The aim of this section is threefold: 1) show difficulties in disentangling feelings from emotions; 2) attempt to give a more precise characterisation of the formal object of feelings, along with some examples; 3) criticize the use of the term “direct” and offer another definition that is consistent with predictive coding. The first problematic point that I see is Proust’s use of the term “affective”, which is ambiguous. She employs at least two different definitions of “affective”:
Feelings that possess valence (p. 20). Yet all kinds of feelings, according to Proust, possess affect and valence[1] (p. 1). Given her distinction between “hot” (emotional) feelings and those that have valence[2] (p. 21), emotional feelings might differ from mere feelings with valence due to the differently-experienced valence, maybe if emotional valence were a richer experience. Thus, the question is about the minimal requirements on valence and intensity in feelings.
Feelings that express emotions.
Difference between feelings and emotions: if agentive and metacognitive feelings can be affective, then the categorization of feelings into bodily, agentive, metacognitive, and affective (p. 5) might be better restricted to the first three, with the fourth being a dimension along which they vary. If affective in this categorization means emotional (p. 2), then there is an ambiguity of terms—affective = having valence and affective = being part of an emotional experience—because the latter seems to be more complex.
Difference between formal objects of feelings and emotions: if “feelings are affective ingredients in emotional awareness” (p. 3), then there is a circularity in understanding affectivity here: feelings are affective in virtue of being part of an emotion, while at the same time they themselves are the affective component in the emotion of which they are part. The first part of this claim can be followed from that defended by Proust, namely that feelings that do not express emotions are not affective (p. 2). The second part of the claim follows from Proust’s claim that feelings are affective ingredients of emotions (p. 3). As elaborated in the previous section, emotions are said by Proust to contain one of two kinds of subjective appraisals: feelings or appraisals cum conative dispositions. Further, if feelings are components of emotions, but both can have a formal object, then those objects might diverge. The consequence is that an emotion and a feeling that is part of it might be directed at different objects. Thus, Proust on the one hand distinguishes feelings from emotions and yet on the other hand claims that not only emotional feelings, but also agentive and metacognitive feelings might be “feeling toward” experiences (p. 3, pp. 20–21). The latter claim that both feelings and emotions are directed at intentional objects has been used as an argument to identify both (see de Sousa 2014 section 2 for a discussion of this question). Given Proust’s claim that there are somatic, affective, agentive, and metacognitive feelings, and given the claim that at least in metacognitive feelings the formal object is not the cognitive disposition itself but the rate of change of its execution above or below discrepancy, an interesting question focuses on the formal object of emotional feelings.[3] For example, can it be that while the formal object of the emotion of fear is some dangerous object, the object of a feeling is a rate of change in the assessment of the situation before and after the change of the formal object of an emotion? This might explain why, e.g., the first bite of a bar of chocolate makes one happier than the following bites.
Bodily phenomenology of feelings as their formal object: Proust argues that while somatic feelings are about bodily sensations (or, more consistently, about the rate of their change), in affective (emotional) and possibly metacognitive feelings “the bodily phenomenology tends to recede to the fringe of consciousness” (this collection, p. 2). The example that Proust gives with respect to metacognitive feelings is that feelings of remembering are correlated with but not about facial muscle activity (p. 3). Proust acknowledges that there might be mixed cases (experience of bodily feeling + intentional content, pp. 2–3), but I want to argue that in some emotional feelings bodily phenomenology is, to borrow a metaphor, in the foreground. There might be emotional feelings whose objects are bodily sensations, e.g., the anxiety that arises during a panic attack: when I concentrate on my accelerated heartbeat, then if I come to associate the heartbeat with some threatening aspects of a situation, such an experience might lead to anxiety, and thus the initial anxiety leads to even more anxiety, leading to a vicious cycle of panic (for a discussion of heartbeat perception in panic disorder see Ehlers & Breuer 1996). This might be a case of an emotion whose formal object is the rate of change of bodily sensations, or maybe a meta-feeling (for a discussion of meta-emotions see Mendonça 2013).
