a crucial difference is that blindsight patients typically report a lack of awareness even when they accurately detect stimuli (Kentridge et al., 1999), whereas subjects in Luypan and Ward’s experiment do report an awareness of the stimuli. Thus, a rejection of the self-reports of the subjects would require that subjects in the experiment are mistaken about their own experiences. This is not an implausible assumption in general, and first-person reports of one’s own experiences are defeasible forms of evidence. For instance, we may misremember our experiences after significant delays or fail to notice significant changes in our experiences, such as in cases of change blindness. However, subjects in Lupyan and Ward’s experiment form immediate judgments concerning their experiences and are not required to form contrastive judgments on the basis of two or more of their experiences, as in the case of change blindness. Therefore, there is no obvious reason to question the veracity of the subjects’ first-person reports. Of course, some more general form of skepticism regarding such first-person reports could be endorsed, but this would be ad hoc, at least without independent motivation. For this reason, a rejection of the self-reports of the subjects is not particularly appealing. On that note, I proceed with the assumption that evidence on the basis of category-based attention supports (1)-(4) and that ATTENTIONAL MODULATION is correct.
Although empirical evidence regarding category-based attention is still forthcoming, I nonetheless speculate that an optimization process underlies ATTENTIONAL MODULATION. The tokening of certain cognitive states results in top down, category-based attention. In turn, category-based attention optimizes the processing of some stimuli over others, resulting in certain perceptual experiences rather than others. This optimization process may work by means of signal enhancement or noise reduction mechanisms. Indeed, there is evidence that top down attention involves both signal enhancement and noise reduction. Top down attention can result in signal enhancement by a multiplicative gain in the responses of neurons preferring the attended stimuli and can result in noise reduction by a suppression of the responses of neurons preferring distractor stimuli (Maunsell, 2015). In addition, top down attention can result in signal enhancement by increasing local gamma-band synchronization and noise reduction by decreasing low frequency synchronization in a range of visual regions (Noudoost et al., 2010). Finally, voxel-based analysis of fMRI data suggests that signal enhancement by category-based attention is also achieved by tuning shifts towards the attended category, thereby increasing sensitivity to the attended category (Çukur et al., 2013). Understanding ATTENTIONAL MODULATION in terms of an optimization process fits nicely with the more general suggestion that the “ultimate aim” of top down processing is the optimization of processing of incoming stimuli by perceptual systems (Teufel and Nanay, 2017, p. 23).
4. Cognitive shifting and cognitive amplification
In this section, I use SPLIT CONSEQUENCES to distinguish two types of cognitive penetration: cognitive amplification and cognitive shifting. Recall the claim given in section 2:
SPLIT CONSEQUENCES: If perceptual/cognitive relations of type A and B: (a) both count as instances of cognitive penetration by the CONSEQUENTIALIST CONSTRAINT; (b) differ in their e-consequences; and (c) differ in their ie-consequences, then tokenings of A and B are instances of distinct types of cognitive penetration.
In what sense could e-consequences differ? Recall that the e-consequences of cognitive penetration emerge from the epistemically beneficial or harmful cognitive influences on the perceptual experiences that serve as the basis for observational beliefs. Thus, e-consequences may differ if there are epistemically relevant differences in how perceptual experience is influenced by cognition.
On one hand, cognition might alter the epistemic status of perceptual experience by changing the content of our experiences. This can be spelled out if we take the content of perceptual representation to involve a singular aspect that serves to refer to contextually-specified, non-repeatable particulars and a general aspect that is responsible for the attribution of some type of kind, relation, or property to these particulars (Burge, 2010). Together, these two elements determine the accuracy conditions of the representation. We can symbolize the accuracy conditions for a given representation as ‘that F’ where “that” indicates the singular aspect while “F” indicates the general aspect (Block, 2014). A change in the general aspect while the singular aspect remains fixed can lead to a change in the epistemic status of a perceptual experience. In other words, a shift from the accuracy conditions as ‘that F’ to ‘that G,’ where the singular aspect remains fixed, can lead to change in the epistemic status of a conscious perceptual representation.
Indeed, this is what we saw in the sea glass example in section 2. In that case, my prior theory concerning sailors affected the content of my perceptual experience by influencing which colors and other surface properties were attributed to the objects in your hand. Suppose that the objects in your hand really did have a frosty surface. If the general aspect of the content of my perceptual experience consists in an attribution of frost-free surface properties because of my prior theory concerning sailors, then my prior theory compromises the epistemic status of my experience. This change in the content of my experience results in the theory-laden observational belief that you are holding plastic, not sea glass. Call e-consequences of this form experiential e-consequences since the epistemic status of perceptual experience itself is, for better or for worse, changed in such cases.
On the other hand, this is not the only way that cognition might influence perceptual experience. Cognition might bias perceptual experience without compromising its epistemic status qua experience. Extending an analogy given by Susanna Siegel, a sports magazine that exclusively prints front page stories on golf may demonstrate a bias towards golf, but it does not follow that those front page stories are factually inaccurate. In line with this analogy, Watzl (2017) suggests that the conscious mind may have a structure whereby some parts of consciousness are prioritized over others. Indeed, he suggests that this priority structure is the result of the activity of attention. If such a view is correct, then experiences with the same accuracy conditions may differ phenomenologically if their priority structures differ. If some such structures are the result of category-based attention, then cognition can influence experience without compromising its epistemic status.
Although I find Watzl’s thesis plausible, the point that cognition can bias perceptual experience without compromising its epistemic status does not depend on a thesis as robust as his. To see this, simply consider cases of selective looking. (Of course, selective looking is not cognitive penetration under the CONSEQUENTIALIST CONSTRAINT, but it nonetheless serves as a useful example.) In such cases, what one perceptually experiences is clearly biased by one’s beliefs or desires. Nonetheless, such biased perceptual experiences can be reliable indicators of those things that one selectively looks at. Indeed, if one is trying to determine whether a particular object is or is not present, then selective looking can lead one to form more accurate beliefs than unbiased, passive observation. That is, selective looking can be epistemically beneficial.
Nevertheless, such biases can also be epistemically pernicious. If one’s experiences are biased, then the observational beliefs formed on the basis of those experiences may reflect such a bias. Inductive generalizations formed on the basis of observational beliefs are justified only if one’s observational beliefs are somewhat representative of what one is exposed to. Therefore, inductive generalizations formed on the basis of such observational beliefs may not be warranted if experiences are biased. The crucial point is that in both the epistemically beneficial and harmful cases, the epistemic status of the experience itself need not be altered. For this reason, call e-consequences of this form downstream e-consequences since in these cases epistemic consequences need only occur downstream of perceptual experiences.
Let us now consider how ie-consequences might differ. Recall that ie-consequences relate to discussions concerning cognitive architecture and modularity of mind. Informational encapsulation, understood in terms of restricted informational flow into a system, is an essential characteristic of a Fodorian module. Widespread cognitive penetration would mark a failure of the informational encapsulation of perceptual systems from cognitive systems. Nonetheless, talk of “informational encapsulation” is ambiguous since there are, at least, two distinct senses of the term ‘information.’ Thus, ie-consequences can differ depending on the notion of “information” under consideration.
“Information” is sometimes understood in terms of causal covariation (Fodor, 1990; Orlandi, 2014). Suppose there is a source S with possible processes s1, s2, …, and sn and a destination D with possible processes d1, d2, …, dm. Under a causal covariation interpretation of “information,” S carries information about D if some S process si causes or would cause some D process dj. In other words, some processes in S causally covary with some processes in D. Under this interpretation, the amount of informational encapsulation of some system D from S depends on the degree to which processes in S causally covary with processes in D. The “degree” to which processes in S causally covary with processes in D may be a function of: the number of processes in S which causally covary with processes in D, the probability that any given S process si causally covaries with some D process dj, or the probability that any given D process dj was caused by some S process si (as opposed to some other S process or some non-S process). I set this matter aside for the remainder of the paper.
