Iuliia Pliushch finally makes an interesting suggestion: when self-deception occurs, the believer senses a metacognitive feeling of uneasiness, indicating that her underlying belief-forming process is faulty. This suggestions offers an account of the tension that arises while forming a belief on the basis of motivational, rather than evidential grounds. It would be wrong to interpret her proposal as the claim that finding faulty a belief, or a belief-forming process, involves an appraisal of the content of the belief, or of the kind of process that has been used to form it. As I understand her, Pliushch is rather claiming, as psychologists and neuroscientists of metacognition do, that the mind is able to detect fault in the dynamical properties of the underlying processes. Pliushch argues further that, in contrast (she claims) with my own proposal, monitoring not only occurs "before or after a cognitive process, but also during it". There is no real conflict, however, about this claim. Presence of intermediate monitoring depends on the temporal extension of the mental action considered. When confronted with perceptual or memorial uncertainty, there is only control-based, mainly unconscious, intermediate monitoring; intermediate becomes prominent, however, in prolonged, effortful actions, such as problem solving Ackerman 2013). I agree with Pliushch, however, that representing a mental action merely in terms of a starting and end points misrepresents the facts: it is based on a serial view that does not fit the dynamic character of metacognition (as already discussed in section 7 above). The evidence presented in Proust (2013) suggests that retrospective evaluation is based on the underlying dynamic of the whole action (the rate of accumulation in favor of a dominant response, as well as the dispersion of the neural responses), while predictive evaluation is based on the dynamics elicited by the command for this action, as compared with a stored standard (the complexity of the feedback used is addressed in Koriat et al. 2006). An epistemic evaluation, however, has two functions: stop the action, and encourage its continued performance, hence the role of polar valence in motivating action, which is reflected in the bi-partition of evaluations in two classes. This is in close agreement with how predictive coding, as any other theory of emotion and action, describes the facts.
Does predictive coding offer new insights on metacognition? The concept of "transition probabilities" mentioned by Pliushch, is shared by all theorists working on neural dynamics, as well by theorists of recurrent feedback; the concept of free-energy minimization, related to the minimization of surprise, seems prima facie to be consonant with Rescorla & Wagner’s (1972) well established model of reinforcement learning. There is an internal connection between free energy minimization and the evaluation of one's own uncertainty, because it is adaptive to predict one's chances of being uncorrect, and hence avoid surprising failures. The concept of free energy, however, is no more equipped to provide any mechanistic account of brain function as any other evolutionary theory. "It is nothing more that principle of least action applied to information theory", Friston recognizes (Friston et al. 2012). Indeed a prominent problem remains to be solved, concerning how priors vary as a function of task demands and of environmental statistics. Unpacking the principle across adaptive time-scales and survival contexts is indeed a complex future goal. Ways in which predictive coding might enrich the analysis of metacognition with new descriptive, operational tools or new functional explanations remain, then, to be specified.
Pliushch claims further that a first step in the proposed metacognitive theory of self-deception consists in recognizing that metacognitive feelings must be "extended to unconscious belief forming processes". If what is meant is that the dynamic properties that elicit feelings belong to such processes, there is universal agreement on this claim (see the so-called “cross-over principle” between unconscious heuristics and representations (including beliefs) and conscious feelings in Koriat 2000). What is meant, then, by the suggested "extension" is unclear. If what is meant, rather, is that the feelings themselves might be unconscious, this is a possibility that is taken seriously in studies of metaperception in blindsight patients (Reder & Schunn 1996). The very existence of such feelings complicates the phenomenologist's task. A second step is claimed to consist in "clarifying the representational content of tension". Although more detailed work needs to be done in order to better understand the contrast between perceptual and conceptual fluency, intuitivity is generally identified as a variety of what experimental psychologists call "feelings of fluency". One suggestion is that what creates feelings of tension or dysfluency in self-deception is not merely the representation that "the cognitive process violates some important goal representation", but rather, that it violates an implicit heuristic of self-consistency, as discussed in Koriat (2012). Another suggestion is that tension has to do with the realization that the effort initially planned for a current task needs to be upgraded, which is a source of anxiety (Ackerman 2013). In summary: belief-forming processes are known to elicit metacognitive feelings. It remains to be shown how a metacognitive analysis of self-deception might enlighten philosophical and epistemological views about it. Self-deception is a good test case for making the point that conceptual-inferential processing also conveys non-conceptual information.