身体活动的双系统理论:一种强化学习的视角

以理性决策为基础的锻炼行为理论被认为是理解身体活动的主导体系, 它提供了与身体活动相关的认知构念作为有价值的信息。基于社会生态模型设计的行为干预措施, 因表现出了更好的效果而备受关注。近期研究表明, 积极的运动认知和当前体育环境都没能很好地促进个人锻炼习惯的养成, 因此有必要探索新的理论体系来阐明个人锻炼习惯的形成机制。解释身体活动的最新体系是双系统理论, 由于其考虑了身体活动的无意识和快乐决定因素, 有望提供一个更广泛的动机视角。一方面, 多个有代表性的身体活动双系统模型, 从简单的自发路径, 到情境线索与锻炼习惯, 再到突出自动情感评价作用的复杂概念模型, 阐明了系统1的构建, 结合锻炼行为理论所关注的系统2, 为模型的构建提供了依据。另一方面, 通过对双系统的竞争、协调和层级控制原则的分析, 为模型的控制提供了建议。经典的强化学习框架解释了双系统模型的构建与控制原则:在模型的构建方面, 无模型与基于模型的强化学习分别表示系统1和系统2。在模型的控制方面, Dyna协作架构与分层强化学习, 为身体活动可能是一种相互协作、分层执行的复杂行动组合提供了合理解释。最后提出强化学习视角下锻炼者-体育环境的互动模式, 试图从一个全新的角度探讨锻炼行为。

Dual system theory of physical activity: A reinforcement learning perspective

Exercise behavior theory based on rational decision making is regarded as the dominant framework for understanding physical activity, which provides valuable information on cognitive constructs linked to physical activity. Behavioral interventions based on social ecological model have attracted much attention due to its better performance. However, recent studies show that exercise benefits cognition and current sport environment do not promote the formation of individual exercise habit, so it is necessary to explore a new theoretical framework to clarify the formation mechanism of individual exercise habit. The latest framework for explaining physical activity is dual system model, which promises to provide a broader perspective on motivation by considering the non-conscious and hedonic determinants of physical activity. Multiple representative dual system models associated with physical activity, on the one hand, from simple spontaneous path to context cues and exercise habit, and then to complex concept model highlighting the role of automatic affective evaluations, the construct of system 1 is elaborated. Combined with system 2, which is concerned by exercise behavior theory, they offer a basis on how to build dual system model for physical activity. On the other hand, through analysis of the principle for competitive, cooperative and hierarchical control of dual systems, proposals are provided for the control of dual system model. Canonical reinforcement learning framework explains the principle for construct and control of dual system model: in terms of construct of the model, model-free and model-based reinforcement learning represent system 1 and system 2, respectively. In terms of control of the model, Dyna cooperative architecture and hierarchical reinforcement learning provide a reasonable explanation for physical activity is more likely to be a complex combination of actions which is hierarchically operated with a cooperative form than competitive control. Finally, the exerciser-sport environment interaction in reinforcement learning is put forward to explore exercise behavior from a brand-new perspective.