Attentional Dynamics Reveal Cognitive Motivation and Effort Allocation

How people decide whether to invest cognitive effort in demanding tasks or shift to easier alternatives is central to understanding motivation and cognitive control, yet the latent mechanisms governing these moment-to-moment decisions remain insufficiently understood. Here, we combine a novel parallel-task reinforcement learning and working memory (RLWM) paradigm with a hierarchical hidden Markov model (HMM) to track trial-level cognitive states as participants freely and dynamically allocate attention between simultaneous low-demand and high-demand learning tasks. We show that our model adequately identifies behavior as reflecting one of three latent states: engaged in the low-demand task, engaged in the high-demand task, or disengaged. Across three independent samples (N=269), including a preregistered replication, we demonstrate that: (i) subjective effort predicts model-estimated disengagement, linking perceived cognitive load to attentional lapses; (ii) intrinsic motivation for cognitive challenge (Need for Cognition) predicts voluntary selection of the demanding task; and (iii) switches into higher demand are accompanied by increased policy entropy, revealing that effort investment reflects exploratory information-seeking. Our framework unifies perspectives from cognitive control, working memory, and curiosity-driven exploration, demonstrating that trial-to-trial fluctuations in attention provide a mechanistic window into how subjective effort, intrinsic motivation, and information-seeking jointly shape the dynamic allocation of cognitive resources.