# Kyoto University: non-kleptomanic theft recidivists show lower learning rate in 4-arm bandit task

- Published: 2026-11-11 10:00 UTC
- Section: Life & Health › Neurotechnology
- Tags: theft recidivists, reinforcement learning deficit, 4-arm bandit task, explore-exploit decision making, fNIRS prefrontal cortex, kleptomania, Q-learning with decay, Kyoto University study, Frontiers in Psychology
- URL: https://realtech.media/news/rt-81cfc8592b053d7072ff3d54_kyoto_university_non/

**In a study with fNIRS monitoring of the prefrontal cortex, repeat thieves without kleptomania updated action values more slowly after feedback than controls.**

![Kyoto University: non-kleptomanic theft recidivists show lower learning rate in 4-arm bandit task](https://realtech.media/media/2026/11/dca28c9bf2f6b618.jpg)
*Walej, Wikimedia Commons, CC BY-SA 4.0*

Researchers at Kyoto University reported that theft recidivists without kleptomania have a lower reinforcement learning rate than people with no criminal record. The paper was published on 9 November 2026 in Frontiers in Psychology. It compared three groups: recidivists with a kleptomania diagnosis (TR+K), recidivists without it (TR-K), and controls with no criminal record (CT) [1]. Prefrontal cortex activity was recorded throughout the task with functional near-infrared spectroscopy (fNIRS) [1].

According to the university's release, theft recidivism reaches about 70 percent within five years. Recidivists fall into two subgroups: kleptomania, driven by uncontrollable urges to take items of little monetary value, and instrumental theft, which is profit-seeking. Why instrumental offenders reoffend despite severe punishment has remained unclear.

## Task and model results

Participants performed a 4-arm bandit task. They repeatedly chose among four options with fluctuating reward probabilities, trying to pick the highest-paying one each time. The task requires balancing exploration of unknown options against exploitation of options with known payoffs.

The TR-K group, but not the TR+K group, accumulated significantly higher cumulative regret and made fewer optimal choices than controls [1]. Model comparison with Variational Bayesian Analysis identified Q-learning with decay as the best fit. In this model, the TR-K group had a lower learning rate than both the CT and TR+K groups. The authors interpret this as a deficit in updating action values after environmental feedback [1].

## Prefrontal activity

The team also tracked trial-by-trial latent reinforcement variables in the fNIRS signal. Prefrontal activity was modulated by these variables in all groups. Group differences appeared as static shifts in baseline hemodynamics, not as changes in the circuits that track value [1].

The authors write that recurrent non-kleptomanic theft may be associated with impaired reinforcement learning, and that this challenges current punitive deterrence models [1]. Corresponding author Yukiori Goto said that instrumental theft was assumed to involve normal or even enhanced reinforcement learning, but the findings show that recurrent instrumental theft involves neurobehavioral deficits. The authors conclude that justice systems may need rehabilitation strategies that target reinforcement learning deficits [1].

## Sources

1. [Frontiers in Psychology](https://doi.org/10.3389/fpsyg.2026.1997570)
2. [EurekAlert](https://www.eurekalert.org/news-releases/1151185)

---

Source: RealTech, realtech.media — https://realtech.media/news/rt-81cfc8592b053d7072ff3d54_kyoto_university_non/