Tomoya Nakai
Cognitive neuroscience aims to link psychological constructs to brain activity, but this goal depends critically on how such constructs are operationalized. Conventional controlled paradigms define constructs through task contrasts or parametric manipulations, providing interpretability by restricting the range of stimuli, behavioral responses, and task demands. Naturalistic paradigms instead use complex, temporally extended stimuli such as movies, narratives, and music, enabling brain activity to be modeled in relation to continuous, multidimensional features of real-world inputs. With the rise of intersubject correlation analyses, encoding models, and NeuroAI approaches, naturalistic neuroimaging has expanded from the characterization of shared brain activity across participants to high-dimensional predictive modeling using features derived from stimuli, behavior, and artificial neural networks. However, this shift does not eliminate theoretical assumptions; it relocates them from task design to feature-space selection, model architecture, and model evaluation. This review argues that controlled and naturalistic paradigms should be treated as complementary strategies rather than competing alternatives. Integrating hypothesis-guided task design with computational modeling may extend neuroimaging beyond passive perception toward active cognition, including reasoning, decision making, and mathematical problem solving. Such an integrative framework can provide a broader basis for testing, refining, and potentially redefining psychological constructs through their relationships to neural and computational representations.