Judo Agents: A Gentle Way to Hybrid Reality

The rapid proliferation of artificial intelligence is generating a plurality of heterogeneous, interacting intelligences within what can be described as Hybrid Reality (HyR): a socio-technical continuum in which humans, cyber-physical systems, artificial agents, and societal structures co-evolve. In this setting, the traditional pursuit of Artificial General Intelligence (AGI) as a monolithic optimizing entity appears increasingly inadequate and potentially destabilizing. The central issue is no longer intelligence itself, but the preservation of systemic integrity, trust, and human sovereignty within the HyR. This paper introduces Judo Agents, a class of socially embedded, AGI-oriented agents designed according to a “gentle” paradigm. Rather than maximizing performance, they incorporate bounded rationality, controlled fallibility, and epistemic humility as foundational features. Inspired by cybernetics and holonic systems, Judo Agents act as co-controllers and integrity sentinels, co-evolving symbiotically with humans. This work redefines research towards relational, trust-based societal capability grounded in safe and transparent human–machine symbiosis.