Active Inference, Epistemic Value, and Uncertainty in Conceptual Disorganization in First-Episode Schizophrenia

Abstract

Background and hypothesis: Active inference has become an influential concept in psychopathology. We apply active inference to investigate conceptual disorganization in first-episode schizophrenia. We conceptualize speech production as a decision-making process affected by the latent "conceptual organization"-as a special case of uncertainty about the causes of sensory information. Uncertainty is both minimized via speech production-in which function words index conceptual organization in terms of analytic thinking-and tracked by a domain-general salience network. We hypothesize that analytic thinking depends on conceptual organization. Therefore, conceptual disorganization in schizophrenia would be both indexed by low conceptual organization and reflected in the effective connectivity within the salience network.

Study design: With 1-minute speech samples from a picture description task and resting state fMRI from 30 patients and 30 healthy subjects, we employed dynamic causal and probabilistic graphical models to investigate if the effective connectivity of the salience network underwrites conceptual organization.

Study results: Low analytic thinking scores index low conceptual organization which affects diagnostic status. The influence of the anterior insula on the anterior cingulate cortex and the self-inhibition within the anterior cingulate cortex are elevated given low conceptual organization (ie, conceptual disorganization).

Conclusions: Conceptual organization, a construct that explains formal thought disorder, can be modeled in an active inference framework and studied in relation to putative neural substrates of disrupted language in schizophrenia. This provides a critical advance to move away from rating-scale scores to deeper constructs in the pursuit of the pathophysiology of formal thought disorder.

Keywords: bayes network; conceptual organization; dynamic causal models; free energy principle; thought disorder.

Fig. 1.

A cartoon summary of the generative model of speech production during the TLI interview. Although there is a degree of communication with another person, the interview used in this study focuses on the ideational function of language, to make sense of our external world, actualizing its epistemic property. Figure created with Toonytools (classic.toonytools.com).

Fig. 2.

Toy example of variational free energy minimization during language production. The agent intents to make sense of a picture via their embodied generative model (equation 1). In 2 scenarios (high, A, and low, B, uncertainty), the agent brings to the table the same pair of possible (prior) representations (3D images) of the world with relevant prior probabilities, P(x), and prior uncertainty (σ²). To reduce the uncertainty, the agent speaks aloud and hears their own utterances. Hearing 2 different productions (with different analytic thinking scores, At) would differentially minimize the uncertainty (ie, VFE). However, hearing the same utterance with high At (y₁) would lead to a more substantial minimization (σ²= 0.06) in the scenario with a low prior uncertainty, σ²= 0.13 (ie, high prior conceptual organization). In this example, the posterior beliefs (uncertainty or conceptual organization) can be computed directly from equation 2. However, p(y) is intractable in real situations. Instead, posteriors are found via VFE minimization (equation 3)—testing different approximate posteriors, q, until converging with the true posterior. Figure created with Toonytools (classic.toonytools.com).

Fig. 3.

Active inference and probabilistic graphical model of the POMDP of speech production in the TLI interview. The subgraph (dashed rectangle) allows us to estimate the distribution of Co over groups at t = 0. At (analytic thinking score), Co (conceptual organization), rsDCM (DCM set of parameters of resting state fMRI), FW (function word), π (Policy -ie, a stream of words), TLI (Thought Language Index scores), P2 (Item P2 representing clinician-rated Conceptual Disorganization score from the Positive and Negative Syndrome Scale PANSS), both providing clinical symptoms scores, Gr (group).

Fig. 4.

Bayes networks testing the hypothesis that the analytic thinking score is causally associated with the unobserved Co. The bar graph shows the Co posteriors given group. The middle and right line graphs show the posteriors (mu and standard deviation—error bars) of analytic thinking given Co and group (Gr) respectively. Estimated distribution of Co in M1 (low = 0.58, high = 0.42) were set as priors in the generative model of figure 5. Probability values inside nodes are priors.

Fig. 5.

Bayes networks testing the hypothesis that the effective connectivity parameters of resting state fMRI of the salience networks encode Co. M2 explaining group differences and rsDCM parameters in terms of differences in rating scales underperformed M1 explaining those differences in terms of conceptual organization. Tail-to-tail edges in the Co (conceptual organization) node indicate that after knowing the Co state rating-scale scores or group membership does not add anything about the effective connectivity of the salience network.

Fig. 6.

Probability distributions of effective connectivity parameters conditional upon conceptual organization (Co) level. *** P < .0001.