If cases of cognitive penetration demonstrate that processes in cognitive systems causally covary with processes in perceptual systems to a high degree, then the consequence is a lower degree of this causal form of informational encapsulation. Call reductions in this form of informational encapsulation causal ie-consequences.
A more robust notion of “information,” commonly referred to as “semantic information,” is often at issue in discussions of modularity of mind. Semantic information is often (non-reductively) understood in terms of semantic content. Under this interpretation of “information,” the amount of informational encapsulation of some system D from S depends on the degree to which processing in D is sensitive to semantic content from S. More specifically, if we take D as a system which computes a function from some characteristic input to some characteristic output, then the degree of this semantic form of informational encapsulation depends on how much semantic content from outside the system can be drawn on in computing values of the function for given inputs (cf. discussion of Pylyshyn, 1999 and Fodor, 2001 in Wu, 2013). If cases of cognitive penetration demonstrate that perceptual systems can directly or indirectly draw on the semantic content of cognitive systems in computing values of functions from their characteristic inputs to their characteristic outputs, then the consequence is a lower degree of this semantic form of informational encapsulation. Call reductions in this form of informational encapsulation semantic ie-consequences.
By SPLIT CONSEQUENCES, we can distinguish two types of cognitive penetration (see fig. 1). Call the tokening of perceptual/cognitive relations resulting in a combination of experiential e-consequences and semantic ie-consequences cases of cognitive shifting. In contrast, call the tokening of perceptual/cognitive relations resulting in a combination of downstream e-consequences and causal ie-consequences cases of cognitive amplification.
|Cognitive Shifting||Cognitive Amplification |
|Epistemic Consequences||Experiential: the epistemic status of experience is altered by a change in the content of experience due to cognitive influence.||Downstream: experience is biased due to cognitive influence but the epistemic status of the experience need not be altered.|
|Informational Encapsulation Consequences||Semantic: perceptual systems can directly or indirectly draw on semantic content from cognitive systems.||Causal: processes in cognitive systems causally covary with processes in perceptual systems.|
Figure 1: Cognitive shifting and cognitive amplification are distinguished by e- and ie consequences.
Most discussions concerning non-attentional cases of cognitive penetration concern cognitive shifting. For instance, take the frequently cited experiments of Levin and Banaji (2006) in which race categorization appears to affect the apparent lightness of black and white faces. Let us suppose for a moment that these experiments are methodologically impeccable demonstrations of cognitive penetration and cannot be explained away in terms of post-perceptual judgments, response bias, shifts in spatial attention, low-level feature interactions, or some other potentially confounding factor. The results of these idealized experiments would show that perceptual experiences are not merely biased so that certain faces are more likely to be noticed or prioritized. Rather, these experiments would show that the epistemic status of subjects’ perceptual experiences is compromised, as reflected by subjects’ distorted lightness judgments in matching tasks. Thus, the resulting e-consequences would be experiential e-consequences.
In their second experiment, Levin and Banaji show that lightness judgments of a racially ambiguous face are distorted due to race category labels. That is, the category label “white” results in subjects overlightening the ambiguous face whereas the category label “black” results in subjects underlightening it. Assuming that the results in this experiment are methodologically impeccable, this suggests that semantic content from cognitive systems concerning the racial category of the face can be drawn on directly or indirectly by perceptual systems as they compute their characteristic input to output functions: a perceptual system darkens an ambiguous face representation when the face falls under the “black” category and lightens it when it falls under the “white” category. There may be other interpretations of these results, but under the assumption that the experiment is methodologically impeccable, this appears to be the most straightforward one. Thus, the resulting ie-consequences are semantic ie-consequences.
In contrast to the idealized Levin and Banaji experiments, most cases of cognitive penetration by attention are cases of cognitive amplification. I consider two counterexamples to this generalization in the next section, but at this point I would like to suggest that the evidence supporting ATTENTIONAL MODULATION is best understood in terms of cognitive amplification. Lupyan and Ward show that category-based attention brings otherwise invisible stimuli into perceptual experience. In these cases, attention plays a biasing role without affecting the epistemic status of perceptual experience. Subjects are more likely to see certain masked stimuli if they attend to those stimuli, but attending to those stimuli does not elevate or depreciate the epistemic status of their experiences. Indeed, these results cohere nicely with biased competition models of attention which understand attentional processes as emerging from distributed mechanisms which bias competition among stimuli representations over limited receptive field space in a range of visual regions (Desimone and Duncan, 1995; Desimone, 1998; Duncan, 1998). Thus, the e-consequences in question appear to be downstream e-consequences.
In section 3, I suggested that an optimization process underlies ATTENTIONAL MODULATION. Such an optimization process entails that certain cognitive states causally covary with certain perceptual states, resulting in causal ie-consequences. However, this optimization process does not appear to result in semantic ie-consequences. Although I do not have a decisive argument for this negative claim, in contrast to the idealized Levin and Banaji experiment, there is no explanatory gain in suggesting that this optimization process requires that perceptual systems be able to directly or indirectly draw on content from cognitive systems. Rather, optimization can be understood as a purely causal process: certain signals are enhanced whereas others are suppressed. Cognition does play a role in determining what is and is not optimized, but there is no evidence of transfer of semantic content from cognition to perception by means of optimization. Content from cognitive systems becomes irrelevant to perceptual processing once attention is appropriately allocated, and perceptual processing does not display sensitivity to how a particular stimulus is cognitively represented in these cases. Although I supported the claims regarding optimization in section 3 with neural evidence, one cannot object that an explanation in terms of optimization simply involves changing levels of description. Signal enhancement and noise reduction, whether described in neural or purely functional terms, do not require that perceptual systems be able to directly or indirectly draw on content from cognitive systems. Thus, it appears that the relevant ie-consequences are merely causal ie-consequences.
5. The Phenomenology of Attention and Ambiguous Images
I conjectured that most cases of cognitive penetration by attention are cases of cognitive amplification. Is there reason to think that some cases of cognitive penetration by attention are cases of cognitive shifting? Let us consider two cases that appear to favor an affirmative answer to this question.
First, consider the following demonstration by Tse (2005). Fixate on any of the dots in fig. 2 and attend to one of the gray disks without moving your eyes.
Figure 2 (from Tse, 2005): Does the attended disk appear darker?
All 16 of Tse’s subjects reported a boost in the apparent darkness of the attended disk, and all subjects reported that they could voluntarily alter the apparent darkness of the disks by shifting their attention from one to another. Tse hypothesizes that voluntary attention alters apparent brightness in this demonstration.
Similarly, experimental work has shown that non-voluntary, bottom-up attention can also boost apparent contrast (Carrasco et al., 2004), size of moving visual stimuli (Anton-Erxleben et al., 2007), color saturation (Fuller and Carrasco, 2006), spatial frequency and gap size (Gobell and Carrasco, 2005), and other appearances. What is important for our purposes is that similar experimental paradigms show that voluntary, top-down attention boosts apparent contrast (Liu et al., 2009) and apparent spatial frequency (Abrams et al., 2010).
Second, reversals in ambiguous images, such as the Necker cube, may be under the control of top-down attention. Top-down attention can increase the amount of time that a desired interpretation is sustained and decrease the amount of time that an undesired interpretation is sustained (Meng and Tong, 2004). This provides evidence that observers can selectively enhance one interpretation while suppressing another. In some cases, these reversals are phenomenologically striking and seem to involve changes in the low-level surface properties attributed to perceived objects (see fig. 3). It is unlikely that all such cases can be explained in terms changes in non-perceptual phenomenology.
Figure 3 (from Krishna, 2016 cited in Lupyan, 2017): Can you see both interpretations?
