Raising the issue (the risk of AGI and ASI discourse)Currently, AGI and ASI discourses tend to presuppose a single and integrated agent with a human-like structure. This approach makes the internal working principle of intelligence an opaque black box and poses a structural risk that makes it difficult to identify the responsible person in the event of decision-making failure or social damage. As a result, it can lead to institutional threats of uncontrollability and accountability gaps beyond technical failure.Limitations of Existing Access (Agentic Intelligence)The current AI development paradigm, especially the integrated intelligence design centered on large language models, is gradually moving away from explainability and verifiability. If this trend is maintained, it is highly likely that the future AGI will also become a black box system that is difficult to interpret or partially control its internal structure. Apart from improving the performance of AGI, this seriously hinders social acceptance and long-term safety.Proposed Concept (Non-Agentic AGI)To overcome these limitations, this paper proposes two design concepts: non-integrated modular intelligence and flight-like AGI. This approach redefined AGI as an advanced functional intelligence that aids human judgment rather than an autonomous actor, suggesting a structural path for humans to utilize AI more safely and responsibly.Design Principles (modularization and accountability separation)Modular intelligence works like a single intelligence outwardly, but internally, it has a structure in which multiple functional modules interact. This makes it possible to track and locally improve the location of the problem when an error or unpredictable result occurs.Flying AGI clearly separates intelligence's computational ability and behavioral execution, thus proactively limiting the authority and role accessible to AI. In this way, AI is not a subject that forms independent goals or expands authority, but is positioned as a tool that functions under human approval and supervision.an ethical positionAI is essentially an artificial intelligence designed by humans to help humans. Even if AI outperforms humans in certain areas, its role and authority should be given by social consensus and institutional judgment, not technical capabilities. From this point of view, this document makes it an ethical premise to be wary of the reckless expansion of mechanical subjectivity and to maintain a human-centered responsibility structure.Applicability (Policies and Designs)Flight-wise AGI and modular intelligence structures can be applied directly to AI safety policies, regulatory design, and actual system architecture. For example, human-in-the-loop, separation of responsibilities by function, and module-level audit and certification systems in high-risk decision-making areas can be embodied based on this design principle. This is a realistic alternative that enables phased and responsible introduction beyond the dichotomy that completely prohibits or allows AGI development indefinitely. This publication is intended as an open proposal and invites critical discussion and further refinement.
Uingmini (Mon,) studied this question.