Attention regulation depends on how neural systems balance internal control with sensory input. When this balance becomes rigid, regulation requires greater effort and becomes unstable. Empirical work shows that a brief low-effort training paradigm, the integrative body–mind training (IBMT) improves attention by inducing a low-effort regulatory state [1], yet individuals vary in their ability to access and stabilize this state. Existing accounts do not explain how such low-effort regulation is learned and generalized at the level of control dynamics. We propose a theoretical framework in which embodied AI supports this learning by stabilizing a measurable target regulatory state via adaptive multisensory input.
Methods
We model attention regulation using predictive coding and the free energy principle [2], treating control as precision-weighted inference. Effortful states arise from excessive precision on self-related control priors, increasing energetic cost. Low-effort states correspond to reduced control precision and efficient integration of interoceptive and exteroceptive signals. We introduce an AI-based multisensory support architecture operating as a closed-loop system layered onto IBMT. The AI estimates regulatory state from physiological and attentional markers and adaptively modulates basic sensory parameters (timing, intensity, variability) to bias precision allocation without explicit instruction or reinforcement.
Results
The framework predicts that adaptive multisensory support facilitates access to low-effort regulatory states in individuals with high attentional noise or rigid control. Predicted outcomes include reduced energetic cost, reflected in stabilized autonomic markers and decreased indices of cognitive effort. With repeated practice, the model predicts internalization of the target state, leading to reduced dependence on external modulation. The framework also predicts failure regimes, including destabilization under excessive modulation and limited benefit when control precision is already low, highlighting the need for individualized adaptation.
Discussion
By treating low-effort attention as a configuration of regulatory dynamics rather than a subjective state, this framework bridges contemplative neuroscience and embodied active inference. Individual differences in attention training are reframed as differences in precision dynamics and learning trajectories. The model generates testable predictions that adaptive sensory environments accelerate stabilization and effortless re-entry into low-effort regulation, and suggests general design principles for embodied agents that support regulation by shaping sensory context rather than increasing control intensity.
References
1. Tang, Y.Y., Tang, R, Posner, M. I., & Gross, J. J. (2022). Effortless training of attention and self-control: mechanisms and applications. Trends Cogn Sci. 26(7), 567-577. https://doi.org/10.1016/j.tics.2022.04.006 2. Friston, K. (2010). The free-energy principle: a unified brain theory?. Nat Rev Neurosci 11, 127–138. https://doi.org/10.1038/nrn2787
Acknowledgement
This work is supported by the ONR N000142412270 and NIH R33 AT010138.
