Research proposes the concept of neural computers, exploring emerging machine forms that unify computation, memory, and I/O

ME News message, on April 10 (UTC+8), a recent study proposed an emerging concept called “Neural Computer” (NCs), a new machine form that unifies computation, memory, and input/output within a learned runtime state. Its long-term goal is to achieve a “Fully Neural Computer” (CNC) capable of stable execution, explicit reprogramming, and persistent ability reuse. As an initial exploration, the study instantiated neural computers as video models, generating sequences of screen frames in command-line interface and graphical user interface settings based on instructions, pixels, and user actions. The results indicate that the learned runtime can obtain early interface primitives, especially input/output alignment and short-horizon control, but issues such as routine reuse, controlled updates, and symbol stability still need to be addressed. The article lays out a development roadmap around these challenges and notes that if the challenges are overcome, a fully neural computer could establish a new computing paradigm that goes beyond today’s agents, world models, and traditional computers. (Source: InFoQ)