From Logic-Centric AI to Material-Driven Physical Intelligence: A Perspective on Co-Design Inspired by Biological Self-Organization

Current AI systems remain largely logic-centric: they attempt to resolve uncertainty in the physical world through software-level computation running on architectures that strictly separate memory, sensing, and actuation. While this paradigm has achieved remarkable performance in digital domains, it appears to face persistent limitations in real-world embodied intelligence, including latency, energy dissipation, and difficulties in assuring safe real-time behavior. Here, we present a conceptual framework for next-generation Physical AI based on material-driven co-design. We argue that by integrating advanced physical substrates—such as organic neuromorphic devices and nonlinear active materials—part of what is conventionally treated as "computation" can be realized directly through intrinsic physical dynamics. Crucially, we contend that this approach need not diminish the role of digital logic; rather, it may elevate it. By delegating high-frequency, local physical interactions to the material layer, centralized software may be better positioned to focus on high-level orchestration, context selection, and rigorous safety supervision. To formalize this synergistic distribution of intelligence, we introduce a three-layer architecture: (1) material computation, (2) trigger and orchestration, and (3) assurance and safety supervision. We outline comparative predictions and argue that distributing intelligence synergistically across materials and logic may offer a potentially robust, energy-efficient, and operationally bounded pathway that could help address SOTIF (Safety of the Intended Functionality: the absence of unreasonable risk from functional insufficiencies under nominal operation) requirements in the real world.