Perceptual presence occurs when “veridical perceptual scenes are world‐revealing” such that “they comprise a world of objects rather than ‘perspectival takes’ on objects.” In spite of Seth’s excellent examples, explanations, and taxonomy, I confess I struggle somewhat to pin down confidently the phenomenology of perceptual awareness of sensory attributes that are strictly unperceived. For example, I think our conception of presence would be helped by examples where two experiences have identical content but one has subjective veridicality (presence) and the other none at all (and where doxastic veridicality doesn’t explain any difference). However, such examples do not seem readily available. Another issue concerns the difference between spatial presence (the back of the tomato), which invites elements of action, and temporal presence (cf. the specious present, with protention of the as yet unpresented tone), which seems far more passive. Presence seems to be one of these familiar and critical, but somewhat elusive, phenomenological concepts, leaving it a difficult target for analysis.

Seth reasonably conceptualizes presence in terms of enactivism. He argues rightly that with mastery of sensorimotor contingencies would come coding of a rich counterfactual element (which is anathema to enactivism). Presence is then underpinned by rich, explicit predictive processing of what sensory input would be like were the agent to move thus and so. Concurrents lack rich counterfactuals so they lack presence.

The counterfactuals are sourced in the mechanism for active inference but Seth gives them a role in phenomenology. It is this move I’ll now probe.

When counterfactuals become more precise than the actual representation, then they begin to drive action. However, actual action would change content rather than presence so Seth stipulates that the counterfactuals be fairly imprecise. But now presence begins to seem somewhat epiphenomenal: World-revealing phenomenology is an effect of counterfactuals that are not yet strong enough to do their real job. It also appears to have the counter-intuitive consequence that sense of presence should be weak or absent for experiences associated with strong, unambiguous SMC-predictions.

Seth’s PPSMC is, however, resourceful: It can insist on the subjunctive or prospective element, such that counterfactual predictive processing is running “off-line.” This aligns presence with mechanisms for imagery and action selection. The key question is then how off-line prospective prediction enters awareness to create the elusive phenomenology of presence. It can’t be as content nor as belief: Actual and counterfactual predictions are contradictory so can’t both determine content, and doxastic veridicality (belief that prospective predictions are encoded) cannot engender sufficiently fine-grained subjective presence. In response, one could say that presence arises as action actually unfolds. Presence would then be associated with increased gain on specific, high-precision expected prediction errors in active inference; its phenomenology would be shaped by increased openness to expected precise input rather than by prospective coding of this input itself. This would nicely parallel a natural understanding of the specious present, and since presence would be shaped “on the go” it becomes a moving target, in line with its elusive phenomenology.

It is possible that PPSMC doesn’t capture all aspects of the elusive phenomenology of presence. It is therefore worth revisiting a conception of presence without an element of mastery, namely the PP account (sidelined by Seth), on which presence is determined by representation of invariants in perceptual hierarchical inference. Real objects interact with each other and thereby cause non-linear flows of sensory input (e.g., occlusion). Hierarchical inference, which convolves representations of higher-order invariants, is necessary to predict such non-linear flow, and the implicit inversion of the model when prediction is successful identifies (deconvolves) these more invariant causes. That is, relatively perspective independent causes are revealed in the very process of prediction error minimization (see Hohwy, 2013). This comes very close to one conception of the sense of presence namely, that it is world-revealing rather than merely perspectival. The lack of presence for concurrents is then easily explained: Concurrents are causes that do not interact on their own with other causes (presumably a fence won’t occlude a concurrent). As such, concurrents produce a fairly linear flow of sensory input and no higher-order invariant is needed to explain away their occurrence. This means their inverted model is very shallow, and reveals nothing much about the causal structure of the world—all of which seems a good fit for an absence of a sense of presence. This PP account may help firm up our conception of presence: An experience with presence and an experience without presence can be identical in terms of the actual sensory input predicted under the generative model at a given moment, but different in terms of depth of the inverted model.

The PP account mimics the “rich-poor” element of Seth’s PPSMC and they could potentially co-exist—perhaps presence is modulated both “upstream” by invariants and “downstream” by counterfactuals and action.