Impaired Reinforcement Learning Underlying Explore-Exploit Decision Making in Theft Recidivists

A recent study explored the behavioral and neurobiological underpinnings of recurrent theft among theft recidivists. Researchers utilized a 4-arm bandit task, while simultaneously monitoring prefrontal cortex hemodynamics via functional near-infrared spectroscopy (fNIRS). The findings suggest a significant impairment in reinforcement learning mechanisms in non-kleptomanic theft recidivists compared to control participants and kleptomanic offenders.

Non-kleptomanic theft recidivists (TR-K) accumulated substantially higher cumulative regret and made fewer optimal choices than both control individuals without criminal records (CT) and kleptomanic offenders (TR+K). The model-based analysis revealed that Q-learning with decay model best fit the observed data. The parameter extraction demonstrated a lower learning rate in the TR-K group compared to the CT and TR+K groups, indicating a deficit in updating action values following environmental feedback.

The fNIRS tracking of trial-by-trial latent reinforcement variables showed that PFC activity was modulated by these variables. However, the group differences were characterized by static baseline hemodynamic shifts rather than rewirings of value-tracking neural circuits. In conclusion, these results suggest that impaired reinforcement learning mechanisms in non-kleptomanic theft recidivists challenge the effectiveness of current punitive deterrence models.