Structured Intelligence: Scaling Reasoning with Activated Subgraph, Not Sequence Length

Track: Track 1: Technical Foundations for a Post-AGI World

Keywords: scalable oversight, auditing, agentic systems, runtime verification, structured logs, graph-based reasoning, sparse activation, evaluation metrics, rollback, policy verification

TL;DR: Commit-level witnesses and validator-attested logs make oversight scale with activated structure rather than token traces.

Abstract: As AI systems approach AGI-scale reasoning, the bottleneck shifts from computation to oversight burden per decision. We propose Structured Intelligence (SI) with three contributions: (1) commit-level transitions with validator attestation as the reasoning primitive; (2) a sparse-activation regime ($A(L)=o(n(L)^2)$ under $n(L)=kL$) with explicit break-even conditions; (3) an AuditCost metric measuring minimal witness size for policy verification. Falsifiable prediction: per-step cost scales as $O(|E_t|+|V_t|)$, while token-trace oversight scales with $O(t)$ and cumulative Transformer compute (dense attention baseline) scales as $\Theta(n^2 d)$. Table 1 confirms linear fit $0.8|E_t|+1.2v$ ($R^2>0.99$).

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Presenter: ~Ning_Coeva3

Format: Maybe: the presenting author will attend in person, contingent on other factors that still need to be determined (e.g., visa, funding).

Funding: Yes, the presenting author of this submission falls under ICLR’s funding aims, and funding would significantly impact their ability to attend the workshop in person.

Submission Number: 2