, 2011). This raises the possibility that model-free and model-based systems are not segregated systems whose influence is weighted at the time of choice. Instead,
choices could also be made by a model-free system in which learning is modulated by transition probabilities. In this study, we cannot unambiguously differentiate between these accounts and further fine-grained investigations, in part motivated by the present data, are required to understand this complex issue. Dopamine itself is a precursor to norepinephrine and epinephrine, potentially contributing to the observed effects. However, L-DOPA administration causes a linear increase in dopamine levels in the brain without affecting norepinephrine levels Ribociclib cost (Everett and Borcherding, 1970). Another possibility would be that L-DOPA exerts effects through interactions with the serotonin system. Such an interaction, between dopamine and serotonin, is known to play a role in a range of higher-level cognitive functions (Boureau and Dayan, 2011). By implicating dopamine in behavioral control, we open the door to further experiments aimed at elucidating the precise neural mechanisms underlying the arbitration between both controllers. While
theoretical considerations afford a number of ways for how this arbitration might be implemented in the brain (Daw, 2011; Keramati et al., 2011), our results find more provide empirical evidence that dopamine influences the relative degree between model-free and model-based control. Eighteen healthy males (mean age: 23.3 [SD: 3.4]) participated in two separate sessions. Data from two additional subjects were not included in the analysis as those subjects misunderstood instructions and performed at chance level. The UCL Ethics committee approved the study and subjects gave written informed
consent before both sessions. Subjects were tested in a double-blind, fully counterbalanced, repeated-measures setting on L-DOPA (150 mg L-3,4-dihydroxyphenylalanine / 37.5 mg benserazide; Madopar, Roche) and on placebo (500 mg calcium carbonate; Calcit, Procter and Gamble) dispersed of in orange squash. The task was administered 55.0 (SD: 4.7) min after drug administration. Sessions one and two were approximately 1 week apart (at least 4, but no more than 14 days), with both sessions at the same time of day. All subjects except one participated in the morning to minimize time-of-day effects. We assessed drug effects on self-reported mental state using a computerized visual analog scale immediately before starting the task (Bond et al., 1974). We drew on Daw et al. (2011)’s two-step choice task to assess the relative degree of model-based versus model-free decision making. Our version of the task was identical to Daw et al.’s except for different stimulus images (semantically irrelevant fractals), a slightly larger dynamic range of reward probabilities, and more rapid trial timings.