Computational Models of Hallucinations

The promises and pitfalls of applying computational models to neurological and psychiatric disorders

“…Yet, the general notion underlying Bayesian models has been invoked to support several profoundly different conceptualizations of how hallucinations may arise. First, an increased weighting of prior expectation in perception, such that expectations generate inaccurate percepts ( Friston, 2005 b ; Corlett et al , 2009 ; Fletcher and Frith, 2009 ; Frith and Friston, 2013 ; Teufel et al , 2015 ), second, a reduction in the relative weighting of prior expectation such that it is the relatively stronger bottom-up signal that generates the aberrant percept ( Adams et al , 2013 ), and finally, a circularity in inferential processing such that a lack of inhibitory control generates a reverberating effect: an expectation enhances a signal which then acts as additional evidence in favour of that expectation ( Jardri and Denève, 2013 , 2014 ). While some of these models might not be mutually exclusive, clearly, a computational framework that is able to encompass such differing possibilities needs close scrutiny.…”

Auditory hallucinations

“…As noted by Jardri and Denève (2013), the attractor network model is a means of investigating the functional stability of networks, and, more specifically, to study how at the synaptic level alterations in ion channels and neurotransmitters may affect spiking activity within such networks. In addition, it offers an unprecedented opportunity to study the mediation of auditory hallucinations by examining the interrelationships between networks and biochemical agents, structural and functional connectivity, phenomenologic characteristics of hallucinations, psychologic factors, and the social -as well as wider environmental -context.…”

Perceptual phenotypes: Perceptual gains and losses in synesthesia and schizophrenia

“…Here, we focus on schizophrenia and synesthesia and investigate whether alterations in perceptual inference may explain their diverse perceptual phenotypes. In schizophrenia, forward propagation of sensory evidence without appropriate top-down corrective priors may cause overreliance on sensory information (Jardri and Denève, 2013a;Jardri and Denève, 2013b;Sterzer et al, in press) thereby explaining for instance the paradoxical immunity to perceptual illusions observed in schizophrenics (Uhlhaas et al, 2004). In synesthesia, excess (top-down) priors may explain the experience of concurrent sensations (e.g., color) without actual sensory input (e.g., black letters) (Seth, 2014;van Leeuwen, 2014).…”

“…Alternatively, excessive neural communication due to increased white matter connectivity might lead to enhanced top-down propagation of priors (Rouw and Scholte, 2007). Excess forward propagation (schizophrenia) or excess priors (synesthesia) may then erroneously be re-interpreted along the cortical hierarchy, leading to 'circular inference' (Jardri and Denève, 2013b;Jardri et al, 2017) and help explain insensitivity to perceptual illusions and the emergence of additional percepts, respectively.…”