Time-consciousness in computational phenomenology: a temporal analysis of active inference

Abstract

Time plays a significant role in science and everyday life. Despite being experienced as a continuous flow, computational models of consciousness are typically restricted to a sequential temporal structure. This difference poses a serious challenge for computational phenomenology-a novel field combining phenomenology and computational modelling. By analysing the temporal structure of the active inference framework, we show that an integrated continuity of time can be achieved by merging Husserlian temporality with a sequential order of time. We also show that a Markov blanket of the present moment integrates past and future moments of both subjective temporality and objective time in an asynchronous manner. By applying the integrated continuity, it is clear that active inference makes use of both subjective temporality and objective time in an integrated fashion. We conclude that active inference, on a temporal note, qualifies as a computational model for phenomenological investigations.

Keywords: active inference; computational phenomenology; consciousness; temporal analysis; temporality.

Figure 1.

A schematic of the Husserlian analysis of subjective temporality (inspired by Husserl 2001b, p. 22). The horizontal x-axis refers to the objective time, while the axes above the x-axis refer to the protention and the axes below refer to retention in subjective time.

Figure 2.

This schematic illustrates three kinds of conditional dependencies. (a) This conditional dependency illustrates the conventional sequential temporal structure in objective time. In the terminology of causal inference, this process is a convention causal path. (b) This conditional dependency illustrates an interpenetrated temporal structure as discovered in Husserl’s model of time. In the terminology of causal inference, this process can be described as a collider. In the context of causal inferences, conditional dependencies within a Bayesian network are usually taken to map onto causal processes (Peters et al. 2017, pp. 105–106). Thus, it may be assumed that some of the conditional dependencies described here map onto causal processes related to objective time. Notice, however, in relation to a phenomenological analysis of subjective temporality, the relationships between nodes are not causal but of ‘motivation’. From the phenomenologists’ perspective, the relationships between phenomena that they are interested in are not of the natural kind because they are subjective. Causality, in contrast, is a natural and objective process in the world. For this reason, phenomenologists often avoid talking about causality and prefer the concept of ‘motivation’: ‘One phenomenon triggers another not through some objective causality, such as the one linking together the events of nature, but rather through the sense it offers’ (Merleau-Ponty 2012, p. 51). The link between causality, motivation, and conditional dependencies may prove to be crucial for the naturalization of phenomenology and the phenomenologization of science from the perspective of computational phenomenology. As stated by Carel and Meacham, it is the ‘relation or non-relation between the orders of causation (nature) and motivation (experience) that remains the decisive issue in any exploration of the relationship between phenomenology and naturalism’ (Carel and Meacham 2013, p. 7). As we see it, in the active inference framework, conditional dependencies may serve as a bridge between motivation and causality. A full-fledged elaboration of this idea remains nevertheless beyond the scope of this paper. (c) This conditional dependency illustrates an integrated continuity taking both models into careful consideration. Note here that objective time is declared in the superscript, whereas subjective temporality is declared in the subscript: . Note also that objective time may be self-referenced, whereas human time is always related to some (conscious) event

Figure 3.

This schematic is a graphical description of the revision of the beliefs about the variables in light of new data. It described the statistical dependencies at a particular node in the integrated continuity. (a) By treating the continuity in terms of conditional dependencies, we may knit a Markov blanket to determine the system constituting the present moment to elucidate the statistical boundaries of the present moment. We effectively consider protention and primal impression as random variables, while the nodes of retention are considered observations or empirical priors in line with active inference. First, it is worth noticing that the blanket exemplifies how the present moment can be described through the integrations of the immediate past and future. Second, it may be observed that an asynchronous structure emerges. The yellow area illustrates the blanket. (b) In computing the probability of the present moment, the DAG suggests that we simply need three other states. The yellow area illustrates the nodes necessary to infer the present. Notice here that everything in retention is signified as as it now belongs to observed states rather than inferred states, whereas the inferred sensory states are signified as , where Ob refers to the objective time, Sb refers to the subjective temporality, and k and n refer to the number of steps in each direction