The neural bases of affective experience remain elusive. Early neuroscience models of affect searched for specific brain regions that uniquely carried out the computations that underlie dimensions of valence and arousal. However, a growing body of work has failed to identify these circuits. Research turned to multivariate analyses, but these strategies, too, have made limited progress. Predictive processing models offer exciting new directions to address this problem. Here, we use predictive processing models as a lens to critique prevailing functional neuroimaging research practices in affective neuroscience. Our review highlights how much work relies on rigid assumptions that are inconsistent with a predictive processing approach. We outline the central aspects of a predictive processing model and draw out their implications for research in affective and cognitive neuroscience. Predictive models motivate a reformulation of “reverse inference” in cognitive neuroscience, and placing a greater emphasis on external validity in experimental design.

情感体验的神经基础仍然难以捉摸。早期的情感神经科学模型寻找特定的大脑区域，这些区域独特地执行了作为效价和唤醒维度基础的计算。然而，越来越多的工作未能识别这些电路。研究转向多变量分析，但这些策略也取得了有限的进展。预测处理模型为解决这个问题提供了令人兴奋的新方向。在这里，我们使用预测处理模型作为镜头来批判情感神经科学中流行的功能神经影像学研究实践。我们的评论强调了有多少工作依赖于与预测处理方法不一致的严格假设。我们概述了预测处理模型的核心方面，并提出了它们对情感和认知神经科学研究的影响。预测模型激发了认知神经科学中“反向推理”的重新表述，并更加强调实验设计中的外部有效性。