Individual differences through self-reinforcement of suboptimal strategies

项目摘要/摘要 是什么导致了个体行为的差异？这个根本性的问题通常被赋予两个 答案：先天和后天。在这里，我们建议这两个答案，虽然都是关键的和正确的，但 不足以完全解释个体的差异。相反，我们认为行为上的巨大差异 在一定程度上，不同的个体之间形成了不同的奖励关联 同样的环境。这是因为世界是复杂的和高维的，因为有 几乎总是有多个可能的行为或事件可以归因于奖励。鉴于...的关键作用 多巴胺神经元作为大脑的正反馈系统进行行为控制，具体假设是 不同个体在初始状态下的微小差异最终会在哪些特征上产生巨大差异 个体归因于奖励的环境。这一假设部分是受到复杂系统的启发。 理论，该理论强调正反馈在产生和放大微小差异、创造 结果似乎是随机的。为了解决这一假设，我们将利用我们最近的发现 多巴胺神经元计算不同环境维度的奖励预测误差。 具体地说，我们将使用多巴胺神经元记录来推断环境的时变特征 每只动物用来预测奖励，然后基于以下条件构建每个个体的强化学习模型 这些功能。最终，这个可测试的框架旨在解释个人之间的正常差异，如 以及多巴胺在调节一系列不同的神经精神疾病方面无处不在的贡献。

Project Summary/Abstract What produces individual differences in behavior? This fundamental question has classically been given two answers: nature and nurture. Here, we suggest that those two answers, while both critical and correct, are insufficient to fully explain individual variability. Instead, we propose that the vast differences in behavior between individuals arise in part from different individuals forming different reward associations within the same environment. This results from the fact that the world is complex and high-dimensional, in that there are almost always multiple possible actions or events that could be attributed to reward. Given the key role of dopamine neurons as the brain’s positive feedback system for behavioral control, the specific hypothesis is that small differences across individuals in initial conditions ultimately produce large differences in which features of the environment that the individual attributes to reward. This hypothesis is inspired in part by complex systems theory, which emphasizes the role of positive feedback in generating and amplifying small differences, creating outcomes that seem stochastic. To address this hypothesis, we will leverage our recent finding that different dopamine neurons calculate reward prediction error across different dimensions of the environment. Specifically, we will use dopamine neuron recordings to infer the time-varying features of the environment that each animal uses to predict reward, and then build reinforcement learning models of each individual based on these features. Ultimately, this testable framework aims to explain both normal variation across individuals, as well as the ubiquitous contribution of dopamine in mediating a disparate range of neuropsychiatric diseases.