This paper has surveyed predictive processing (PP) from the unusual viewpoint of cybernetic origins in active homeostatic control (Ashby 1952; Conant & Ashby 1970). This shifts the perspective from perceptual inference as furnishing representations of the external world for the consumption of general-purpose cognitive mechanisms, towards model-based predictive control as a primary survival imperative from which perception, action, and cognition ensue. This view is aligned with the free energy principle (Friston 2010); however it attempts to account for specific cognitive and phenomenological properties, rather than for adaptive systems in general. Several implications follow from these considerations. Emotion becomes a process of active interoceptive inference (Seth 2013)—a process that also recruits autonomic regulation and influences intuitive decision-making through behavioural allostasis. A common predictive principle underlying interoception and exteroception also provides an integrative view of the neurocognitive mechanisms underlying embodied selfhood, in particular the experience of body ownership (Apps & Tsakiris 2014; Limanowski & Blankenburg 2013; Suzuki et al. 2013). In this view, the experience of embodied selfhood is specified by the brain’s “best guess” of those signals most likely to be “me” across exteroceptive and interoceptive domains. From the perspective of cybernetics the embodied self is both that which needs to be homeostatically maintained and also the medium through which allostatic interactions are expressed.
A second influential line deriving from cybernetics sets PP within the broader context of model-based versus enactivist perspectives on cognitive science. On one hand, cybernetics has been cited in support of non-representational cognitive science in virtue of its showing how simple mechanisms can give rise to complex and apparently goal-directed behaviour by capitalizing on agent-environment interactions, mediated by the body (Pfeifer & Scheier 1999). On the other, the cybernetic legacy shows how PP can put mechanistic flesh on the philosophical bones of enactivism, but only by embracing a finessed form of representationalism (Seth 2014b). A key concept within enactive cognitive science is that of mastery of sensorimotor contingencies (SMCs). This concept is useful for understanding the qualitative character of distinct perceptual modalities, yet as expressed within enactivism it lacks a firm implementation basis. “Predictive Perception of SensoriMotor Contingencies” (PPSMC) addresses this challenge by proposing that SMCs are implemented by predictive models of sensorimotor relations, underpinned by the continuity between perception and action entailed by active inference. Mastery of sensorimotor contingencies depends on predictive models of counterfactual probability densities that specify the likely causes of sensory signals that would occur were specific actions taken. By relating PP to key concepts in enactivism, this theory is able to account for phenomenological features well treated by the latter, such as the experience of perceptual presence (and its absence in cases like synaesthesia).
Considering these issues leads to distinct readings of active inference, which at its most general implies the selective sampling of sensory signals to minimize uncertainty about perceptual predictions. At a finer grain, active inference can involve performing actions to confirm current predictions, to disconfirm current predictions, or to disambiguate competing predictions. These different senses rest on the concept of counterfactually-equipped predictive models; and they generalize the free energy principle to include Bayesian-model comparison as well as optimization and inference.
In summary, the ideas outlined in this paper provide a distinctive integration of predictive processing, cybernetics, and enactivism. This rich blend dissolves apparent tensions between internalist and enactivist (model-based and model-free) views on the neural mechanisms underlying perception, cognition, and action; it elaborates common predictive mechanisms underlying perception and control of self and world; it provides a new view of emotion as active interoceptive inference, and it shows how “counterfactual” predictive processing can account for the phenomenology of conscious presence and its absence in specific situations. It also finesses the concept of active inference to engage distinct forms of hypothesis testing that prescribe different sampling actions (one bonus is that the “dark room problem” is elegantly solved). At the same time, new and difficult challenges arise in validating these ideas experimentally and in distinguishing them from alternative explanations that do not rely on internally-realised inferential mechanisms.
Acknowledgements
I am grateful to the Dr. Mortimer and Theresa Sackler Foundation, which supports the work of the Sackler Centre for Consciousness Science. This work was also supported by ERC FP7 grant CEEDs (FP7-ICT-2009-5, 258749). Many thanks to Thomas Metzinger and Jennifer Windt for inviting me to make this contribution, and for the insightful and helpful reviewer comments they solicited. I’m also grateful to Kevin O’Regan and Jan Dagenaar for inviting me to speak at a symposium entitled “Consciousness without inner models?” (London, April 2014), which provided a feisty forum for debating some of the ideas presented here.