In this chapter I have scrutinized the types of cognitive penetration of perception and the loci of the cognitive influence. Both of these aspects are
intrinsically connected to the epistemic consequences that cognitive pene-tration has on perceptual justification. I primarily presented two stages at which cognitive penetration might happen: early and late vision (section 3.1). In section 3.2.3 I also included cognitive penetration of the (visuo-)motor function. Therefore, cognitive penetration of perception is a phe-nomenon which could influence any stage of the visual system: visuo-motor functions, early vision and late vision.
Next, I have succinctly distinguished three types of cognitive penetration of perception: ‘cognitive penetration of the experience’, ‘cognitive penetra-tion of perceptual content or content-related mechanisms’ and ‘cognitive penetration of non-content-related processes’. Cognitive penetration of the perceptual experience implies cognitive influences on perceptual operations responsible for perceptual content which necessarily have an impact on the content of perceptual experiences. This is the type of cognitive penetration held by Macpherson (2012, ming); Siegel (2012, 2013c); Siegel and Silins (ming); Stokes (2013); Stokes and Bergeron (ming). (Section 3.2.1) Sec-ond, cognitive penetration of perceptual content is the sort of influence that modulates information processing necessary for visual representation (per-ceptual content intended for object and scene recognition) and for action (perceptual content required for action), regardless whether this can mod-ify the perceptual experience. This view is maintained byPylyshyn (1980, 1984, 1999a, b, 2003); Raftopoulos (2001c, a, 2009a, 2014a); Raftopoulos and M¨uller (2006b, a) (section 3.2.2). The third form is cognitive pene-tration of non-content-related mechanisms; that is, the modulation of those operations that are not linked to perceptual content, such visuo-motor func-tions or other influences at the visual processing level. This view is shared byLyons (2011) and Wu (2013). (Section 3.2.3)
I also assumed that these forms of cognitive penetration can happen at early or late vision (although I have not provided empirical support yet).
Moreover, I have briefly characterised two forms of cognitive penetration:
beneficial and pernicious. Good cases of cognitive penetration are those that increase the epistemic status or justification of the subject. On the contrary,
bad cases of cognitive penetration decrease justification by undermining or defeating the subject’s epistemic conditions. Cognitive penetration of per-ception represents a serious problem for epistemic theories of perceptual justification, either internalists or externalists. Internalism is mainly influ-enced by cognitive penetration of the experience. Externalism is basically threatened by all forms of cognitive penetration.
Hereafter I will consider that cognitive penetration can influence the modular system — this includes both visuo-motor mechanisms and early vision —, as well as late vision. Cognitive penetration of motor functions depends on top-down influences on non-content-related mechanisms. At the early and late stages, cognitive influences can affect content- and non-content-related mechanisms. In other words, cognitive penetration of early or late vision will involve higher influences either on perceptual content processing or on visual operations independent from this content.
Therefore, we can distinguish three levels at which cognitive penetra-tion occurs: before, during, and after early vision. The distincpenetra-tion between early and late vision allows us to distinguish two forms of cognitive penetra-tion: strong and weak cognitive penetration of perception. Strong cognitive penetration (strong CP) refers to the cognitive influences that affect early vision. Because early vision is defined as an informationally encapsulated module, strong cognitive penetration implies failures in encapsulation of the early visual system. Weak cognitive penetration (weak CP) refers to the cognitive effects on late vision. Due to the fact that the late visual system is non-modular, cognitive influences at this stage of processing represent a weak form of cognitive penetration. Before early vision cognitive penetra-tion influences visuo-motor mechanisms responsible for eye movements and attention. Let’s name this weak cognitive penetration.35
A final aspect discussed in this chapter was the content of perception
35The label ‘strong’ and ‘weak’ depend on the hard core of cognitive penetration: while early vision is modular, late vision is not. Moreover, even if visuo-motor functions are part of the visual module, I include them in the weak CP group. The reason is that, apart from Fodor, even the staunch upholders of cognitive impenetrability of perception consider that visuo-motor mechanisms are penetrated (Pylyshyn (1999a, 343-344;2003, 68-73, 122-123, 157, ch. 4); Raftopoulos(2009a, 277-290); see section2.1.1).