These two cases seem importantly different from the earlier cases of cognitive amplification. However, this alone does not entail that they are cases of cognitive shifting. Indeed, I think there is reason for rejecting both cases as instances of cognitive shifting: neither results in semantic ie-consequences.
Consider the range of cases where voluntary attention alters apparent contrast and apparent spatial frequency. Block (2010) argues that such phenomenological shifts due to attention cannot be understood in terms of changes in representational content and uses such cases to argue against the thesis that representational content determines perceptual phenomenology. If Block is correct, then these cases do not result in experiential e-consequences. However, setting this matter aside for the time being, let us focus on whether these cases involve semantic ie-consequences. Why think that such cases show that perceptual systems can draw on the content of cognitive systems? In section 4, I suggested that an optimization process underlies ATTENTIONAL MODULATION and that this process entails causal but not semantic ie-consequences. However, it seems that these phenomenological changes in perceptual experience due to attention are the result of the same optimization processes: increases in apparent spatial frequency are the result of enhanced spatial resolution and increases in apparent contrast are due to enhanced contrast sensitivity (Carrasco, 2011). If these optimization processes do not entail semantic ie-consequences, then changes in perceptual phenomenology resulting from these optimization processes also do not entail semantic ie-consequences.
Consider now the range of cases in which attention appears to control reversals in ambiguous images. It is arguable that at least some reversals involve changes in the content of perceptual experience. For instance, the low-level surface properties attributed to the legs in fig. 3 seem to change when a reversal occurs. Again, however, there is no reason to think that such cases entail semantic ie-consequences. Contextually driven top-down effects and adaptation driven bottom-up effects on ambiguous image reversals are additive, pointing towards independent underlying mechanisms (Intaitė et al., 2013). Moreover, ERP evidence indicates that signals correlated with reversals can originate from early visual regions (Kornmeier and Bach, 2005). This suggests that perceptual systems need not rely on semantic content from cognitive systems in order to generate either of the two interpretations of the ambiguous image. This makes it more likely that top down attention’s role in these cases is to merely tip the scales in favor of one of several already entertained interpretations rather than to help generate an interpretation by means of some transfer of semantic content from cognitive systems to perceptual systems. This suggestion fits with the view whereby top down signals selectively gate intrinsic cortical connections in visual regions (Gilbert and Sigman, 2007).
At this point, one might grant that these two cases are not instances of cognitive shifting since they involve causal rather than semantic ie-consequences. However, some of these cases may be instances of a third type of cognitive penetration involving the tokening of perceptual/cognitive relations resulting in a combination of experiential e-consequences and causal ie-consequences. Although I am open to this possibility in principle, more would need to be said in order to secure this claim. Cognitive influence on perceptual experience is a necessary condition for cognitive penetration. But what evidence is there that these cases involve any cognitive influence? The form of category-based, top-down attention in ATTENTIONAL MODULATION is clearly cognitively influenced since it requires the possession of stimulus independent, linguistic representations, such as “chair” and “dog.” However, there does not seem to be any analogous requirement in these two cases, both of which appear to involve top-down spatial attention. It might be suggested that top-down spatial attention requires the possession of perceptually-based demonstratives (see Wu, forthcoming), but it is not clear that this amounts to cognitive influence in any substantive sense. For this reason, I am inclined to altogether dismiss these cases as cases of cognitive penetration, although I admit that these considerations are not decisive.
I conclude with a handful of speculative questions:
- Are there any actual, as opposed to merely nomologically possible, cases of cognitive shifting?
- One issue in answering (1) stems from the difficulty in determining whether a case involves semantic ie-consequences as opposed to merely causal ie-consequences. Is there an experimental paradigm which can help to distinguish these two types of ie-consequences?
- The possibility was raised of there being third type of cognitive penetration. Might one extend a consequentialist approach to show that there are four, five, six, etc. types of cognitive penetration?
- It is sometimes said that perception is “quasi-encapsulated” from cognition. Can the distinction between semantic and causal ie-consequences help to specify what is meant by “quasi-encapsulation”?
Abrams, J., et al. 2010. “Voluntary Attention Increases Perceived Spatial Frequency.” Attention, Perception, & Psychophysics. 72.6: 1510-1521.
Anderson, J. 1978. “Arguments Concerning Representations for Mental Imagery.” Psychological Review. 85.4: 249-277.
Anton-Erxleben, K., et al. 2007. “Attention Changes Perceived Size of Moving Visual Patterns.” Journal of Vision. 7.11: 1-9.
Block, N. 2007. “Consciousness, Accessibility, and the Mesh between Psychology and Neuroscience.” Behavioral and Brain Sciences. 30.5-6: 499-548.
Block, N. 2010. “Attention and Mental Paint.” Philosophical Issues. 20.1: 23-63.
Block, N. 2014. “Seeing-As in the Light of Vision Science.” Philosophy and Phenomenological Research. 89.1: 560-572.
Brogaard, B. and Gatzia, D. 2017. “Is Color Experience Cognitively Penetrable?” Topics in Cognitive Science. 9.1: 193-214.
Burge, T. 2010. Origins of Objectivity. New York, NY: Oxford University Press.
Burnston, D. 2016. “Cognitive Penetration and the Cognition-Perception Interface.” Synthese. 1-24.
Carrasco, M. et al. 2004. “Attention Alters Appearance.” Nature Neuroscience. 7.3: 308-313.
Carrasco, M. 2011. “Visual Attention: The Past 25 Years.” Vision Research. 51.13: 1484-1525.
Carruthers, P. 2006. The Architecture of the Mind. New York, NY: Oxford University Press.
Cohen, M. and Dennett, D. 2011. “Consciousness Cannot Be Separated from Function.” Trends in Cognitive Science. 15.8: 358-364.
Deroy, O. 2013. “Object-Sensitivity Versus Cognitive Penetrability of Perception.” Philosophical Studies. 162.1: 87-107.
Çukur, T. et al. 2013. “Attention during Natural Vision Warps Semantic Representation across the Human Brain.” Nature Neuroscience. 16.6: 763–770.
Delk, J. and Fillenbaum, S. 1965. ‘‘Differences in Perceived Colour as a Function of Characteristic Color.’’ The American Journal of Psychology. 78.2: 290–293.
Desimone, R. and Duncan, J. 1995. “Neural Mechanisms of Selective Visual Attention.” Annual Review of Neuroscience. 18: 193-222.
Desimone, R. 1998. “Visual Attention Mediated by Biased Competition in Extrastriate Visual Cortex.” Phil. Trans. R. Soc. Lond. B. 353: 1245-1255.
Duncan, J. 1998. “Converging Levels of Analysis in the Cognitive Neuroscience of Visual Attention.” Phil. Trans. R. Soc. Lond. B. 353: 1307-1317.
Fodor, J. 1983. The Modularity of Mind. Cambridge, MA: MIT Press.
Fodor, J. 1990. “Information and Representation,” in P. Hanson (ed.), Information, Language, and Cognition. Vancouver, BC: University of British Columbia Press.
Fodor, J. 2001. The Mind Doesn’t Work That Way: The Scope and Limits of Computational Psychology. Cambridge, MA: MIT Press.
Firestone, C. and Scholl, B. 2015. “Can You Experience ‘Top-Down’ Effects on Perception?: The Case of Race Categories and Perceived Lightness.” Psychonomic Bulletin & Review. 22.3: 694-700.
Firestone, C. and Scholl, B. 2016. “Cognition Does Not Affect Perception: Evaluating the Evidence for ‘Top-Down’ Effects.” Behavioral and Brain Sciences. 39: 1-77.
Fuller, S. and Carrasco, M. 2006. “Exogenous Attention and Color Perception: Performance and Appearance of Saturation and Hue.” Vision Research. 46.23: 4032-4047.
Gilbert, C. and Sigman, M. 2007. “Brain States: Top-Down Influences in Sensory Processing.” Neuron. 54.5: 677-696.
