Intelligence, traditionally associated with the ability to perceive, retain, and utilize information for environmental adaptation, extends to systems of living cells. This research introduces a novel framework suggesting that non-equilibrium tuning and compartmentalization are fundamental principles sufficient to model cellular intelligence. By employing the mathematical language of operads, the study encapsulates complex cellular behaviors—such as specialization, division, fusion, and intercellular communication—within a unified theoretical model. Operads enable the representation of nested and hierarchical systems, offering a cohesive understanding across all scales of life, from subcellular organelles to multicellular organisms.

To operationalize this model, the researchers developed IntCyt, an unsupervised learning algorithm that embodies the principles of non-equilibrium compartmentalization and operadic structures. IntCyt was applied to benchmark machine learning datasets, engaging in generative and self-supervised tasks. The algorithm showcased exceptional capabilities in memorizing, organizing, and abstracting data, surpassing a wide array of existing machine learning algorithms in terms of interpretability, plasticity, and accuracy.

This innovative approach provides a biologically inspired computational paradigm that mirrors the emergent properties inherent in living systems. By harnessing the adaptive strategies of cellular composites, IntCyt offers a practical and lightweight tool for advanced data mining and analysis. The implications of this work are far-reaching, potentially revolutionizing our understanding of life processes and paving the way for new methodologies in AI systems.

This framework aligns seamlessly with Daice Labs' mission to innovate through the computing of natural intelligence principles. We are building upon non-equilibrium tuning and compartmentalization to develop advanced computational models that emulate composite cellular behavior. By leveraging the operadic approach to model nested and hierarchical multi-agentic systems, we are extending the capabilities of algorithms like IntCyt to enhance machine learning applications. Our work focuses on creating adaptable, and efficient computational tools that mirror the emergent properties of living cells, thereby advancing our understanding of collective intelligence.