Converging evidence suggests that reinforcement learning (RL) signals exist within the human brain and that they play a role in the modification of behaviour. According to RL theory, prediction errors are used to update values associated with actions and/or predictive cues, thus facilitate decision-making. For example, the reward positivity-a feedback-sensitive component of the event-related brain potential (ERP)-is thought to index an RL prediction error. An unresolved question, however, is whether or not action is required to elicit a reward positivity. Reinforcement learning theory would predict that the reward positivity should diminish or disappear in the absence of action, but evidence for this claim is conflicting. To investigate the impact of cue, choice, and action on the amplitude of the reward positivity, we altered a two-armed bandit task by systematically removing these factors. The reward positivity was greatly reduced or absent in the altered versions of the task. This result highlights the key role of agency in producing learning signals, such as the reward positivity.

中文翻译：

---

代理在人类奖励处理中的重要性。

越来越多的证据表明，强化学习（RL）信号存在于人脑中，并且在行为改变中起作用。根据RL理论，预测错误用于更新与动作和/或预测线索相关的值，从而有助于决策。例如，奖励积极性-事件相关脑电势（ERP）的反馈敏感组件-被认为可以索引RL预测误差。然而，一个尚未解决的问题是是否需要采取行动才能激发奖励的积极性。强化学习理论可以预测，如果不采取任何行动，则奖励积极性应该减少或消失，但是这种说法的证据是矛盾的。为了研究提示，选择和动作对奖励积极性幅度的影响，我们通过系统地消除这些因素来改变了两臂匪徒的任务。在任务的变更版本中，奖励积极性大大降低或缺失。该结果突出了代理在产生学习信号（例如报酬积极性）中的关键作用。