Gobell, J. and Carrasco, M. 2005. “Attention Alters the Appearance of Spatial Frequency and Gap Size.” Psychological Science. 16.8: 644-651.
Intaitė, M. et al., 2013. “Interaction of Bottom-Up and Top-Down Processes in the Perception of Ambiguous Figures.” Vision Research. 89.30: 24-31.
Kang, M. et al. 2011. “Semantic Analysis Does Not Occur in the Absence of Awareness Induced by Interocular Suppression.” The Journal of Neuroscience. 31.38: 13535–13545.
Kentridge, R. et al. 1999. “Attention without Awareness in Blindsight.” Proceedings of the Royal Society B: Biological Sciences. 266.1430: 1805-1811.
Kornmeier, J. and Bach, M. 2005. “The Necker Cube—An Ambiguous Figure Disambiguated in Early Visual Processing.” Vision Research. 45.8: 955-960.
Kravitz, D. and Behrmann, M. 2011. “Space-, Object-, and Feature-Based Attention Interact to Organize Visual Scenes.” Attention, Perception, & Psychophysics. 73.8: 2434-2447.
Krishna, R. 2016. “People Are Losing It Over This Photo of Someone’s Legs.” Retrieved on March 9, 2017. <https://www.buzzfeed.com/krishrach/people-are-losing-it-over-this-photo-of-a-girls- legs>.
Levin, D. and Banaji, M. 2006. “Distortions in the Perceived Lightness of Faces: The Role of Race Categories.” Journal of Experimental Psychology: General. 135.4: 501–512.
Ling, S. and Carrasco, M. 2007. “Transient Covert Attention Does Alter Appearance: A Reply to Schneider (2006).” Perception & Psychophysics. 69.6: 1051-1058.
Liu, T. et al. 2009. “Voluntary Attention Enhances Contrast Appearance.” Psychological Science. 20.3: 354-362.
Lombardi, O. et al. 2016. “What is Shannon Information?” Synthese. 193.7: 1983-2012.
Lupyan, G. et al. 2010. “Conceptual Penetration of Visual Processing.” Psychological Science. 21.5: 682-691.
Lupyan, G. and Spivey, M. 2010. “Redundant Spoken Labels Facilitate Perception of Multiple Items.” Attention, Perception, & Psychophysics. 72.8: 2236-2253.
Lupyan, G. and Ward, E. 2013. “Language Can Boost Otherwise Unseen Objects into Visual Awareness.” Proceedings of the National Academy of Sciences. 110.35: 14196–14201.
Lupyan, G. 2017. “Changing What You See by Changing What You Know: The Role of Attention.” Frontiers in Psychology. 8: 1-15.
Macpherson, F. 2012. “Cognitive Penetration of Colour Experience: Rethinking the Issue in Light of an Indirect Mechanism.” Philosophy and Phenomenological Research. 84.1: 24-62.
Maunsell, J. and Treue, S. 2006. “Feature-Based Attention in Visual Cortex.” Trends in Cognitive Science. 29.6: 317-322.
Maunsell, J. 2015. “Neuronal Mechanisms of Visual Attention.” Annual Review of Vision Science. 1: 373–391.
Meng, M. and Tong, F. 2005. “Can Attention Selectively Bias Bistable Perception? Differences Between Binocular Rivalry and Ambiguous Figures.” Journal of Vision. 4.7: 539-551.
Mole, C. 2015. “Attention and Cognitive Penetration,” in A. Raftopoulos and J. Zeimbekis (eds.), Cognitive Penetrability. New York, NY: Oxford University Press.
Nanay, B. 2015. “Cognitive Penetration and the Gallery of Indiscernibles.” Frontiers in Psychology. 5: 1-3.
Noudoost, B. et al. 2010. “Top-Down Control of Visual Attention.” Current Opinion in Neurobiology. 20: 183-190.
Orlandi, N. 2014. The Innocent Eye: Why Vision is Not a Cognitive Process. New York, NY: Oxford University Press.
Prinz, J. 2012. The Conscious Brain. New York, NY: Oxford University Press.
Pylyshyn, Z. 1999. “Is Vision Continuous with Cognition? The Case for Cognitive Impenetrability of Visual Perception.” Behavioral and Brain Sciences. 22: 341-423.
Pylyshyn, Z. 2002. “Mental Imagery: In Search of a Theory.” Behavioral and Brain Sciences. 25: 157-238.
Schneider, K. 2006. “Does Attention Alter Appearance?” Perception & Psychophysics. 68.5: 800-814.
Scholl, B. 2001. “Objects and Attention: The State of the Art.” Cognition. 80.1-2: 1-46.
Shannon, C. 1948. “A Mathematical Theory of Communication.” The Bell System Technical Journal. 27: 379-423.
Siegel, S. 2012. “Cognitive Penetrability and Perceptual Justification.” Nous. 46.2: 201-222.
Stokes, D. 2012. “Perceiving and Desiring: A New Look at the Cognitive Penetrability of Experience.” Philosophical Studies. 158.3: 479-492.
Stokes, D. 2015. “Towards a Consequentialist Understanding of Cognitive Penetration,” in A. Raftopoulos and J. Zeimbekis (eds.), Cognitive Penetrability. New York, NY: Oxford University Press.
Stokes, D. Forthcoming. “Attention and the Cognitive Penetrability of Perception.” Australian Journal of Philosophy.
Teufel, C. and Nanay, B. 2016. “How to (and How Not to) Think About Top-Down Influences on Visual Perception.” Consciousness and Cognition. 47: 17-25.
Tse, P. 2005. “Voluntary Attention Modulates the Brightness of Overlapping Transparent Surfaces.” Vision Research. 45.9: 1095-1098.
Tsuchiya, N. and Koch, C. 2005. “Continuous Flash Suppression Reduces Negative Afterimages.” Nature Neuroscience. 8: 1096-1101.
Watzl, S. 2017. Structuring Mind: The Nature of Attention and How It Shapes Consciousness. New York, NY: Oxford University Press.
Wu, W. 2013. “Visual Spatial Constancy and Modularity: Does Intention Penetrate Vision?” Philosophical Studies. 165.2: 647-669.
Wu, W. Forthcoming. “Shaking up the Mind’s Ground Floor: The Cognitive Penetration of Visual Attention.” Journal of Philosophy.
Yuval-Greenberg, S. and Heeger, D. 2013. “Continuous Flash Suppression Modulates Cortical Activity in Early Visual Cortex.” The Journal of Neuroscience. 33.23: 9635-9643.
 There is also recent philosophical discussion concerning cognitive penetration that is independent of actual instances of cognitive penetration. For instance, Siegel (2012) highlights the challenges that cognitive penetration poses for epistemic views according to which the perceptual experience that p provides an immediate justification for the belief that p. These epistemic views are threatened even if there are only hypothetical cases of cognitive penetration. In addition, the philosophical discussion of cognitive penetration is not limited to philosophy of mind or epistemology. Nanay (2015) invokes considerations relating to cognitive penetration to challenge a classic thought experiment in aesthetics.
 For example, Macpherson (2012) suggests that the Delk and Fillenbaum (1965) study she cites cannot be explained away in terms of post-perceptual judgments, shifts in spatial attention, or several other non-cognitive factors that may influence the experience of hue. Nevertheless, Deroy (2013) maintains that the results can be explained in terms of high-level, multimodal sensory representations. That is, Deroy suggests that some non-cognitive factor overlooked by Macpherson provides a viable alternative explanation. Alternatively, Brogaard and Gatzia (2017) suggest that the experience of determinate hues may itself result from post-perceptual processes. Under this interpretation, Macpherson does not offer an instance of the cognitive penetration since there is no perceptual state that is penetrated.
 My focus in this paper is on visual attention. I am referring to visual attention whenever I use the term ‘attention.’
