Gambling disorder is a behavioral addiction that is linked to a range of financial, social and mental health problems, and can result in substantial costs for individuals and society. From a neurocognitive perspective, disordered gambling is linked to reduced behavioral flexibility and impairments in goal-directed control. In a first study (n=30 gambling group, n=30 control group), we show that individuals suffering from problem gambling symptoms show reduced performance in a two-step sequential decision-making task. Computational modeling revealed this this was due to impairments in goal-directed (“model-based”) control, as well as changes in perseveration (“habitual” control). In a second online study (n=491 individuals with gambling experience), we then tested the hypothesis that gambling would also be linked to rigid and inflexible beliefs. Endorsement of irrational beliefs increased with increasing problem gambling severity for delusional, superstituous and conspiratorial ideation, as well as for gambling-related cognitive distortions. Likewise, a previously identified transdiagnostic symptom domain (“compulsive behavior and intrusive thought”, CIT) was positively correlated with problem gambling severity. The effect of CIT on problem gambling severity was partially mediated by gambling-related cognitive distortions. Together, results reveal that problem gambling severity is linked to both behavioral (study 1) and cognitive inflexibility (study 2), as reflected in the increased endorsement of irrational beliefs across domains. Results are compatible with an etiological model in which a high compulsivity phenotype may predispose individuals to develop maladaptive gambling-related cognitive distortions during gambling exposure.

Habits are integral to daily life, yet their development remains poorly understood. We studied the role of behavioral repetition in habit formation. Using a computational model in an instrumental-learning binary-choice task, we investigated whether choice frequency affects behavior in addition to and independently from reinforcement. We developed a novel behavioral task based on computational models that operationalize habits by the frequency of past behavior and trained 220 participants on each of five consecutive days in the lab. Linear mixed-effects models and computational modeling of choice behavior revealed that past choice frequency influenced current choice in addition to and independently of reinforcement. Hierarchical Drift Diffusion Modeling of response times confirmed that a model mapping choice frequency onto the drift rate provided a superior fit to the data compared to models mapping it to response bias or not including choice frequency as a source of variation. In a subsequent pre-registered fMRI study, 60 participants completed a modified version of the Reward Pairs task. Computational modeling of task behavior replicated previous findings. At the neural level, BOLD responses in the ventral striatum correlated with reinforcement values and those in the dorsolateral striatum with choice frequency. Furthermore, an ongoing study examines task behavior of participants with regular cocaine use and preliminary results of the comparison with non-users will be presented. This pattern of results suggests that choice frequency influences subsequent behavior. Furthermore, choice-frequency mechanisms may work side-by-side with reinforcement learning and rely on partially distinct neural circuits that jointly contribute to decision making.

Habits are automatic behaviors that facilitate efficient performance but can also lead to behavioral inflexibility. Neuroimaging studies suggest that the posterior putamen is involved in human habit expression, but findings have not been consistently replicated. Since many habitual behaviors rely on motor activity, motor learning may provide a useful framework for reliably investigating human habit formation. We therefore developed an fMRI-adapted motor sequence learning task (MSLT) to investigate the neural mechanisms underlying motor automatization. Twenty-nine participants (16 female, mean age (SD) = 24.4 (6.4) years) completed the MSLT across two days. In learning blocks, participants implicitly learned a fixed sequence of visual stimuli, while goal-directed behavior was examined in test blocks by omitting single items of the sequence at random positions. During these skip-events, we measured whether participants executed the omitted response despite its absence (“action slip”), interpreted as behavioral inflexibility. With prolonged training, reaction times within learning blocks decreased, while action slips within test blocks increased, indicating a shift towards automatic and inflexible responding. fMRI analyses of prolonged training revealed reduced activity in regions involved in sensory and motor processing (p_FWE <.05). In contrast, activation in the bilateral posterior putamen increased with prolonged training (SVC, p_FWE <.05). Action slips in test blocks were associated with greater putamen activation than correct skip-events (SVC, p_FWE < .05), suggesting that inflexible responding was related to activity in this region. These findings suggest the MSLT as a promising paradigm for investigating the development of motor automaticity related to human habit formation.

Evidence for an altered balance between habitual (model-free, MF) and goal-directed (model-based, MB) control in individuals with alcohol use disorder remains mixed (Doñamayor et al., 2022). This may be due to how human studies often assess habits: via paradigms that take place in disorder-irrelevant contexts and that focus narrowly on MF-MB distinctions, overlooking alternative strategies.

In a preregistered (doi.org/10.17605/OSF.IO/9TUZE) online study in n=520 alcohol drinkers (preprint: doi.org/10.17605/OSF.IO/9TUZE), we examined whether high-risk users (n=260) rely more on successor representation (SR) strategies (Dayan, 1993), or their goal-based dimension-reduced variants (redSR; Shimomura et al., 2021), than low-risk users (n=260). Such SR-based strategies have been proposed as a more subtle proxy of habitual behavior (Momennejad et al., 2017). We also tested whether group differences in SR-based strategy use are amplified in virtual alcohol-related contexts.

While our results support the idea that SR strategies contribute to human learning and decision-making (Momennejad et al., 2017), high- and low-risk drinkers did not differ in their reliance on SR or redSR strategies. Unexpectedly, both high- and low-risk users showed lower reliance on (dimension-reduced) SR-based strategies and increased reliance on a MF strategy in alcohol-related contexts.

The reasons for this context effect remain to be further investigated, but one possible explanation is that cognitive map-based strategies have previously proven unreliable under alcohol influence, leading even low-risk drinkers to default to simpler MF control in alcohol-related environments. More broadly, these findings underscore the importance of context and the need to assess a wider range of decision-making strategies in addiction research.

Addiction is often characterized by maladaptive behavior: individuals repeatedly engage in behavior despite knowing that these are not in their best interest. Such behavior is commonly attributed to a shift from goal-directed to stimulus-driven, habitual processes. In this talk, we propose an alternative account. Building on Moors’ goal-directed theory, we explain maladaptive behavior in addiction as the product of biased goal-directed processes in which addictive behavior is selected because it is expected to best reduce discrepancies between current and desired states. We identify five factors that may bias this process and thereby contribute to maladaptive behavior in addiction: the relative value of addiction-related goals, the frequency and magnitude of perceived discrepancies, preferences for discrepancy-reduction strategies, the availability of behavioral alternatives, and rigid expectancies about the outcomes of addictive behavior. We argue that this framework can accommodate variability, risk factors, and intervention outcomes, and offers a promising path towards a novel understanding of maladaptive behavior in addiction.