New BHI framework distributes computation between biological tissue and AI

In a new arXiv preprint (arXiv:2608.18748), researchers Michael Taynnan Barros, Sergio Lopez Bernal, and Reinhold Scherer introduce Biological-Hybrid Intelligence (BHI), a conceptual framework for distributing computation across biological and artificial substrates. Existing biohybrid systems optimize the biological component, the AI model, or their interface separately, but never address how computation should be allocated, reassigned, or evaluated. BHI closes that gap by treating both the living tissue and the AI as computational entities coupled through a bioelectronic interface and coordinated by an orchestrator. Crucially, BHI requires reciprocal co-adaptation: unlike systems that merely decode neural signals or stimulate a substrate, this framework allows computational responsibilities to shift dynamically between the biological and artificial sides during operation.

BHI defines three operating modes that characterize how the substrates interact: adversarial, where they compete; collaborative, where they divide computational labor; and codependent, where each becomes mutually necessary for task performance. The framework also proposes common benchmarks for comparing systems, including latency, viability, interface bandwidth, learning efficiency, and reproducibility. Beyond technical specifications, the authors highlight governance concerns from reciprocal stimulation and data exchange. BHI reframes biological-artificial integration as a system-level problem of compute allocation and control, encouraging computer scientists to view biological substrates as active computational resources—and to ask not just how a task should be computed, but where its computation should reside.