 Wu (forthcoming) pursues a strategy along these lines.
 The first of these motivations is explicit in Pylyshyn (1999). As for the second, although Pylyshyn does not see attention as analogous to a simple spotlight, he nonetheless states that attention “is an extremely primitive mechanism that provides selection by only the most elementary properties, such as location and a few other primitively transduced properties” (ibid., p. 411). Moreover, he appears to amalgamate attending with cases of selective looking and squinting (ibid.). These two motivations are implicit in much of the cognitive penetration literature as well (see Mole, 2015 for discussion).
 It is worth noting that some recent massive modularity theorists require that modules are informationally encapsulated in a much less robust sense (see, for instance, Carruthers, 2006, ch. 1 on “narrow-scope encapsulation” and “wide-scope encapsulation”). Nevertheless, even with a less stringent requirement, massive modularity theories would be empirically falsified by a thoroughgoing failure of informational encapsulation.
 A signal detection sensitivity measure d’ is calculated using targets hits during stimulus present trials and false alarms during stimulus absent trials.
 In addition, a further series of experiments suggests that a post-perceptual account of category-based attention cannot account for differences in category-based interference in physical similarity judgment tasks due to stimulus onset asynchrony (see Lupyan et al., 2010).
 Similarly, fMRI evidence indicates that masking by continuous flash suppression results in significantly degraded stimulus responses in V1-V3 (Yuval-Greenberg and Heeger, 2013).
 The distinction between cognitive amplification and cognitive shifting bears some similarity to Burnston’s (2016) distinction between the external effect view (EEV) and the internal effect view (IEV), respectively. The independent development of such distinctions suggests that there is something substantive at play. Nonetheless, there are a number of significant differences between Burnston’s view and my own. Burnston takes there to be an ambiguity in the term ‘cognitive penetration’ and distinguishes two different readings. I suggest that there are two different phenomena, one actual (i.e. cognitive amplification) and one, perhaps, merely possible (i.e. cognitive shifting). The importance of this distinction arises as Burnston aims to rule out IEV on relatively a priori grounds. I see two issues with this approach.
First, Burnston’s dismissal of IEV relies on a distinction between the form of cognitive and perceptual representations. Although no one disputes that there is a difference between the content of cognitive and perceptual representations, a distinction on the basis of their form is controversial. For instance, Pylyshyn (2002) argues that there is no compelling evidence to overturn the “null hypothesis” that there is only one representational form, and Anderson (1978) argues that the dispute may amount to an underdetermination problem, making it, in principle, unresolvable. Unfortunately, Burnston provides little motivation for endorsing the distinction in representational form. Indeed, the core of Burnston’s argument against IEV—that there is a failure of informational interaction among the two forms of representation—is equally appealing as a reason in favor of there being only one representational form. For that reason, Burnston’s argument is potentially self-undermining.
Second, I take it to be a substantive empirical question whether there are cases of cognitive shifting. Under my view, aiming to provide evidence in favor of cognitive shifting would be an interesting research project. Burnston’s discussion, however, suggests that cognitive penetration is a homogenous phenomenon. That is, “. . . there is no deeper question to be asked about the relationship between cognition and perception other than determining the ways in which EEV-style relationships are brought about in particular contexts” (ibid., p. 22). From Burnston’s point of view, this leads to a pragmatic gain in clarity. From my point of view, this amounts to a pragmatic harm for future empirical investigations concerning cognitive penetration. I hope to more thoroughly explore the differences between Burnston’s point of view and my own in future work.
 An even weaker notion of information is given by Shannon (1948). Shannon’s notion of “mutual information” serves as a measure of the interdependence of the two systems. An increase in the amount of mutual information between two systems results in less uncertainty of one system given knowledge of the other. For more detailed discussion, see Lombardi et al. (2016).
A system will indirectly draw on semantic content if the semantic content is drawn on by means of an indirect mechanism, such as the indirect mechanism suggested by Macpherson (2012). Macpherson’s indirect mechanism involves two steps. First, some cognitive state generates some non-perceptual state with phenomenal character. Second, this non-perceptual state with phenomenal character alters the phenomenal character of one’s perceptual state by merging with it. The important point is that this indirect mechanism can involve a transfer of semantic content: “the content of such states [the non-perceptual states with phenomenal character] is overwhelmingly content dependent on one’s cognitive states . . .” (ibid., p. 51), and these non-perceptual states with phenomenal character “can affect the phenomenology and content of one’s perceptual experience” (ibid., p. 54).
 Follow up experiments suggest that the experiments of Levin and Banaji are methodologically flawed due to low-level feature interactions and response bias (see Firestone and Scholl, 2015). To my knowledge, it is still an open question whether there are any actual cases of cognitive shifting.
 There are methodological criticisms of the paradigms used in the experiments cited in this paragraph (see Schneider, 2006). For a compelling reply to these concerns, see Ling and Carrasco (2007). Due to issues of length, for the remainder of the paper I assume that attention alters experience in these cases.
 It is worth noting that this third type of cognitive penetration would not, strictly speaking, follow from SPLIT CONSEQUENCES.
Comments on Greyson Abid’s “Two Varieties of Cognitive Penetration”
Dimitria Electra Gatzia
University of Akron, University of Antwerp
I found Grayson’s paper very interesting. I am mostly in agreement with the taxonomy he is proposing. In what follows, I first discuss a couple of potential problems with Grayson’s argument and then make some suggestions that may help strengthen it.
There are two debates pertaining to the issue of cognitive penetration. One is about the possibility of cognitive influences such as concepts or beliefs on perceptual experience. For example, one question within this debate may be whether my belief that I am hearing a high-pitch sound influences my auditory experience of that sound. The other is about the interaction between cognitive and perceptual processes. Specifically, whether cognitive processes influence perceptual processes. This debate focuses primarily on the question of whether the cognitive system influences the perceptual system in such a way as to alter perceptual outputs. Fodor’s notion of “information encapsulation” is typically used to describe the information flow associated with modular systems: when a system A is informationally encapsulated from another B, the information from B does not flow to A.
Within the literature of cognitive penetration, attention has largely been seen as a pre-perceptual process. On this view, attention is not an active controlling influence of perception processing and hence attentional influences on perception do not count as instances of cognitive penetration. Recently, however, philosophers and psychologists have suggested that attention can play a different role. These proposals are primarily concerned with endogenous overt attention that corresponds to our ability monitor stimulus information (typically based on a cue) voluntarily without reorienting the body (thereby, eliminating the worry that the change in the perceptual output is due to changes in the distal stimulus). On this view, attentional influences count as influences of cognitive penetration because attention is an active controlling influence of perception processing.
In “Two Varieties of Cognitive Penetration”, Grayson Abid seems to be concerned with both of these debates. He distinguishes between two types of consequences of cognitive penetration: epistemic (e-consequences) and informational encapsulation (ie-consequences). While e-consequences relate to the first debate about cognitive penetration, ie-consequences relate to the second. On the basis of e- and ie-consequences, Grayson stipulates the following sufficient condition for cognitive penetration (CC):
CC: If a tokening of the perceptual/cognitive relation of type A results in both epistemic (e-) and informational encapsulation (i.e.-) consequences that are not due to movement or changes in proximal stimulus, then the tokening is an instance of cognitive penetration.
This allows Grayson to distinguish between different tokens that may belong to the same type of cognitive penetration. On his view, if two instances of cognitive penetration differ in both e and ie-consequences, then they are instances of distinct types of cognitive penetration (SC):
SC: If perceptual/cognitive relations of type A and B: (a) both count as instances of cognitive penetration by the CC; (b) differ in their e-consequences; and (c) differ in their ie-consequences, then tokenings of A and B are instances of distinct types of cognitive penetration.
