What exactly is it that makes the experiential content of some, but not all, conscious states appear as irrevocably real to us? And how can there be degrees in experiential realness? Seth pursues what is known as an “intentionalist” strategy in philosophy of mind (Brentano, 1874; Crane, 2009). His strategy is to reduce a specific phenomenal property (namely, “perceptual presence”) to a specific form of representational content, which is carried by dynamical neural representations as described under the predictive processing approach (PP). A phenomenal property is a property of conscious experience, as experientially accessed from a subjective and individual first-person perspective (1PP). Any convincing neuroscientific theory of consciousness will therefore, first, have to tell us what a 1PP is and, second, describe the fine-grained mechanics by which a given phenomenal target property is instantiated in the brain. A first advantage of Seth’s strategy is that it offers us a novel and more precise understanding of what it means that the perceptual contents of conscious experience is appearance only, however immediate and mind-independent it may seem. Perceptual content always is counterfactual content, because the underlying generative models incorporate explicitly counterfactual elements related to how sensory inputs would change on the basis of a broad repertoire of possible actions, even if those actions are not in fact performed. Second, Seth then usefully applies the idea of maximizing the salience of counterfactual representations and thereby arriving at explicit percepts though a process of actively induced circular causality (Friston, Adams, Perrinet, & Breakspear, 2012) to the special case of conscious experience. Third, his proposal is clearly internalist, in good keeping with a widespread consensus among philosophers that phenomenal properties locally supervene on contemporaneous functional and/or physical properties of the brain. Therefore, his proposal opens a route to localist, reductive explanations of high-level phenomenal properties like “realness,” “presence,” or “mind-independence.” While Noё’s persisting conceptual difficulty is to show that interaction with the physical world is metaphysically necessary for presence (and not only causally enabling; cf. Block, 2005), the great advantage of Seth’s new idea is that it allows us to tell a fully internalist story. “Presence” can now be seen to be an entirely local property of brain processes only. Anil Seth’s contribution to the second question mentioned above is that he takes a genuine step forward in isolating the minimal metaphysically sufficient condition for perceptual experience.

One conceptual problem with taking “perceptual presence” seriously as a well-defined explanandum for the cognitive neuroscience of consciousness is that it has two major semantic components: “nowness” (i.e., localization in a temporal frame of reference) and “realness” (i.e., being non-representational, and therefore mind-independent). “Veridicality,” strictly speaking, refers to a property of representations and is exactly not what one wants in explaining either of those two components, because the target phenomenology is not one of having a “subjectively veridical” representation of reality, but of now being directly in touch with reality itself. Interestingly, only if the conscious brain explicitly treats information as unreliable do we find the converse effect, namely, subjective non-veridicality. In experiencing synesthetic concurrents or pseudo-hallucinations, subjects are typically aware that they are experiencing a misrepresentation, something that is unreliable and somehow “in the mind.” A third, and related, conceptual complication is that we also find presence and the associated gradient of realness in the human self-model, with the bodily self being perceived as real and present, while the cognitive self-model is experienced as comprised of representations.

However, there are now at least two competing internalist approaches for the conscious experience of “realness.” Both are grounded in empirical data, and both tell a specific microfunctional story about what is sufficient for perceptual presence. Seth claims that “unreal” conscious contents depend on counterfactually poor generative models of their external causes, lacking a corresponding rich world-related statistical structure for such models to learn. But the older theory of “phenomenal transparency” says that it is the accessibility of earlier processing stages for attentional processing that leads to phenomenal “unrealness” (see Metzinger, 2003a, 3.2.7 and Metzinger, 2003b for details).

“Transparency” refers to a property of conscious representations, namely, that they are not experienced as representations. Therefore, the subject of experience feels as if being in direct and immediate contact with their content. Transparent conscious representations create the phenomenology of naive realism. An opaque phenomenal representation is one that is experienced as a representation, for example, in pseudo-hallucinations, synesthetic concurrents, or lucid dreams. Unconscious representations are neither transparent nor opaque. Moreover, there exists a graded spectrum between transparency and opacity, determining the variable phenomenology of “mind-independence” or “realness.” This phenomenology of “realness” can be analyzed as the phenomenal transparency of certain conscious representations, i.e., as the fact that the system has no introspective access to non-intentional properties of its own representations, that it is necessarily unaware of the construction process.

If we now move from the representationalist to the functionalist level of analysis, we may interpret experiential “realness” as the Bayes-optimality, counterfactual richness, or plain dynamical stability of the generative models employed, or perhaps also as the claim that further optimization of precision expectations is not possible. My own claim is that transparency is the phenomenal signature of epistemic reliability.

Two other excellent, empirically well-documented examples are avatars in experimentally induced full-body illusions (Blanke, 2012; Lenggenhager, Tadi, Metzinger, & Blanke, 2007) and the transition from two-dimensional hypnagogic images at sleep-onset to the fully realistic and immersive dream state (Windt, 2010, forthcoming). The neural avatar model is counterfactually poor, just as visual hypnagogic imagery is, and their “realness” exactly increases together with the sense of identification and the degree of experienced self-location in a spatiotemporal reference frame. Just like mental imagery, they initially lack perceptual presence, but may gradually gain this quality (cf. Table 2, bottom line). We may interpret shifts of this kind as the weakening of a high-level prior of image-hood, for example, by saying that the visually perceived avatar is a counterfactually shallow body model, because interoceptive sensations cannot be predicted (Seth, 2013). But we could also say that the brain has recognized that, although visually compatible, it is only a unimodal precursor to a full-blown self-model, just as two-dimensional hypnagogic imagery or the synesthete’s concurrents are immediately detected as probably belonging to an earlier processing stage, in which spatial properties are still poorly expressed by intermediate-level HGMs. Insofar as the brain has to extract features that are invariant under counterfactual manipulation, the corresponding epistemic reliability—i.e., the degree of certainty that the system has successfully isolated and latched onto an invariant property of reality—increases with the counterfactual richness of a model. What decreases is the probability for the existence of features that actually do change under non-represented counterfactual manipulations.

Therefore, the degree of counterfactual richness may be an indicator of the overall processing level currently achieved—a highly relevant functional property that directly correlates with epistemic reliability. This property can (and must) itself be predicted by high-level models. Perceptual presence as well as transparency therefore express epistemic reliability on the level of conscious processing. My second, and much more speculative, proposal is that, at any given point in time, it is exactly the region of maximal invariance and reliability that constitutes the origin of the 1PP.