Habits of thought generate candidate actions for choice

Abstract

Effective decision-making in high-dimensional choice spaces requires a computationally efficient mechanism to restrict consideration to a small set of viable candidates. This “pre-planning” phase relies on a hybrid cognitive architecture where context-free habits of thought generate a subset of actions based on their general historical value. Deliberative, model-based planning is then selectively applied to this reduced set to evaluate candidates according to context-specific features. Simulations demonstrate that this approach optimizes the trade-off between the computational costs of planning and the accuracy of outcomes. Empirical evidence from eight experiments, including food choice and novel associative learning tasks, confirms that generalized value estimates primarily determine which options enter the consideration set, whereas specific value estimates guide the final selection. Notably, high-value habitual candidates persist in coming to mind even when they are contextually inappropriate or explicitly counter-indicated by the current task. This integration suggests that habits are not merely alternatives to deliberation but are essential precursors that render complex planning tractable by providing the raw materials for choice. – AI-generated abstract.