Grayson distinguishes between two types of e-consequences: e-consequences may differ if there are epistemically relevant differences in how perceptual experience is influenced by cognition – he calls these consequences “experiential e-consequences”. For example, cognition (e.g., my belief that an object in front of me is red) may influence the epistemic status of perceptual experience by altering the content of perceptual experience (e.g., I experience the object as being more red than it actually is). However, cognition might bias perceptual experience without compromising its epistemic status, e.g., in cases of selective looking of, say, when I try to determine whether an object is present or not – Grayson calls these consequences “downstream e-consequences”.
Grayson distinguishes between two types of ie-consequences. As he rightly notes, how ie-consequences differ depends on the notion of “information” one uses. If “information” is understood in terms of causal covariation, then cases of cognitive information would involve cases in which processes in cognitive and perceptual systems causally covary – he calls these consequences “causal ie-consequences”. But if “information” is understood in terms of semantic content, then cases of cognitive information would involve cases in which perceptual systems directly or indirectly draw on the semantic content of cognitive systems when computing values of functions from inputs to outputs – he calls these consequences “semantic ie-consequences.”
Grayson SC to distinguish between cases that involve what he calls “cognitive shifting” and cases that involve what he calls “cognitive amplification.”
- Cognitive shifting involves the tokenings of perceptual/cognitive relation that result in a combination of experiential e-consequences and semantic ie-consequences.
- Cognitive amplification involves the tokenings of perceptual/cognitive relation that result in a combination of downstream e-consequences and causal ie-consequences.
Grayson’s taxonomy is very helpful in thinking about the possibility of cognitive penetration. However, I would like to point to a potential problem involving Grayson’s sufficient condition for cognitive penetration (CC). Recall that CC states the following:
CC: If a tokening of the perceptual/cognitive relation of type A results in both epistemic and informational encapsulation consequences that are not due to movement or changes in proximal stimulus, then the tokening is an instance of cognitive penetration (emphasis added).
What Grayson seems to have in mind here is that as long as the proximal stimulus remains the same, the tokening of a perceptual/cognitive relation of a type A is an instance of cognitive penetration. The problem with this definition, however, is that the proximal stimulus (i.e., the retinal stimulation) is inherently ambiguous. The same proximal stimulus can be consistent with various distal stimuli (i.e., raw stimuli in the actual world). Ambiguous figures such as the young/old woman, for example, are caused by the same distal stimulus. However, because the same proximal stimulus is consistent with more than one distal stimulus (e.g., the young woman and the old woman), the visual system interprets it either as a young woman or an old woman depending on whether we attend to one or the other. CC would, therefore, be more accurate if “proximal stimulus” was replaced with “distal stimulus.” In other words, CC would be more helpful if it were to read as follows:
CC: If a tokening of the perceptual/cognitive relation of type A results in both epistemic and informational encapsulation consequences that are not due to movement or changes in distal stimulus, then the tokening is an instance of cognitive penetration (emphasis added).
This modification is consistent with studies showing that attention aids in the segmentation of the retinal image by increasing both first- and second-order sensitivity to the attended location. It is also consistent with the current literature on attentional effects. For example, as I mentioned earlier, the type of attention that has been singled out in the literature as purportedly involving cognitive penetration is covert attention. Covert (endogenous) attention involves shifting the direction of attention without reorienting the body. Overt attention, by contrast, involves shifting the direction of attention by reorienting the body. Cases of overt attention are not treated as cases of cognitive penetration because the distal stimulus changes. Some argue that cases of covert attention do not count as cases of cognitive penetration because they involve post-perceptual processes or, as Raftopoulos (2015) puts it attention “rigs up” perception without altering it (although see Lupyan, 2017).  Others argue that they are better construed as instances of perceptual learning (see Gatzia & Brogaard, 2017).
Grayson argues that most discussions of non-attentional cases of cognitive penetration (e.g., cases that involve race categorization) concern cognitive shifting while most cases of cognitive penetration by attention are cases of cognitive amplification. Although Grayson’s focus is attentional cases of cognitive penetration, one may object that his requirement for what counts as cognitive penetration is too strong. Recall, that cognitive shifting involves “the tokenings of perceptual/cognitive relation that result in a combination of experiential e-consequences and semantic ie-consequences.” On this view, for a token to count as an instance of cognitive penetration, it must involve both experiential e-consequences and semantic ie-consequences. In other words, the experience must be altered in a way that its epistemic status is affected and it must be altered in a semantically coherent way. However, a proponent of cognitive penetration may argue that an instance that involves experiential e-consequences need not be accompanied by semantic ie-consequences; what matters in this case is not semantic coherence but causality (see, for example, Stokes (2013). On this view, these cases would count as genuine instances of cognitive penetration. However, on Grayson’s view they would not since they do not involve cognitive shifting.
Attentional Modulation, i.e., (5) below, is best understood in terms of cognitive amplification. His argument for Attentional Modulation is as follows:
- There are non-spatial forms of attention.
- Attention is directed on a cognitive basis.
- Attention affects perceptual processing, not just pre- or post-perceptual processing.
- The changes in perceptual processing in (3) are reflected in perceptual experience.
- Therefore, (Attentional Modulation) non-spatial forms of cognitively directed attention affect perceptual experience by affecting perceptual processing itself.
Grayson discusses studies utilizing masking stimuli through the use of continuous flash suppression (CFS) – it involves presenting a rapidly changing stimulus to one eye renders a static stimulus that is presented to the other eye temporarily rendered invisible (see Lupyan & Ward, 2013). He uses these cases as evidence for (1)-(4), which in turn provide support for (5) Attentional Modulation.
He notes that the results of a study by Lupyan and Ward (2013) indicate that category-based attention brings otherwise invisible stimuli into perceptual experience. However, while subjects are more likely to see certain masked stimuli when they attend to them, attending to them does not affect the epistemic status of their experiences. So the e-consequences here seem to be downstream e-consequences.
Moreover, Grayson argues that the optimization process that underlies Attentional Modulation entails that certain cognitive states causally covary with certain perceptual states. This optimization process, however, results in causal ie-consequences, not in semantic ie-consequences. In this case, while cognition plays a role in determining what is optimized, the content of cognitive systems seems to be irrelevant to perceptual processing once attention is allocated.
Grayson discusses two counterexamples against his claim that most cases of cognitive penetration by attention are cases of cognitive amplification. The first involves a demonstration by Tse (2005). When we fixate on the dot that appears in the space shared by three gray discs arranged in a triangular fashion, the disc we attend to appears to be darker than the rest. This seems to indicate that voluntary attention alters apparent brightness. The second involves reversals in ambiguous images such as the Necker cube. In this case, our experience of the cube changes when we attend to the far or near corners or the cube. Both cases seem to involve changes in perceptual phenomenology.
In the first case, Grayson argues that the phenomenal changes in perceptual experience resulting from attention can be attributed to the optimization that underlies Attentional Modulation. However, he argues that the optimization process that underlies Attentional Modulation entails causal and not semantic ie-consequences. The idea is that the changes in perceptual experience due to attention are the result of the same optimization processes associated with increases in apparent spatial frequency are the result of enhanced spatial resolution and increases in apparent contrast are due to enhanced contrast sensitivity. Assuming that optimization processes do not seem to entail semantic ie-consequences, this case does not involve semantic ie-consequences.
But why assume that optimization processes do not seem to entail semantic ie-consequences? Grayson admits that he does not have a decisive argument for this claim but suggests that optimization could be understood as a purely causal process:
“there is no explanatory gain in suggesting that this optimization process requires that perceptual systems be able to directly or indirectly draw on content from cognitive systems. Rather, optimization can be understood as a purely causal process: certain signals are enhanced whereas others are suppressed. Cognition does play a role in determining what is and is not optimized, but there is no evidence of transfer of semantic content from cognition to perception by means of optimization. Content from cognitive systems becomes irrelevant to perceptual processing once attention is appropriately allocated, and perceptual processing does not display sensitivity to how a particular stimulus is cognitively represented in these cases.”