(section 3.4). I spelled out that nonconceptualism appears to be the most intuitive hypothesis for the explanation of perceptual content. Some authors argue that nonconceptualism is incompatible with cognitive penetration of perception, subpersonal early perceptual content is purely nonconceptual if and only if it is cognitively impenetrable; on the contrary, because personal content is cognitively penetrable it can be both conceptual and nonconcep-tual (Raftopoulos 2009a, 2014a; Raftopoulos and M¨uller 2006b, a). Other authors argue that the early visual system can process both conceptual and nonconceptual content while being cognitively impenetrable (Fodor 2007;
Pylyshyn 1984, 1999a, 2003). Finally, a third group holds that even though the visual system is cognitively penetrable its content could remain non-conceptual (Macpherson ming; Toribio 2014). However, I have explained that all these perspectives, in apparent mutual contradiction, are in fact complementary because they take into consideration different definitions of nonconceptual content and cognitive penetration.
In the next chapter I will examine penetrating and penetrated aspects and how the phenomenon of cognitive penetration can be defined.
4
Characterisation of Cognitive Penetration
Contents
4.1 Penetrating Aspects . . . 129 4.2 Penetrated Aspects: Architecture and Content . . . 136 4.3 Consequences of Cognitive Penetration . . . 148 4.4 Defining Cognitive Penetrability of Perception . . . 155 4.4.1 Semantic Criterion . . . 156 4.4.1.1 Objections to the Semantic Criterion 159 4.4.2 Counterfactual Criterion . . . 161
4.4.2.1 Objections to the Counterfactual Cri-terion . . . 162 4.4.3 Causal Relation Criterion . . . 167
4.4.3.1 Objections to the Causal Relation Criterion . . . 167
4.4.4 Informational Encapsulation Criterion . . . 170 4.4.4.1 Objections to the Informational
En-capsulation Criterion . . . 172 4.4.5 The Informational Encapsulation Criterion
(Re-formulated) . . . 173
The present chapter is composed of two parts. In the first part I would like to scrutinize further aspects of cognitive penetration, to wit, the po-tential penetrators, the penetrated dimension of the system, and the con-sequences on the perceptual system (see Siegel (2012, 204) for a similar enumeration). First, I argue that cognitive, emotional or affective mental states at the origin of the penetration are not the penetrators of the visual system but rather the triggers of the penetrating subpersonal states which eventually modulate perceptual processes (section 4.1). Second, I show that the penetrated aspects in visual perception are the architecture or the content of the visual system (section 4.2). Hence, there is architectural cognitive penetration when the higher influence modifies how the system processes (selects, retrieves, transforms, substitutes, stores, etc.) percep-tual content; andcontent cognitive penetration when top-down states affect the perceptual content (adds, modifies, or substitutes any piece of informa-tion). Furthermore, I explain that these two forms of influence can occur synchronically, when the cognitive effect is simultaneous to the perceptual act, or diachronically, when it occurs before visual perception. And finally (section4.3), I analyse the consequences that cognitive penetration can have on the visual system.
The second part of the chapter focuses on the definition of cognitive penetration (section 4.4). Even though philosophers and psychologists fre-quently use the term ‘cognitive penetration’, what they mean by this term is sometimes very different. The common core of the concept could be
for-mulated as follows: cognitive penetration refers to the kind of cognitive influences on perceptual systems which have consequences on the epistemic status of the subject. Then, I will scrutinize different definitions of cog-nitive penetration — to wit, Pylyshyn’s semantic criterion (section 4.4.1), Siegel’s and Macpherson’s counterfactual criterion (section 4.4.2), Stokes’
causal relation definition (section 4.4.3), and the informational encapsula-tion criterion by Wu (secencapsula-tion 4.4.4). I will finish the chapter by proposing a reformulation of the informational encapsulation criterion which will be capable of bringing cognitive influences on content- and non-content-related processes at the (visuo-)motor, early and late visual levels under the label of cognitive penetration of perception (section 4.4.5).
