Figure 1. Distinct mechanisms by which motivational valence may bias behavioural activation.
(A) Pavlovian response bias: appetitive cues (green edge) elicit generalized behavioural activation (‘Go’), whereas aversive cues (red edge) elicit behavioural inhibition (‘NoGo’). This Pavlovian response bias is introduced in model M3a as the parameter π (c.f. Figure 3). (B) Instrumental learning bias: rewarding outcomes (upper panel) facilitate learning of action (‘Go’, thick arrow) relative to inaction (‘NoGo’, thin arrow). Thus, learning effects at the individual trials t will result in a cumulative selective increase of the rewarded action on later trials tn. Punishment outcomes (lower panel) hamper the unlearning of inaction (‘NoGo’, dashed arrow) relative to action (‘Go’, solid arrow), resulting in sustained inaction. Neutral outcomes are equally well associated with actions and inactions, and are not illustrated here. The instrumental learning bias is introduced as the parameter κ in model M3b (c.f. Figure 3). We assess whether these two mechanisms (i) act in parallel, and (ii) are modulated by the catecholamine system. To test the latter, we administered methylphenidate (MPH), which prolongs the effects of catecholamine release via blockade of the catecholamine receptors. We first assess whether MPH affects the strength of the Pavlovian response bias, introduced as the parameter πMPH in model M5a, and instrumental learning bias, implemented as the parameter κMPH-selective in model M5b (c.f. Figure 5). (C) We hypothesise that prolonged effects of dopamine release following reward outcomes will reduce (temporal) specificity, leading to spread of credit: Credit is assigned to other recent actions (thin arrow), in addition to the performed (and rewarded) Go response (thick arrow), resulting in additional learning of the alternative (not-performed) Go response. This MPH-induced diffuse learning bias is implemented by the parameter κMPH-diffuse in model M5c (c.f. Figure 5).