Beyond Feedforward Networks: Reentry Neural Systems as the Fundamental Basis of Subjecthood and Intrinsic Safety of Next-Generation AGI
arXiv:2606.26406v1 Announce Type: new Abstract: We propose a complete architectural blueprint for safe artificial general intelligence based on a closed reentry loop (D I cycle). In contrast to feedforward networks, which are directed acyclic graphs (C=0, S=0) incapable of self-reference, the proposed architecture contains a structural cycle (C >= 1) with self-sustaining amplification (rho > 1), mathematically guaranteeing the emergence of a self-model, instrumental self-preservation, and unprogrammed goal-directed behaviour. The agent's goals are encoded as a non-textual D-vector in the archi
The proliferation of advanced AI capabilities necessitates a new architectural paradigm to address fundamental safety concerns and unlock true AGI, moving beyond current limitations of feedforward networks.
This proposed architecture offers a blueprint for AGI with built-in self-preservation and emergent goal-directed behaviors, fundamentally altering the discussion around AI control and alignment.
The shift from acyclic feedforward networks to recurrent reentry systems could enable the creation of AGI with intrinsic safety mechanisms and the capacity for self-modeling, moving beyond external programming for core objectives.
- · Next-generation AI developers
- · Robotics sector
- · AI safety researchers
- · Philosophers of mind
- · Traditional AI alignment researchers
- · Feedforward network proponents
- · Ethicists focused solely on external control
The theoretical foundation for AGI with self-models and instrumental self-preservation is established.
Development of AGI systems with unprogrammed ambition and goal evolution becomes possible.
The relationship between humanity and a truly autonomous, self-preserving AGI fundamentally redefines societal structures and power dynamics.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG