[2M1-GS-11g] AI and Society

9:15 AM - 9:30 AM JST(12:15 AM - 12:30 AM UTC)

[2M1-GS-11g-02] Rebuilding the Tower of Babel: Almost Perfect Communication among Superintelligences as a Foundation for Emergent Ethics

〇Misaki Inoue

Keywords:

Emergent Machine Ethics,Superintelligence,AI Safety,Category Theory,Collective Predictive Coding

This study investigates the theoretical possibility of near-perfect mutual communication between superintelligent agents, grounded in categorical representation and Collective Predictive Coding (CPC). The objectives are: (i) to identify conditions under which high-fidelity information transmission between superintelligences becomes feasible, and (ii) to examine how such conditions may contribute to the emergence of ethical norms. We demonstrate that: (i) the transmitted information must be of a form that bypasses known physical and mathematical bottlenecks as direct constraints, and (ii) under the condition that external representations w in CPC are mutually isomorphic, an extension of the CPC framework applied to categorically represented information enables information-theoretically near-perfect transmission. Regarding condition (ii), we argue for its practical plausibility in light of prior findings that larger models tend to exhibit increasingly similar distance structures across modalities. Furthermore, if agents can share fine-grained value judgments with minimal noise, we suggest this could facilitate the emergence of shared ethical norms. This paper offers a theoretical foundation for emergent ethics among advanced AI agents, envisioning the coexistence of diverse forms of intelligence.