4.1 Penetrating Aspects
In this section I will argue that while cognitive mental states (e.g., be-liefs, desires, emotions) are at the origin of the cognitive penetration, they should be understood as triggers of the penetrating cognitive mechanisms rather than as penetrating states. While the triggers are personal states, the penetrators remain at the subpersonal level.
Cognitive penetration is frequently formulated as a case in which the subject’s affective or cognitive mental states penetrate perceptual systems (e.g., Lyons 2011;Macpherson 2012;Siegel 2012,2013c;Stokes 2012,2013).
Cognitive mental states can be beliefs (creedal feelings and dispositions, be-liefs and assents, and judgements and acceptances)1, knowledge, concepts, hypotheses or suppositions, presumptions, doubts, suspicions, intentions, expectations, goals, aims, needs, priming, context-sensitive conventions, hopes, wishes, desires. But also personality traits, attitudes, and affective andemotional states such as moods, emotions, feelings, fears, aversions, and the like. (See Engel (1988a, 11; 2012b, 20); Fodor (1983, 68, 76); Gilbert
1This enumeration is possible if we consider that there is a hierarchy of continuous degrees of belief, “starting at a low level with creedal feelings and dispositional beliefs, moving upwards to degrees of belief and to assents, with judgments and acceptances at the top level” (Engel 2012b, 22).
and Sigman (2007, 679); Bar (2009b, 1239); Macpherson (2012, 27, fn. 3);
Siegel (2012, 201; 2013c, 698, 719-720); Siegel and Silins (ming, 30); Lyons (2011, 290); Pylyshyn (1980, 119; 2003, 65); Stokes (2012), for these lists.) Other states that must be added to the list are memories, dispositions, and wishful thinking.
Cognitive mental states are considered as propositional attitudes with conceptual content. Beliefs and desires possess representational content which corresponds to a state of affairs. The subject’s beliefs aim at fitting the world (beliefs aim at truth), they describe the world as it is. On the other hand, desires describe the world as it should be (desires aim at their own satisfaction); the world should fit the subject’s desire. That is, the
“direction of fit” (Searle 1983) is different in this attitudes: while the belief’s direction of fit is mind-to-world, the desire’s direction of fit is world-to-mind.
(See Dokic and Engel (2002, 55) for this explanation.) The truth of these states depends on its relation to the states of affairs and having the states suppose to have concepts. Cognitive states share these properties.
Emotional states (emotions and feelings) are more difficult to define than beliefs. Their characterisation depends on the theory adopted, either psy-chological and evolutionary, or cognitivist, or perceptual (de Sousa 2014).
I do not intend to discuss these issues here but briefly focus my attention on the content of these states from a perceptualist view point. Emotions and feelings appear to be different from cognitive states. While the lat-ter are propositional and conceptual, the former are not. The content of emotional states, as it appears to be the case for perception, appear to be non-propositional (de Sousa 2014): they are not susceptible of being true or false. Second, their content does not seem to be conceptual; a subject can have an emotional state even though she does not possess any concepts to characterise its content (Gunther 2003, 287-288; Peacocke 2001a; Tappolet 2003).
Something similar happens with memory. Some types of memories are regarded as mental states which share similar characteristics with percep-tion: they can be non-propositional and non-conceptual (Bernecker 2008,
35-36; 2010, 26; Martin 1992, 755). For instance, Bernecker (2008, 35-36) distinguishes between “fact memory” (memory of previous facts) which presupposes the possession of concepts, and “object-, property-, and event-memory” which, like experiences, are non-conceptual (Bernecker 2010, 11-19). He considers that while object-, property-, and event-memory are subdoxastic states (e.g., implicit, tacit, or unconscious knowledge or sub-personal states), fact-memories are dispositional beliefs. (Bernecker 2008, 35-36) This difference is close to the distinction between a form of propo-sitional memory (e.g., the red chair is in the room) and a pictorial form of memory (e.g., yesterday’s sunset); the former share the characteristics of beliefs and the latter those of the experience.
