Dissecting how psychopathic traits are linked to learning in different contexts: A multilevel computational and electrophysiological approach

Abstract

Previous studies suggest that elevated psychopathic traits, linked to social norm violations and personal gain-seeking, may be caused by impairments in associative learning. Recent advances in computational modelling offer insight into the unobservable processes that are thought to underly associative learning. Using such a model, the present study investigated the associations between psychopathic traits in a nonoffender sample and the cognitive computations underlying adaptive behavior during associative learning. We also investigated the potential engagement of adaptive control processes by measuring oscillatory theta activity in the prefrontal cortex. Participants performed a reinforcement learning task in which the trade-off between using social and nonsocial information affected task performance and the associated monetary reward. The findings indicated that increasing levels of psychopathic traits co-occurred with reduced learning from social information and suggested that antisocial traits were linked to a reduced ability to track changes in the trustworthiness of social advice over time. This did not affect the preference for one information source and the risk taken to obtain a high reward. Furthermore, midfrontal theta power was negatively linked to levels of psychopathic traits, aligning with indications that theta is involved in volatility tracking of social information. Importantly, we consider that the task design may reflect reduced sensitivity to secondary, rather than specifically social information. The current study provides support for a relationship between associative learning, theta power, and psychopathic traits and contributes to our understanding of the mechanisms that may explain reduced responsiveness to current treatment interventions in individuals with psychopathy.

Keywords: Associative learning; Computational modelling; Psychopathy; Social learning; Theta power.

Fig. 1

a Overview of the Wager task, adapted from Diaconescu et al., . Each trial consisted of four phases: 1) Advice: advisor recommended a card; 2) Response: participant chose a card; 3) Investment: participant decided how many points to invest; and 4) Feedback: outcome was presented. b Probability scheme for the blue card being correct (top) and advisor trustworthiness (bottom) across trials. Stable phases are highlighted in light grey (Nonsocial: trials 1–25 and 100–160; Social: trials 1–49 and 70–99). Volatile phases are in grey (Nonsocial: trials 26–99; Social: trials 50–69 and 100–159). Adapted from Diaconescu et al. (2020)

Fig. 2

Overview of the analysis pipeline. Note that the correlations are Spearman rank correlations (1000 bootstrap samples)

Fig. 3

Model of the 3-level HGF and the response model. The diamonds represented quantities that change in time (i.e., that carry a time or trial index (k)) but that do not depend on their previous state. The hexagons represented states that change in time but additionally depend on their previous state. Circles denoted fixed parameters. The perceptual model had three layers: (1) χ1 represented the accuracy of the current advice or the rewarded card color; (2) ${\mathrm{\chi }}_2^{(\text{k})}$ represented the perceived likelihood of obtaining trustworthy advice or of the card-outcome contingency on trial (k); and (3) ${\mathrm{\chi }}_3^{(\text{k})}$ represented the participant’s representation of the rate of change in the advisor’s trustworthiness or the card-outcome contingency on trial (k). Parameter $\kappa$ captures how strongly ${\mathrm{\chi }}_2^{(\text{k})}$ and ${\mathrm{\chi }}_3^{(\text{k})}$ are coupled, and $\vartheta$ indicates the estimated rate of change in ${\mathrm{\chi }}_3^{(\text{k})}$. The response model had two layers: 1) the computation of the probability of the outcome given both the nonsocial and the social source; 2) the chosen action. Parameter $\zeta$ reflected the weight of the social- compared to the nonsocial source. $Y$ represents the subject’s binary response ($y$= 1: responding according to the advice, $y$ = 0: responding against the advice). The results on the inverse decision temperature parameter beta ($\beta$) were not considered in the current study. Figure adapted from Diaconescu et al., (2017, 2020)

Fig. 4

Behavioural variables influenced by the volatility. A Performance accuracy; B advice-taking; and C amounts of points wagered. Black diamonds represent the means. Significant effects are flagged (***p <.001). A Participants performed significantly better in phases of the task in which social information was stable as compared to when social information was volatile (t(85) = 11.197, p <.001, d = 1.21, 95% CI [.93, 1.48]). B Participants took advice into account more often in phases of the task in which social information was stable compared with when social information was volatile (t(85) = 6.062, p <.001, d =.654, 95% CI [.42,.89]). C Participants wagered significantly more points during stable phases of the task (F(1,85) = 98.409, p <.001, η_p² = .537, 95% CI [.39,.64]), regardless of the information source

Fig. 5

Model internal validity. The predicted wager was highly correlated with the number of points participants actually wagered across all four phases of the experiment

Fig. 6

a Learning model parameters. Results showed that there was a significant difference between kappa ($\kappa$) for social information and kappa ($\kappa$) for nonsocial information. This was not the case for theta. b Distribution of values for response model parameter zeta ($\zeta$)

Fig. 7

Mean theta power (4–8 Hz). a Topoplot of mean theta power across 100–400 ms after trial outcome representation (0 ms). b Time–frequency plot of mean theta power (4–8 Hz) over the midfrontal electrodes (Fz, FCz, Cz) across 0–1 s after trial outcome representation. c Mean theta power over the midfrontal electrodes for correct and incorrect trials. d Topographic representation of the TFR of the difference in theta power between incorrect and correct trials