Bio-CPC: Collective Predictive Computation in Biological Organization

Daichi G. Suzuki , Shigeto Dobata, Koryu Kin, Reina Kobayashi, Yusuke Hayashi

Biological organization is characterized by hierarchically nested levels in which relatively autonomous units become integrated into higher-order systems. Major evolutionary transitions such as multicellularity and eusociality therefore raise a common problem: how can locally informed, goal-directed units coordinate heterogeneous responses so that they collectively support a larger whole? Here, building on collective predictive coding (CPC), originally developed to explain the emergence of shared symbolic representations through decentralized Bayesian inference, we propose Bio-CPC, a broader framework of collective predictive computation for biological organization. In this framework, information originating in multiple constituent units can be integrated into a collectively informed latent variable that carries information about states relevant to the higher-order system. Unit-specific mappings translate this collective variable into potentially heterogeneous but mutually compatible responses whose joint effects regulate the collectively relevant state toward a preferred or reference state. Bio-CPC thus extends the original CPC architecture by coupling collective inference to an additional biological alignment problem: the establishment, calibration, or maintenance of compatible mappings from collectively informed states to unit-specific responses. Importantly, Bio-CPC is not restricted to decentralized architectures. The same computational organization can be realized along a continuum from predominantly local inference among constituent units to increasingly centralized integration by specialized subsystems. Such centralization may strengthen functional integration within multicellular organisms, as illustrated by the evolution of nervous systems in some animal lineages. The framework also has an evolutionary implication: as higher-order regulation becomes increasingly dependent on information that is available only through multi-unit integration, natural selection should favor communication and integration architectures that improve how such information is transmitted and combined across units. Developmental interactions may coordinate inherited response capacities and communication mechanisms, with recurrently effective mappings subsequently stabilized through genetic accommodation. Bio-CPC is specifically supported when information originating in multiple units contributes causally to a higher-order state estimate and when compatible response mappings are established, calibrated, or maintained. Bio-CPC predicts distinguishable perturbation effects where these two processes can be selectively targeted; conflict-suppression mechanisms help maintain response-mapping alignment. This provides a testable information-processing framework for investigating both the emergence and progressive integration of higher-order biological organization.