I believe that emotional and affective states should be distinguished from cognitive states. So we should talk about ‘emotional or affective pen-etrability’ and ‘cognitive penpen-etrability’ respectively. First, it is doubtful that emotions are cognitive states, for they could be nonpropositional and nonconceptual. Second, emotions and other cognitive states are the result of different and well-identified neuroanatomico-functional systems. How-ever, for the sake of simplicity, and following the normal usage (Lyons 2011;
Siegel 2012; Macpherson 2012; Pylyshyn 1999a; Stokes 2013), I will use the label ‘cognitive’ for both cognitive and emotional penetration regard-less of whether they are propositional or non-propositional, conceptual or non-conceptual.2
Let me now suggest how cognitive states might penetrate perceptual systems. In order to do so, I would like to distinguish two types of cognitive levels: personal and subpersonal.3 From a functional viewpoint cognitive states involve two stages: a subpersonal and a personal (Engel 1996, ch.
2One of the main reasons to use the same label for both kinds of penetration is the extreme difficulty to determine when the penetrating state is a belief, an emotion or a feeling. When cognitive penetration occurs, both cognitive and emotional brain networks interact and influence together perceptual systems. I consider that my explanation of cognitive penetration is perfectly plausible for penetrating conceptual or nonconceptual states as well as for propositional or nonpropositional states.
3For the sake of simplicity I will focus on cognitive states and leave aside emotions and feelings.
III).4 The subpersonal computational state of a belief involves the part of the process which is by definition unconscious to the subject; to wit, brain processes that make possible conceptual or cognitive content processing.
The personal state is the conceptual content the subject can have conscious access to.5 In other words, cognitive states involve functional states respon-sible for the process of cognitive content at the subpersonal (this comprises the inner workings of cerebral processes) and at the personal (the conscious outputs of cognitive states (Dehaene and Naccache 2001, 16)) levels.
To the subpersonal category also belong internal cognitive processes that do not directly aim at the production of personal states. Those are functions responsible for selecting, retrieving, transforming, controlling, and storing information; choosing, predicting, and planning prospective actions; and so
4I take this distinction to be similar to the characterisation of doxastic and subdoxastic (or subrational) states (Bernecker 2008, 36;Engel 1988a, 15;1996, 241-242;Stich 1978).
According to Stich, subdoxastic states are isolated: their information cannot be used for reasoning, speech, and other cognitive functions because we have no conscious access to them; “[s]ubdoxastic states occur in a variety of separate, special purpose cognitive subsystems”. On the contrary, doxastic states “form a consciously accessible, inferentially integrated cognitive subsystem”Stich(1978, 508).
5We are never conscious of the inner workings of cerebral processes responsible for cognitive functions but solely of their conscious outputs: beliefs, desires, judgements (Dehaene and Naccache 2001, 16). Prinz (2006, 30) claims that “Nisbett and Wilson (1977) have shown that human judgment is often driven by processes that operate below the level of consciousness”. When we infer a conclusion from some premisses we are only conscious of the propositions implicated in the procedure but we have no access to the rules the brain carries out to solve the inference. Speakers are conscious of the sentences of a language but they only have tacit knowledge of grammar and the rules of syntax.
(Berm´udez and Cahen 2012;Dehaene and Naccache 2001, 16;Engel 1996, 226-231;Prinz 2006, 30)
There are other processes which might not always reach consciousness. Blindsight subjects, who claim that they cannot see the stimulus in their blind hemifield, are able to reach and grasp the object at this location or even identify it (Weiskrantz (2003);
alsoLyons(2009, 57);Plantinga(1993, 93)). Similar cases are noticed in visual agnosias, prosopagnosia, neglect, aphasia, alexia, and other brain damages (Dehaene and Naccache 2001, 5). This kind of unconscious (or potentially conscious) visual processes is also seen in normal subjects: masked priming (Dehaene and Naccache 2001, 6); the blindspot — we are not (directly) conscious of the fact that visual information is not captured by this retinal region (Dehaene and Naccache 2001, 17-18) —; subliminal priming at visual, semantic and motor levels (Dehaene and Changeux 2011, 202-207), and so on.
The fact that an observer may be unconscious of some stimuli does not mean that such information has not been processed by the system. On the contrary, unconscious processes affect subsequent cognitive and motor operations (Baars and Franklin 2003, 167;Dehaene and Naccache 2001, 5-6;Pylyshyn 2003, 165, 365).
on. Those are processes carried out by executive functions (Fuster (2008, 178-192); Miller and Wallis (2009), and section 2.1.3).
