Uncertainty, epistemics and active inference

Abstract

Biological systems—like ourselves—are constantly faced with uncertainty. Despite noisy sensory data, and volatile environments, creatures appear to actively maintain their integrity. To account for this remarkable ability to make optimal decisions in the face of a capricious world, we propose a generative model that represents the beliefs an agent might possess about their own uncertainty. By simulating a noisy and volatile environment, we demonstrate how uncertainty influences optimal epistemic (visual) foraging. In our simulations, saccades were deployed less frequently to regions with a lower sensory precision, while a greater volatility led to a shorter inhibition of return. These simulations illustrate a principled explanation for some cardinal aspects of visual foraging—and allow us to propose a correspondence between the representation of uncertainty and ascending neuromodulatory systems, complementing that suggested by Yu & Dayan (Yu & Dayan 2005 Neuron 46, 681–692. (doi:10.1016/j.neuron.2005.04.02610.1016/j.neuron.2005.04.026)).

Learning of complex auditory patterns changes intrinsic and feedforward effective connectivity between Heschl’s gyrus and planum temporale

“…In two-state DCM, this gain modulation arises from the excitation/inhibition balance of excitatory pyramidal cells and inhibitory interneurons. This interpretation entails that, besides generating prediction error signals, melodic deviants are afforded a higher precision than standards, which is consistent with the role of salient stimuli in the orientation of attention (Polich and Criado, 2006;Hannon and Trainor, 2007;Parr and Friston, 2017) . Thus, unexpected sounds would point to the sources in the auditory scene that are most informative and relevant and need to be prioritized for processing through gain mechanisms.…”

Eye pupil signals information gain

“…Pupillary responses, because of their relation to the noradrenergic system [71], have previously been linked to unexpected uncertainty [27,84], sometimes taken as synonym to surprise [7,28,6] and sometimes as an equivalent of volatility, i.e. how likely the environment dynamics is to change [84,27,21,85,86,87]. These two definitions are strongly related since surprising observations suggest that the statistical structure of the environment may have changed [88].…”

“…While surprise is event-related and could be linked to phasic pupillary changes [28], volatility varies slowly and could be related to tonic pupil size [27]. Unexpected uncertainty relates also strongly to the problem of exploitation/exploration trade-off, another concept linked to pupillary responses [30,29,83,89]: when confidence in the internal model of the environment drops following surprising observations, exploitation strategies lose value with respect to alternative exploration strategies [84]. However, recent data has shown that variations of tonic pupil size are not indicative of unexpected uncertainty, but are rather a signature of reducible uncertainty (ambiguity resulting from poor model of environment, caused by undersampling; Krishnamurthy et al [21]) or expected uncertainty (related to the variance of the task; De Berker et al [12]).…”

“…However, since volatility is a slow-changing property of the environment, this observed correlation with phasic pupillary changes must reflect the fact that, when prior knowledge on environment is unreliable (i.e. volatility is high), more weight is given to new sensory evidence, as opposed to prior biases [90,84,27], and model updates between prior and posterior beliefs are more expansive [90], leading to larger pupillary dilations. Overall, current evidence does not seem to favour the view that pupil dilation would be indicative of specific types of uncertainty but, as I argue in the present work, would rather signal information processing, which itself depends strongly on uncertainty conditions.…”

Simulating emotions: An active inference model of emotional state inference and emotion concept learning

“…Changes in precision were implemented via a temperature parameter of a softmax function applied to a fully precise version of likelihood mappings between emotion concepts and the 5 types of lower-level information that the agent could attend to (where a higher value indicates higher precision). For a more technical account of this type of manipulation, please see (Parr & Friston, 2017a). under different levels of precision for two parameters (denoted by temperature values for a softmax function controlling the specificity of the A and B matrices for hidden state factor 1; higher values indicate higher precision).…”