In the given panic example it might have seemed as if I had embraced the analogy between feelings and perception that Proust denies, so I will explain why it may be more beneficial to use the term “direct” in another sense to that used by Proust. Proust makes a sharp distinction between feeling and perceiving: “[w]hile percepts allow recognition and identification of external objects and properties, feelings express specific affordances in a perceived, imagined, or remembered situation” (this collection, p. 10). Non-conceptual parts of perceptions are said to relate to “objective, external contrastive cues” (Proust this collection, p. 10), while in feelings they relate to evaluative states. Perception is said to involve “direct sensory access to the world” (p. 10), while the access of feelings to the world and the body is claimed to be indirect. Proust’s evidence for a disanalogy between feeling and perception is based on the neuroscientific research of Barrett & Bar, who say that absence of “internal affective context” impairs the categorization of objects (2009, p. 12).[4] Their evidence for this hypothesis is based on reviewing the anatomic connections involved in affective processing and that of object perception. One critique of this might be that the time of activation of certain regions responsible for emotional processing and perception might justify the claim that emotional processing comes before perception, but not how direct such processing is. Moreover, in light of predictive coding, perception, emotion, and cognition might all be indirect (Hohwy 2014; for more technical elaboration Friston et al. 2014). In other words, predictive coding provides the term “direct” with a meaning other than that used by Proust. In predictive coding directness is an absence of the evidentiary boundary, where the evidentiary boundary is the inferential isolation between the model of the world and the hidden causes of sensory input (Hohwy 2014). This means that causes beyond the boundary have to be inferred on the basis of independent evidence (ibid., p. 6), or, in Hohwy’s words, “[t]he brain doing the inference is secluded at least in the sense that certain kinds of doubt about the occurrence of the evidence are unanswerable without further, independent evidence” (p. 7). Relating this observation to Proust, on the premise of accepting predictive coding, there might not be a sharp distinction between feeling and perceiving such as Proust postulates, or at least not in the form presented in the target article. If interoception as perception involves inferences about circumstances beyond the (same) evidentiary boundary, as suggested by Hohwy (2014), then feeling and perceiving would both be indirect (to the same degree).[5] If interoception does not go beyond the evidentiary boundary, feelings might be direct, even if perception is not.
In this part of the review I will point out the dangers of interpreting the relation between feelings and concepts too simplistically and argue that it is possible that at least some kinds of feelings are influenced by concepts, even if they themselves are non-conceptual. Proust argues that for metacognitive feelings to arise an important affordance, as well as an implicit heuristic, has to be present (this collection, p. 18). This heuristic is based on cues about the dynamics of information processing, but not its contents (p. 15). The dichotomies that Proust uses in the description—implicit–explicit, unconscious–conscious, evolutionarily-old–evolutionarily-new, associative–rule-based (pp. 3-4, p. 17)—have often been mapped onto two different kinds of processes in dual processing theory (e.g., Frankish & Evans 2009). Dual processing theory states that there are two kinds of processing that possess the dichotomous characteristics mentioned above. A minimal description provided by Evans (2009) for type 1 is “fast, automatic, high processing capacity, low effort”, and for type 2 “slow, controlled, limited capacity, high effort” (p. 33). Along these lines, “implicit”, “unconscious”, “evolutionarily old”, “associative” have been also used as descriptors for type 1 and “explicit”, “conscious”, “evolutionarily new”, “rule-based” as descriptors for type 2. A belief bias (accepting more believable than unbelievable conclusions) might serve as an example for type 1 (ibid., p. 41), and the conscious correction thereof for type 2. The worry I have is adding to those dichotomies another one: non-conceptual (meaning in this case non-propositional; Proust this collection, p. 7)–conceptual (propositional, belief-like). Proust holds that “cues (associative heuristics) dictate how an affordance is detected, assessed and exploited in a context, but these cues are not consciously available, and hence do not depend on a naïve theory of the task” (p. 17). This inference is not valid in the given form. I agree with Proust that “[a] cue-based, non-analytic heuristic is not inferential in the interpretive, first-person sense” (p. 17), but I hold that there is at least one step to consider in between non-conceptual[6] affordances and consciously evaluated affordances. And this is automatic concept-based activation (the existence of automatic appraisal is acknowledged by Proust; footnote 7).
