MAIS: an in-vitro sandbox enables adaptive neuromodulation via scalable neural interfaces

Abstract Brain-machine interfaces (BMIs) predominantly rely on static digital architectures to decode biological neuronal networks, a paradigm that is incompatible with natural neural coding in the human brain 1–4 . Bridging this gap is a critical step in combating neuronal dysfunction, enhancing brain functionality, and refining the precision of neuroprosthetics 5 . The integration of brain organoids with microelectrode array (MEA), as a class of BMIs, offers a humanized in vitro platform with unique biological compatibility advantages for dynamic neuronal decoding. This study resolves the biological-electronic encoding incompatibility of brain organoid-MEA Integration through three progressive breakthroughs. First, a human-machine hybrid agent is developed as a newly proposed bioengineered platform that couples brain organoids together with high-density MEAs and computational chips, enabling closed-loop perturbation of biological neuronal networks via exogenous signals. Second, through plasticity-driven real-time tracking of neuronal activity, we establish dynamically reconfigurable stimulation nodes that self-align with the electrophysiological states of the organoids. This resolves the exogenous-endogenous encoding mismatch by implementing plasticity-driven adaptation principles that ensure biological compatibility through spatially adaptive coordination. Finally, through shared plasticity rules rather than centralized control, we construct the first scalable multi-agent interactive system (MAIS) and demonstrate its real-world applications. Through designed scenarios of pathological/normal neuronal network interaction, we validate that MAIS achieves stable cross-network coordination. MAIS embodies a self-evolving neural coding sandbox in which plasticity-driven dynamic decoding bridges the compatibility gaps between biological and digital systems, providing a scalable and foundational infrastructure for human-centered neural interfaces.