The Embodied Octopus: Distributed Intelligence and Active Inference in a Flexible Body

Octopuses exhibit a striking form of embodied intelligence: a nervous system in which over two-thirds of neurons reside not in a central brain, but in their highly flexible arms [Hochner, 2012, 2023]. This distributed architecture enables each arm to perform local sens- ing, prediction, and motor control, while the central brain coordinates global behaviours such as hunting, navigation, and camouflage [Hanlon and Messenger, 1996]. We argue that this organisation is best understood through the lens of hierarchical Active Inference, in which local generative models operate at the periphery of the body while higher-level priors govern goal-directed policy selection [Friston, 2010, Buckley et al., 2017]. Unlike vertebrate organ- isms, the octopus must control thousands of degrees of freedom in a soft-bodied morphology, making purely centralised control computationally intractable [Flash and Zullo, 2023, Kier and Smith, 1985]. Instead, its nervous system exhibits a multi-scale division of labour be- tween local and global inference, offering a powerful biological blueprint for distributed, sensorimotor AI [Hale, 2025, Carls-Diamante, 2022]. In this perspective, we explore how octopus cognition provides a living model of decentralised, hierarchical predictive processing and discuss implications for artificial systems that seek to embody similar principles [Costa et al., 2022, Mazzaglia et al., 2022].