Evans (2009) distinguishes between different kinds of dual processing theories, among which are the sequential (first automatic processing, then controlled) and the parallel theory. Proust seems to embrace a sequential kind of dual processing theory, given the functional role she ascribes to metacognitive feelings (evaluation of mental actions before and after their execution; Proust 2013). Yet how far implicit heuristics are independent of concepts is in question. Proust (this collection) denies that “a concept-based interpretation will affect the experienced feeling itself” (p. 17). As mentioned in section 1, she also denies that feelings have a conceptual format. Thus, she seems to deny both that concepts play a causal role in the emergence of feelings and that feelings themselves possess a conceptual format. I will briefly demonstrate that the term “implicit heuristic” does not preclude automatic concept activation, if it implies the activation of knowledge or goal representations. Thompson (2009) argues that heuristic processes are contaminated by background knowledge, as well as by beliefs and expectations (p. 172, p. 174). Frankish (2009) notes that “the concepts of belief and desire correspond to the psychologist’s concepts of knowledge (or memory) and goal structure” (p. 91). Hence, activation of knowledge that may provide the context for feelings could also be conceptual. Goal representations might also be activated in the course of context creation, provided that unconscious goal pursuit is flexible and context-sensitive (Aarts & Custers 2012). Further, unconsciously activated goals not only depend on context, but also create context by influencing the accessibility of knowledge, evaluations, and emotions (Fishbach & Ferguson 2007, p. 496). It follows that if goal representations are activated, then they might lead to the activation of conceptual knowledge. Another interesting point is that if there is a continuous interplay between goal representations and affordances (opportunities in the environment; Huang & Bargh 2014, p. 125) and if goal representations can change the experience of the world (ibid., p. 124), then goal representations might change sensing of affordances and, hence, the feelings associated with it. Further, there has been a proposal to distinguish between associative and rule-based processes by the kind of architecture they operate upon: namely connectionist vs. classical computational (for a short discussion see Samuels 2009, pp. 141–142). Thus, implicit heuristics might be understood as certain connected representations in a network being activated by some cues, where the question is about the representational format of such knowledge, or a more precise description of the relational nature of the feeling affordance. Last, a general note about the similarity between feelings and other kinds of representations: if Bliss-Moreau & Williams (2014) are correct in defending the claim that all kinds of representations possess an affective component (valence + arousal in their definition), then affect is something that expressive and conceptual representations share.
Of course, Proust’s claim that in the case of feelings those cues relate to the dynamics, but not to the contents of processes, indicates a more specific understanding of the kind of implicit heuristic in question. My point, though, is that if humans can “enrich their noetic feelings through concepts, and thereby revise their reliance on fluency where it is not justified” (Proust 2013, p. 144), then in humans implicit heuristics may also be influenced by concepts (in an automatic way) and in such a way influence feelings. Needless to say, the independent existence of such a schema (be it cognitive or emotional) is hard to prove (Eysenck & Keane 2010, p. 597). According to Koriat & Levy-Sadot (1999), as cited by Proust (this collection, p. 15), metacognitive feelings arise as a result of nonanalytic inferential processes (described as the implicit or unconscious application of heuristics), in distinction to the direct memory trace hypothesis, according to which feelings have direct access to memory traces (Koriat & Levy-Sadot 1999, p. 487). Koriat & Levy-Sadot (1999) argue that the presence of dissociations between knowing and the feeling of knowing speaks against the second hypothesis. Even if heuristics in feelings are non-conceptual, the fact that through feelings emotion gets its valence necessitates that we consider how concepts and memory traces influence feelings, given that they play a role in emotions. Lane et al. (forthcomming), for example, argue that psychotherapeutic change is made possible by updating prior emotional experiences, for which memory traces of those experiences have to be reactivated and reconsolidated. Thus, even if feelings are non-propositional (Proust this collection, p. 20), activation of concepts and their expression in propositional terms are to be distinguished. The point is not that metacognitive feelings themselves cannot have indexical formats,[7] or that an agent could not possess expressive and conceptual representations at the same time, but that in humans the generation of (at least) metacognitive and emotional feelings might be preceded by an automatic concept activation that influences them. If this is the case, then one could ask again whether feelings are transparent (see section 1).