Let’s apply the personal/subpersonal distinction to cognitive penetra-tion. It is frequently claimed that personal cognitive states penetrate the visual system. However, I prefer to understand them as triggers of the sub-personal cognitive mechanisms that will ultimately penetrate perception.
(See Siegel(2013c, 716, 720) andMacpherson (2012) who suggest that cog-nitive mental states are triggers.) Thus, cogcog-nitive penetration of perception seems to occur because personal states trigger subpersonal cognitive mech-anisms that eventually penetrate the visual system.
An illustration of how cognitive states influence perceptual processes is given by Michael Tye. He explains how the subject’s conceptual equipment modifies the way the visual scene or ambiguous figures will be decomposed.
He writes:
Where a figure has an ambiguous decomposition into spatial parts, concepts can influence which decomposition occurs. This is one way in which top-down processing can make a phenomenal difference.
But once a particular decomposition is in place, the way in which an ambiguous figure phenomenally appears is fixed. [...] These seg-mentations are no doubt conceptually influenced, but the resulting representations of spatial parts are not themselves automatically con-ceptual. [...] [T]he concepts do not enter into the content of the sensory representation, and they are not themselves phenomenally relevant. (Tye 1995, 140)
I accept that conceiving of a visual scene or an ambiguous figure in one way rather than another may sometimes influence how we break it up cognitively into spatial parts, for example, and the shapes we then experience may not be the ones we would have experienced under a different conceptualization. Even so, the sensory experiences of shapes (at the most basic level) do not require shape concepts.
Seeing a cloud, for example, I will likely have an experience of a shape for which I have no corresponding concept. (Tye 2000, 61)
Tye claims that the visual system can disambiguate ambiguous figures (e.g., duck/rabbit) if the concept of one of the figures (e.g., rabbit) primes visual perception. I believe that the decomposition of the figure by the conceptual influence might happen in two ways. One form involves cogni-tive penetration of non-content-related visual processes: the cognicogni-tive state including the concept ‘rabbit’ disambiguates the visual representation by guiding attention to a specific part of the figure. This might happens as follows: the (personal) conceptual state triggers (subpersonal) cognitive functions or states which at some point affect non-content-related visual mechanisms (the motor system). Eventually, the eyes are oriented to a spe-cific location and the visual system selects rabbit’s features from the scene.6 The other form of cognitive penetration that may be involved in the de-composition of the ambiguous figure is cognitive penetration of perceptual content or content-related mechanisms. While keeping the attentional focus fixed, the (personal) cognitive state could sensitize visual neural networks responsible for the representation of rabbit’s features. A similar example seems to occur when our knowledge of objects’ colour (e.g., bananas) affects how we see these objects (e.g., more yellow than they are) (Hansen et al.
2006; Olkkonen et al. 2008). The cognitive penetration process in these two latter cases seems to happen as follows: the cognitive penetrating state (e.g., the concept of rabbit or the belief that bananas are yellow) triggers some (subpersonal) cognitive mechanisms that influence visual structures in control of perceptual content processing. The consequence of this process is that we see rabbit’s features or a strong hue of yellow.7 In sum, cognitive penetration occurs because (personal) cognitive states trigger the subper-sonal cognitive mechanisms which eventually affect perceptual processes.
6From an empirical point of view, subpersonal cognitive functions might influence cortical areas such as the frontal eye field (FEF) and the dorsolateral prefrontal cortex (DLPFC) as well as subcortical structures such as the superior colliculus (SC) both responsible for allocation of visual attention and eye movements. See section2.1.1.
7Cognitive influences may increase visual networks’ sensitivity to long elliptic vertical shapes (rabbit’s ears), curvilinear triangular shapes (head) (e.g., visual areas V1 and V2), in the first case, and the assembly of neurons responsible for colour detection (e.g.,
7Cognitive influences may increase visual networks’ sensitivity to long elliptic vertical shapes (rabbit’s ears), curvilinear triangular shapes (head) (e.g., visual areas V1 and V2), in the first case, and the assembly of neurons responsible for colour detection (e.g.,