I would like to suggest that one of the reason for thinking that the optimization can be best understood as a purely causal process is that most of these cases involve perceptual learning (see, e.g., Gatzia and Brogaard, 2017 – footnote 4). Top-down influences associated with the optimization of information processing allow the system to exploit the statistical properties of natural scenes, especially when retinal information is inadequate for the system to interpret the distal stimulus. In other words, as the organism comes into contact with various distal stimuli, it learns to rely more heavily on prior probability distributions of different possible environmental states and adjusts its estimates accordingly. If attentional cases involve the sort of optimization process associated with perceptual learning, then the experiential e-consequences would not be accompanied with semantic ie-consequences. The same can be said about the second case.
In the second case, “contextual driven top-down effects and adaptation drive bottom up effects on ambiguous image reversal are additive…[and] signals correlated with reversals can originate from early visual regions”, which suggests that “the perceptual systems need not rely on semantic content from cognitive systems” to generate either interpretation. The general idea is here is that neither of these cases results in a combination of experiential e-consequences and semantic ie-consequences and as such they are not cases of cognitive shifting.
Contrary to Grayson, Lupyan (2017) argues that some cognitive states influence perception via attention, and these cases count as genuine cases of cognitive penetration because the influence is semantically coherent. According to Lupyan,
“it is not possible to characterize attentional effects as non-semantic changes in input of the kind that occur when we look at one location versus another. Rather, attention can and often does operate over dimensions that we normally think of as reflecting meaning and these attentional effects should be counted as genuine instances of CPP.”
Lupyan’s argument suggests that Grayson’s assumption that optimization can be understood as a purely causal process is very implausible. Lupyan discusses cases of ambiguous figures such as the young/old lady, which involve the same visual inputs being perceived differently due to the perceiver’s knowledge. According to Lupyan, the perceiver in these cases learns to associate certain visual inputs with meaningful categories such as faces. In addition, Lupyan discusses cases involving the ability to cue knowledge using perceptual and linguistic cues such as seeing the ambiguous image as a young lady with linguistic cues such as “young woman facing left”. According to Lupyan, our visual system can evaluate the likelihood that an input corresponds to an object and a category:
“We (our visual system) can evaluate the likelihood that an input corresponds not just to a visual object, but to a category such as a face…These alternatives are preferred to the extent that they offer stronger predictive power, explaining for example, the observed placement of the various visual features. Allowing vision to benefit from these higher-level hypotheses helps make meaning out of noise.”
Although I agree with Lupyan that prior experiences can affect how a stimulus will be interpreted, I do not think that this is a good reason for counting such attentional cases as genuine cases of cognitive penetration. For example, I agree that hearing the description “the chin of the young lady is the nose of the old lady” can lead a subject to attend to a certain segment of the distal stimulus which may then lead the visual system to assign an alternative interpretation to the same ambiguous figure. In this case, while your visual system was initially interpreting the image as a young lady, the verbal description allows it to interpret it as a old lady. However, these cases can be more plausibly explained in terms of perceptual learning than in terms of cognitive penetration. If I am right that this sort of optimization is best understood as involving the effects of perceptual learning, then it is not likely that the consequences are semantic ie-consequences (see e.g., Brogaard and Gatzia, 2017a). That would also explain why we (the perceivers) are often surprised when we realize that our visual system fell prey to an illusion.
This discussion may also be helpful in non-attentional cases, which Grayson takes to be cases of attentional shifting. As I noted earlier, some proponents of cognitive penetration may argue that an instance that involves experiential e-consequences need not be accompanied by semantic ie-consequences since what may matter in this case is not semantic coherence but causality (see, for example, Stokes (2013). So these cases would count as genuine instances of cognitive penetration although they would not count as cases of cognitive shifting. Perhaps, treating these cases to involving perceptual learning may also allow Grayson count these cases as cases of cognitive amplification.
 See Barbot, A., Landy, M. S., and Carrasco, M. (2012). Differential effects of exogenous and endogenous attention on second-order texture contrast sensitivity J. Vis. 12:6. doi: 10.1167/12.8.6
 Raftopoulos, A. (2015b). The cognitive impenetrability of perception and theory-ladenness. J. Gen. Philos. Sci. 46, 87–103. doi: 10.1007/s10838-015-9 288-6
 Lupyan, G. (2017) Changing What You See by Changing What You Know: The Role of Attention. Front. Psychol. 8:553. doi: 10.3389/fpsyg.2017.00553
 Gatzia, D. E. and Brogaard, B. (2017) Pre-cueing, Perceptual Learning and Cognitive Penetration. Front. Psychol. 8:739. doi: 10.3389/fpsyg.2017.00739
 Stokes, D. (2013). Cognitive penetrability of perception, Philosophy Compass 8(7):646-663.
 Brogaard, B. and Gatzia, D. E. (2017). Is Color experience cognitively penetrable, Topics in Cognitive Science 9(1):193-214. doi: 10.1111/tops.12241
 Stokes, D. (2013). Cognitive penetrability of perception, Philosophy Compass 8(7):646-663.
Informational Encapsulation and Epistemic Consequences of Cognitive Penetration are Inextricably Related
University of Cyprus
Abid distinguishes between two kinds of cognitive penetration due to cognitively driven attentional modulation of perception, namely, cognitive amplification and cognitive shifting (henceforth, I discuss cognitively driven attention). In cognitive shifting, attention alters the epistemic status of the experience, that is, the kind of support perception offers to the ensuing perceptual beliefs, by changing the phenomenological content of the experience. There is a semantic relation between cognitive and perceptual processes in that the perceptual systems draw on information pertaining to the representations processed over in cognition.
In cognitive amplification, attention plays a causal role in the formation of the percept affecting the phenomenology of the visual scene, but does not compromise the epistemic status of the experience. Cognitive amplification may have ‘down-stream’ epistemic consequences for perception, but these differ from the epistemic consequences of cognitive penetration usually addressed, because the down-stream consequences are related to the way perception is used by cognition and not to the way cognition affects perception. In cognitive amplification, the cognitive processes causally covary with the perceptual processes that are affected by the cognitive processes but there is not a semantic relation between cognitive and perceptual contents since perceptual processes do not use cognitive information. Abid dismisses cases of cognitive amplification from being cases of genuine cognitive penetration.
With some qualifications, I agree with all of these theses. I offer a more fine-grained analysis of perception and explain why and how this affects Abid’s arguments. I distinguish between two stages of perceptual processing, namely early and late vision, and argue that while cognition affects early vision only in what Abid describes as cognitive amplification, attention affects late vision through cognitive shifting. Early vision is cognitively impenetrable, but late vision is cognitively penetrated. Finally, I extend Abid’s account to show that the epistemic consequences and the informational encapsulation consequences of cognitive effects are inextricably related.
2. Attention affects differently Early Vision from Late Vision
2.1 Early vision
Early vision includes a feed forward sweep (FFS) in which signals are transmitted bottom-up. In visual areas FFS lasts for about 100ms. It also includes local recurrent processing at which lateral and recurrent processes restricted within the visual areas and not involving cognitive signals occur. Recurrent processing starts at 80–100 ms and culminates at 120–150 ms. Studies show that there are early feedback loops, from LGN or V1 to MT/V5 and back, where the recurrent signals engage V1’s neurons to perform different tasks from those performed when V1 received feedforward signals from the LGN.