Further, instead of describing cognitive processes as serial, their dynamic (continuous) nature might be more worthy of emphasis. In the target article, Proust mentions that “[i]ncreased activity in the smile muscle, the zygomaticus major, produces feelings with a positive valence” (this collection, p. 15). This suggests that facial expression influences emotions. She also argues for the transparency (impenetrable nature) of feelings and the against two-factor theory, thus against the possibility that appraisal influences the valence of feelings (see section 1). I want to offer for clarification purposes a short review of the recent literature on which factors are supposed to influence feelings and factors feelings influence themselves Rogers et al. (2014) emphasize the dynamic nature of emotions insofar as they depend on the social appraisal of a situation. Brosch (2013) also emphasizes the dynamic nature of appraisal that plays a causal role in eliciting emotions. The definition of appraisal that Brosch (2013) accepts also encompasses low-level appraisal based on learned schemata (p. 370). Brosch (2013) argues that first an initial low-level appraisal affects the physiology (1), action tendency (2), expression (3), and feeling (4) of an emotional experience, and then those changes in turn affect an on-going (low- and high-level) appraisal, establishing an appraisal loop. Here, the direction of influence is still in question, e.g., whether feelings influence expressions or the other way around. Laird & Lacasse (2014) defend the James–Lange theory of emotion, namely that facial expressions (e.g., BOTOX patients being less responsive to mild positive emotional stimuli; for the reference see ibid., p. 29), expressive behaviour (e.g., romantic attraction as a result of shared, mutual gaze; ibid., p. 29), and visceral responses that are interpreted according to situational cues (e.g., misattribution of emotion) are causes of emotions (for a critique of their evidence see Reisenzein & Stephan 2014). As such, they may influence emotional feelings too, which Proust acknowledges by pointing out the causal connection between measures in facial muscles and affective character of feelings (this collection, p. 25). Yet the direction of influence may also go the other way around (from feelings to facial expressions). Thus, the nature of feelings may also be dynamic, as are the nature of the underlying cognitive processes. Interestingly, Thagard & Schröder (2014) argue for a neurocomputational theory of emotions as semantic pointers (term introduced by Chris Eliasmith). They argue that physiological, appraisal, social, and psychological components of emotions can be integrated into one unified account: emotion tokens can possess both shallow and deep meanings. The compressed (shallow) form of emotions is reportable, while at the same time pointing to the uncompressed deep form that binds together situational, physiological, and appraisal components.
In the preceding paragraph I considered literature supporting the claim that feelings are embedded in continuous cognitive processes. The purpose of this was to show that how appraisal might influence feelings in some form is complex and might even be circular. In this paragraph I offer some additional evidence against a discontinuous interpretation of the connection between feelings and propositional descriptions thereof. The existence of affective blindsight (ability to discern emotional stimuli despite inability to consciously perceive them; Eysenck & Keane 2010, p. 581) would stand in line with the assumption that emotional and cognitive processing is based on different kinds of information. This is because affective blindsight demonstrates the dissociation between two different kinds of processing and, thus, a dissociation between the information needed for the one kind and for the other. Further, Scott et al.’s (2014) experiment demonstrating blind insight (accurate metacognitive accuracy in the absence of discriminative accuracy) on the one hand supports Proust’s hypothesis that metacognition and first-order cognition are not based on the same kind of information, yet on the other it speaks against a serial interpretation according to which feelings arise out of automatic processes and are then re-described in propositional terms and used in first-person inferential reasoning. Liu & Wang (2014), for example, argue that motivational intensity influences the effect of positive affect on cognitive control: low-approach motivated positive affect enhances cognitive flexibility and distractibility, while high-approach motivated positive affect (associated with goal pursuit) enhances cognitive stability. Thus, the role of feelings might be broader than just the indicators that may or may not be used in conscious reasoning.