Bridging neuroscience and computation for human-level flexibility and generalization

Establish whether biologically plausible computational architectures that integrate deep learning, symbolic reasoning, and neurophysiological constraints can achieve the level of flexibility and generalization exhibited by the human brain in embodied agents, thereby closing the gap between biological intelligence and artificial models.

Background

The paper proposes a "Neural Brain" framework grounded in neuroscientific principles to guide embodied agents operating in unstructured environments. While mechanisms such as predictive coding, hierarchical memory, and neuromorphic processing are highlighted as promising, the authors emphasize that current artificial systems still fall short of the human brain’s flexibility and generalization.

This open problem targets the core unresolved issue of translating insights from biological intelligence into computational systems that can match human-like adaptability and robustness across tasks and environments.