Opponent learning with different representations in the cortico-basal ganglia pathways can develop obsession-compulsion cycle
Fig 2
Environmental model describing possible development of obsession-compulsion cycle, adopted from the previous study [10].
There are two states: the relief state and the obsession state. At the relief state, the agent can take the "abnormal reaction" (to an intrusive thought, i.e., spontaneously arising anxiety e.g., about door lock), which induces a transition to the obsession state, or the "other" action (e.g., just ignore or forget about the intrusive thought), with which the agent stays at the relief state. At the obsession state, the agent can take the "compulsion" action (e.g., confirms door lock), which requires a small cost (0.01) but induces a transition back to the relief state with a high probability (50%), or the "other" action, which requires no cost but induces a transition back to the relief state only with a small probability (10%). Every stay at the obsession state imposes punishment (negative reward −1).