Presence, objecthood, and the phenomenology of predictive perception

Abstract

Can perceptual presence be explained by counterfactually-rich predictive models linking perception and action? Considering an unusually rich range of responses to this idea has led me to (1) re-emphasize the core conceptual commitment of “predictive processing of sensorimotor contingencies” (PPSMC) to predictive model-based perception, (2) reconsider the relationship between presence and objecthood, and (3) refine the phenomenological target by differentiating between perceptual presence and the phenomenology of absence-of-presence, or “phenomenal unreality.” It turns out that this requires blue-sky thinking.

Keywords:

• Perceptual presence
• Objecthood
• Predictive processing
• Active inference

I am grateful to the Dr. Mortimer and Theresa Sackler Foundation, which support the work of the Sackler Centre for Consciousness Science. Many thanks also to my commentators for providing challenging and insightful commentaries.

Notes

¹ The idea that perception results from interactions between top-down and bottom-up (inside-out and outside-in) signaling is, by itself, agnostic about Bayesianism. Non-Bayesian examples of this idea include Grossberg’s adaptive resonance theory (ART), where top-down perceptual “expectations” reciprocally reinforce compatible bottom-up inputs (Grossberg, 2013; see also for an early neurorobotic study of feature binding, Seth, McKinstry, Edelman, & Krichmar, 2004). In these cases top-down signals excite rather than inhibit bottom-up responses, which may seem contrary to PP. But even within PP, some top-down input can be facilitatory, when increasing the “gain” on sensory inputs via precision weighting (Feldman & Friston, 2010). At the same time, theories like ART can easily be given a Bayesian gloss. These points underline the need to translate theoretical frameworks into neurocomputational mechanisms which make empirically testable predictions (Gotts & Martin, 2014; Summerfield & De Lange, 2014).

² On the Laplace assumption probability distributions are encoded simply by their mean and variance—these being the so-called “sufficient statistics.”

³ The dependency between presence and objecthood explains why it is hard (perhaps impossible) to keep perceptual content absolutely constant while modifying presence (Di Paolo, 2014; Hohwy, 2014).

⁴ Predictions about counterfactual richness are predictions about higher-order properties of probability distributions. It is useful to think of this by analogy with “precision expectations,” which are predictions about the precision (inverse variance) of target signals, and which are associated with attention. (See Hohwy, 2013, Chapter 9, for a summary.) In the original Discussion Paper, predictions about counterfactual richness can be equated with the more informal “high-level priors” of “objecthood” (rich) and “image-hood” (poor), described on pp. 105–106.

⁵ This seems similar to Hohwy’s suggestion that “[presence] would be shaped by increased openness to precise input rather than by prospective coding of this input itself” (Hohwy, 2014, p. 128).

⁶ Another speculative connection concerns Friston’s concept of Markov blankets, which refers to the sequestration or causal encapsulation of parts of predictive hierarchies (Friston, 2014). Possibly, signal exchanges lying comfortably beneath Markov blankets may be unavailable for attentional processing.

⁷ Van Leeuwen (2014) suggests that “if synaesthetic experiences would become manifest by top-down connections alone, all synesthetes would be ‘associators’” (p. 125). This seems wrong on two counts. First, on most predictive processing theories (including PPSMC) perceptual content rests on both top-down and bottom-up processes. Second, on PPSMC the difference between “projectors” and “associators” lies in the different kinds of counterfactual top-down predictions involved.

Additional information

Funding

I am grateful to the Dr. Mortimer and Theresa Sackler Foundation, which support the work of the Sackler Centre for Consciousness Science. Many thanks also to my commentators for providing challenging and insightful commentaries.