Interoceptive Attention as Dynamic Homeostatic Prioritization in a Foraging Agent

A new preprint on arXiv (2608.04232) presents a computational model where a foraging agent dynamically allocates its interoceptive precision to prioritize the most-needed bodily channel, improving survival in a four-channel gridworld called AffectWorld. The study, using active inference, shows that selective allocation more than doubles learning-phase survival compared to uniform precision (0.414 vs 0.199 across 11 layouts, n=32 seeds each, paired cluster-bootstrap p≤10⁻⁴). The mechanism involves reading body-state beliefs, identifying the most-needed channel, and reallocating a fixed budget of interoceptive precision toward it. The benefit runs through the precision-shaped likelihood affecting both belief update and planning. This research addresses how biological systems regulate competing needs under limited perceptual bandwidth, suggesting that dynamic prioritization is key to survival. The findings have implications for understanding homeostatic regulation and could inform AI systems designed to manage multiple objectives. The paper is authored by researchers (names not specified in the abstract) and was announced as new on arXiv. The study's results are statistically significant and highlight the importance of selective attention in interoceptive processing.