It is easy to think of attention as a purely sensorimotor, exogenous mechanism divorced from the influence of an agent’s preferences and needs. However, according to the active inference framework, such a strict reduction cannot be straightforwardly invoked, since _all_ cognitive and behavioural processes can at least be described as maximising the evidence for a generative model entailed by the ongoing existence of that agent; that is, the minimisation of variational free energy. As such, active inference models could cast an (...) (embodied) cognitive mechanism like attention, described in this paper as a relevance filter, as constrained (or enslaved) by these prior preferences for which an agent must seek evidence, whether or not such priors are having direct, real-time neurocognitive effects on the sensorimotor loops that couple the attending agent and her surrounding environment. This duality with respect to the role of priors corresponds to a wider, ongoing debate in the active inference community regarding the framework’s explanatory power. More specifically, the debate centres on whether the notion of a generative model and the priors embedded ubiqitously therein act as a purely useful instrumental tool for scientists aiming to model the behaviours of self-organising entities, or, rather, the brain (and body) is genuinely constituted by a predictive hierarchy within which higher-order dynamics constrain and contextualise activity unfolding at lower levels. With a focus on the second (ontologically realist) construal of active inference presented here, this paper argues that in cognitive systems endowed with attentional schema, higher-order preferences do, indeed, impose a demonstrable and powerful modulating effect on the way attention unfolds. Furthermore, these preferences in question transcend the contingent, task-relevant goals that have already been shown to bias attention. Rather, attention is powerfully tuned by the most-deep rooted priors the agent possesses, such that, when sensory evidence against these priors is observed and free energy spikes, the agent attentionally prioritises the homeostatic restoration of these preferred states over their shorter-term desires. This suggests that, at its core, attention is a goal-driven process, which calls into question the putative dichotomy that exists between endogenous (goal-directed) attention and exogenous (stimulus-driven) attention. What emerges in its place is a symbiotic relationship between attention and preferences, whereby the fulfilment of the latter rests on successful application of the former, and the former derives its function from the organismic need to find evidence for the latter. (shrink)

We propose a pluralist account of content for predictive processing systems. Our pluralism combines Millikan's teleosemantics with existing structural resemblance accounts. The paper has two goals. First, we outline how a teleosemantic treatment of signal passing in predictive processing systems would work, and how it integrates with structural resemblance accounts. We show that the core explanatory motivations and conceptual machinery of teleosemantics and predictive processing mesh together well. Second, we argue this pluralist approach expands the range of empirical cases to (...) which the predictive processing framework might be successfully applied. This is because our pluralism is practice-oriented. A range of different notions of content are used in the cognitive sciences to explain behaviour, and some of these cases look to employ teleosemantic notions. As a result, our pluralism gives predictive processing the scope to cover these cases. (shrink)

The frame or relevance problem is a classic problem in cognitive science and philosophy. We attempt to resolve this problem by appealing to predictive processing, a growing theory of cognition. As such, it ought to explain one of the central processes of cognition, that is, how an agent context-sensitively determines relevance. Our solution begins by appealing to Bayesian prior probabilities, which intuitively reflect relevance for a predictive agent. However, prior probabilities are necessary but insufficient for solving the problem with predictive (...) processing. We then turn to the broader predictive processing toolbox, leveraging the concepts of prediction, prediction error, and precision in order to explain relevance. This move reveals that the processes that optimize for prediction error minimization are crucial for realizing relevance. Although, they do not yet solve the entire problem, which also demands an agent select relevant actions, based on considerations about their consequences. By appealing to active inference, decision-making, and planning can be brought to bear on relevance, in addition to perceptual inference. With this final inclusion of action (as inference), we suggest predictive processing has the tools to comprehensively solve the problem of relevance. (shrink)

The predictive processing theory refers to a family of theories that take the brain and body of an organism to implement a hierarchically organized predictive model of its environment that works in the service of prediction-error minimization. Several philosophers have wondered how belief-like states of prediction account for the conative role desire plays in motivating a person to act. A compelling response to this challenge has begun to take shape that starts from the idea that certain predictions are prioritized in (...) the predictive processing hierarchy. We use the term “first priors” to refer to such predictions. We will argue that agents use first priors to engage in affective sense-making. What has been missing in the literature that seeks to understand desire in terms of predictive processing is a recognition of the role of affective sense-making in motivating action. We go on to describe how affective sense-making can play a role in the context-sensitive shifting assignments of precision to predictions. Precision expectations refer to estimates of the reliability of predictions of the sensory states that are the consequences of acting. Given the role of affect in modulating precision-estimation, we argue that agents will tend to experience their environment through the lens of their desires as a field of inviting affordances. We will show how PP provides a neurocomputational framework that can bridge between first-person phenomenological descriptions of what it is to be a desiring creature, and a third-person, ecological-enactive analysis of desire. (shrink)

Proponents of the predictive processing (PP) framework often claim that one of the framework’s significant virtues is its unificatory power. What is supposedly unified are predictive processes in the mind, and these are explained in virtue of a common prediction error-minimisation (PEM) schema. In this paper, I argue against the claim that PP currently converges towards a unified explanation of cognitive processes. Although the notion of PEM systematically relates a set of posits such as ‘efficiency’ and ‘hierarchical coding’ into a (...) unified conceptual schema, neither the frameworks’ algorithmic specifications nor its hypotheses about their implementations in the brain are clearly unified. I propose a novel way to understand the fruitfulness of the research program in light of a set of research heuristics that are partly shared with those common to Bayesian reverse engineering. An interesting consequence of this proposal is that pluralism is at least as important as unification to promote the positive development of the predictive mind. (shrink)

How are new Bayesian hypotheses generated within the framework of predictive processing? This explanatory framework purports to provide a unified, systematic explanation of cognition by appealing to Bayes rule and hierarchical Bayesian machinery alone. Given that the generation of new hypotheses is fundamental to Bayesian inference, the predictive processing framework faces an important challenge in this regard. By examining several cognitive‐level and neurobiological architecture‐inspired models of hypothesis generation, we argue that there is an essential difference between the two types of (...) models. Cognitive‐level models do not specify how they can be implemented in brains and include structures and assumptions that are external to the predictive processing framework. By contrast, neurobiological architecture‐inspired models, which aim to better resemble brain processes, fail to explain important capacities of cognition, such as categorization and few‐shot learning. The “scaling‐up” challenge for proponents of predictive processing is to explain the relationship between these two types of models using only the theoretical and conceptual machinery of Bayesian inference. (shrink)

Convincing narratives are not confabulations. Presumably they “feel right” to decision-making agents because the probabilities they assign intuitively (i.e., implicitly) to potential outcomes are plausible. Can we render explicit the calculations that would be performed by a decision-making agent to evaluate the plausibility of competing narratives? And if we can, what, exactly, makes a narrative “feel right” to an agent?