Proust (this collection, as well as 2013) argues that mental actions are preceded and followed by metacognitive feelings indicating the appropriateness of the cognitive process in question. I want to argue that tension in self-deception fits the characterisation of a metacognitive feeling. Tension is described as a feeling of uneasiness and distress, and as such I think that it is precisely this tension that is said to indicate to the self-deceiver that her belief-forming process is faulty.
Self-deception (SD) is a motivated (1) kind of typically subpersonal hypothesis-testing (2) that results in an evidence-incompatible mental representation of reality (3) which fulfils a belief-like role (4) (Pliushch & Metzinger 2015). Self-deception is usually discussed in the context of biased belief-forming processes and it is argued that phenomenological tension arises as a result of the execution of such processes (e.g., Lynch 2012). Thus, the same function has been ascribed to tension in self-deception as the one ascribed by Proust to metacognitive feelings, namely a comparison of the cognitive process to certain criteria. In self-deception, rationality criteria are typically emphasised.
I want to argue that metacognitive feelings apply to self-deception, insofar as they might also monitor unconscious cognitive processes and arise not only before or after a cognitive process, but also during it. In case of self-deception these cognitive processes are belief-forming processes. Proust (this collection, 2013) considers conscious mental actions: her argument is that unconscious comparison processes that give rise to metacognitive feelings precede and follow conscious mental actions. She argues that the “attentional-supervisory system” emerges from “distributed metacognitive abilities” (Proust 2013, p. 263). Ignorance of epistemic norms such as relevance, coherence, fluency, and informativeness lead to (pathological) errors in belief acquisition (Proust 2013, pp. 260–261). My argument in favour of the extension of metacognitive feelings to monitor unconscious cognitive processes is of a phenomenological nature. I agree with Proust (this collection) that the term “inference” has been used loosely in the literature and does not always indicate a first-person inference (p. 21). Yet the more basic problem might be that there is no sequential first-person inference as such in the first place. If the shift between mind wandering (task-unrelated cognitive activity) and task-directed cognitive activity goes unnoticed (Metzinger 2013), then there might be other shifts that we do not notice, e.g., the shift from unconscious to conscious cognitive processes, or some changes in the given process. Thus, the phenomenology of a cognitive process might be more complicated than a unified sequence with a starting point and an end. Further, given, for example, mood-state dependent cognition (Eysenck & Keane 2010, pp. 584), I doubt the plausibility of the assumption that only in breaks between conscious cognitive processes do subjects experience affective feelings.
In the previous paragraphs I argued that the functional role of metacognitive feelings fits that of tension in self-deception, and that metacognitive feelings arise not only before and after mental actions, but also before, after, and during unconscious (possibly self-deceptive) cognitive processes. In this paragraph I want to link Proust’s idea that feelings possess valence only if the rate of change of progress is unexpected to predictive coding, in order to provide a functional description of metacognitive feelings. Proust (this collection) argues that the affective quality of feelings arises only if the cognitive process violates expectations: if it progresses quicker towards the goal, positive feelings arise, if slower, negative feelings arise[8] (p. 21). Given that the terms “expectation” and “prediction error” have gained popularity in virtue of being key terms in predictive coding, which is a modelling strategy explaining perception, cognition, and action (Clark 2013), I will shortly discuss Proust’s claim about affect in metacognitive feelings in the context of predictive coding. According to predictive coding, prediction errors (deviation between expectation and outcome) are precision-weighted. Precision is the property of prediction errors (errors between the top-down prediction and the bottom-up signal one receives) that can be described as the weight of a prediction error that plays the role of selection: the more precise the prediction error, the more it will change the hypothesis about causes of input. Switching between perception and action depends on the precision of prediction errors: precise prediction errors change hypotheses, while imprecise ones lead to action (Brown et al. 2013). Precision[9] is also argued to play a dual biasing role: biasing perception toward goal states and enhancing confidence in action choices (Friston et al. 2013). Low precision of prediction errors has been argued to cause anxiety (Mathys et al. 2011, p. 17).[10] I argue that Proust’s proposal that violations of expectations of “a given rate of reduction of the discrepancies toward her [agent’s] cognitive goal” (this collection, p. 26) produce affective feelings might be described in predictive coding terms as violations of transition probabilities of reaching the goal state:[11] if a state conducive to the goal state or a goal state itself has been reached, despite a low probability of changing into that state from the current state, then positive affective feelings might arise.[12]
The first step in the categorisation of tension as a metacognitive feeling has been an extention of the application of metacognitive feelings to unconscious belief-forming processes. The second is to clarify the representational content of tension. To do the latter, it might be beneficial to consider which other kinds of metacognitive feelings arise out of belief-forming processes. Those are intuitivity, counter-intuitivity, and anxiety, if one classifies them according to the phenomenology and not according to the norm that they control. Intuitivity indicates the appropriateness of a given belief-forming process.[13] The reason for the ascription of the given functional role to intuitivity is that intuivitity signals 1) a good fit with respect to the network of our explicit background beliefs and 2) a good fit with respect to our conscious and unconscious model of reality (Metzinger & Windt 2014). An appropriate belief-forming process provides a good fit with respect to 1) and highly likely also with respect to 2). I further argue that counter-intuitivity represents that a certain cognitive process violates the chosen criterion of appropriateness, but is neutral with respect to the system’s goal representations, while tension or anxiety represents that the cognitive process violates at least some important goal representations. The reason for this distinction is to account for the effect of motivation on belief-forming processes.
Thus, if feelings accompany our belief-forming processes, then readers might have experienced some while reading this commentary: hence the title. To conclude, I think that Proust has offered interesting ideas on the nature of feelings that will greatly contribute to the clarification of the matter: the indexical (affordance-sensing and non-conceptual) format of feelings, their transparency, the taxonomy of feelings into sensory, emotional, agentive, and epistemic, the predictive and retrospective function of feelings signalling the appropriateness of the cognitive process they monitor, and the degree of change of expectation as the origin of valence of feelings. In this review I have tried to extend Proust’s account. To do this, I attempted to provide some conceptual clarifications on the distinction between feelings and emotions, the formal object of feelings, and the conceptual influences to which they might be subject. Last, I argued that tension in self-deception is a kind of metacognitive feeling.
Acknowledgements
I am very grateful to the Barbara-Wengeler foundation for generously supporting my PhD project and all the reviewers of this commentary for their helpful improvements.