In Raftopoulos (2009; 2017), I argued that early vision processing is not affected by attention although attention may affect pre-early vision and post-early vision stages. Spatial attention determines where one focuses before the presentation of the stimulus and, thus, before the onset of early vision. In cases of pre-cueing, feature/object based attention prepares the perceptual systems to process some items in the visual scene faster and more effectively by setting up the values of some parameters of the rules governing the state transformations during perception but the processes themselves of early vision are not affected by attention; attention sets up the initial conditions in the transformation equations but the equations themselves are not affected; signal transmission during early vision is not affected by top-down cognitive signals and is restricted within the visual areas of the brain. Thus, the processes of early vision do not use cognitive information as an information resource and this renders early vision cognitively impenetrable (Raftopoulos 2009; Wu, 2013).
I cannot go in the details of the arguments that early vision is cognitively impenetrable and restrict myself to discussing briefly Abid’s suggestion that in cognitive amplification:
The tokening of certain cognitive states results in top down, category-based attention. In turn, category-based attention optimizes the processing of some stimuli over others, resulting in certain perceptual experiences rather than others.
Abid’s analysis is correct if restricted to early vision. Endogenous attention does act externally to early vision, but these forms of attention directly modulate perceptual processing during late vision since late vision processes draw on cognitive information, as we shall shortly see.
I have argued (Raftopoulos 2017) that pre-cueing reflects a change in background neural activity and has anticipatory effects that are established prior to viewing the stimulus. Pre-cueing does not modulate processing during stimulus viewing but they bias the process before it starts; they do not affect perceptual processing on-line. A variety of mechanisms may be available and which one is chosen depends on the task at hand, which means that attention can flexibly solicit different ways to modulate the activity of neurons so as to change visual representations at a cellular level and affect the functional properties of neurons.
Is this cognitive amplification or cognitive shifting? Abid, rightly, thinks that it is cognitive amplification. The processes of early vision are not affected directly by attention and there is no semantic relation to cognitive contents. As I argue in the last section, attention does not affect the epistemic status of early vision either. Thus, attention affects early vision in the manner of cognitive amplification. According to Abid’s analysis, however, if attention affects early vision this way, the contents of early vision should causally covary with the cognitive contents driving attention. In pre-cueing, spatial attention does not determine the percept and in feature/object attention early vision retrieves from the environment all the information in the visual scene. The cognitive information driving attention, therefore, does not covary causally with any of the early visual contents, which means that Abid’s analysis of cognitive amplification does not apply to pre-cueing. It does not even apply to ambiguous figures because, according to Abid’s own account, early vision retrieves both interpretations and, therefore, the product of early vision does not covary with any semantic contents.
2.2 Late vision
The cognitively modulated stage of visual processing is called late vision. Starting at 150–200 ms, signals from higher executive centers including mnemonic circuits intervene and modulate perceptual processing in the visual cortex and this signals the onset of global recurrent processing. In 50ms low spatial frequency (LSF) information reaches IT and in 100 ms high spatial frequency (HSF) information reaches the same area. Within 130 ms post-stimulus, parietal areas in the dorsal system but also areas in the ventral pathway (IT) semantically process LSF information and determine the gist of the scene based on stored knowledge that generates predictions about the most likely interpretation of the input, even in the absence of focal attention.
This information reenters the extrastriate visual areas and modulates at about 150 ms perceptual processing facilitating the analysis of HSF, for example, by specifying certain cues in the image that might facilitate target identification. Thus, shortly after 150 ms specific hypotheses concerning the identity of the object(s) in a scene are formed using HSF information in the visual brain and information from visual working memory (WM). These hypotheses are tested against the detailed iconic information stored in early visual circuits through top-down signals reentering the early visual areas, including V1. Evidence shows that V1 is reentered by cognitive signals mediated by object/feature-centered attention at 235 ms post-stimulus.
The role of N1 ERP component illuminates the way cognitive factors affect directly perceptual processing. N1 arises from multiple generators in posterior parietal areas (140-160 ms) and in ventral occipital-temporal areas (160-200 ms). N1 is enhanced at the attended locations but is not suppressed at the unattended locations. N1 is enhanced only in endogenous attention, that is, is elicited only when subjects view a scene and decide where to attend and do not just when they passively view the scene. N1 is insensitive to the type of the target and occurs before the identification of the target. N1 is considered to be an index of the orientation of spatial attention to objects that are related to the task at hand and are found at the attended locations. Task-relevant objects are distinguished from target irrelevant objects at about 140-200 ms after stimulus onset. Thus, cognition directly affects perceptual processing since the indices of cognitive effects are found in late vision. Since late vision processes operate over cognitive information, the ie-consequences of cognition’s influences on perception are semantic, which means that the attentional modulation of late vision results in cognitive shifting.
3. E- and IE- Consequences are interwoven
Abid distinguishes between two sorts of effects that cognitive modulation through attention may have on perceptual processes. It may affect their epistemic status, or it may affect perception’s standing as an informational encapsulated process. Abid treats these two issues separately. I think that these two kinds are interwined, a view that is fully compatible with Abid’s discussion. In fact, when Abid discusses the epistemic effects on perception that cognitive amplification may have, which he calls downstream consequences, Abid comes very close to the view that I defend next.
In discussing early vision, I claimed that cognition does not affect perceptual processing itself. Thus, early vision processes are not affected by cognitive factors, and the whole processing in early vision is data-driven; the processes of early vision retrieve from the environment all the information that is there, given the limitations of our perceptual system. It follows that cognitive effects do not diminish the sensitivity of early vision to the distal data, that is, the extent to which the information early vision retrieves from the scene is faithful to what there really is there since all data in the visual scene are retrieved and stored into the proximal image. Thus cognition does not affect the information that early vision retrieves from a visual scene and is subsequently used in late vision as evidence for the various hypotheses concerning object identity formed there. In view of the fact that early vision’s sensitivity to the data is not affected, its contribution to the epistemic role of perception is not affected by cognition and, thus, any indirect cognitive effects on early vision do not have any epistemic consequences. According to the consequentialist criterion (Raftopoulos 2009; Stokes 2015), early vision is not cognitively penetrated.
The discussion of late vision shows that late vision uses cognitive information in its processes. When hypotheses concerning the identity of the stimulus are tested, semantic information drives attention to those areas of the proximal stimulus in which pertinent information for the identity of the stimulus is more likely to be found. The conceptual framework of the perceiver influences perceptual processing and affects its sensitivity to the data; according to the consequentialist criterion, late vision is cognitively penetrated.
In discussing the epistemic effects of cognition on early and late vision, the reader has noticed that whether such effects exist or not crucially depends on whether the processes of early and late vision processes draw on cognitive information. If cognition modulates a perceptual stage, it affects the epistemic role of that stage rendering it cognitively penetrated, because cognition involves in the perceptual processing of that stage the conceptual framework of the viewer and this affects the sensitivity of the stage to the data. If the latter obtains, the cognition perception relation is semantic. Thus, Abid’s e- and ie-consequences of cognitive effects on perception are inextricably related. The ie-consequences determine whether there are e-consequences and the e-consequences determine in turn whether the cognitive effects that have ie-consequences are case of cognitive penetration.
Lyons, J. (2011). Circularity, reliability, and the cognitive penetrability of perception. Philosophical issues, 21, The Epistemology of Perception, 289-311.
Raftopoulos, A. (2009). Cognition and perception: How do psychology and neural science inform philosophy? Cambridge, MA: MIT Press.
Raftopoulos, A. (2017). Pre-cueing, the epistemic role of early vision, and the cognitive impenetrability of early vision. Frontiers in Science, Psychology. doi: 10.3389/fpsyg.2017.01156.
Siegel, S. (2013). The epistemic impact of the etiology of experience. Philosophical Studies, 162, 697-722.
Stokes, D. (2015). Towards a consequentialist understanding of cognitive penetration. In J. Zeimbekis & A. Raftopoulos (Eds.), Cognitive Penetrability of Perception: New Philosophical Perspectives (pp. 75-100). Oxford; Oxford University Press.
Wu, W. (2013). Visual spatial constancy and modularity: does intention penetrate vision? Philosophical Studies, 165, 647-669.