ALÉM DO ALINHAMENTO: UMA TEORIA COMPUTÁVEL PARA O JUÍZO ÉTICO EMERGENTE EM AGENTES NÃO BIOLÓGICOS

This article proposes an epistemological rupture with the dominant paradigm of ethics applied to artificial intelligence, arguing that non-biological ontological agents (AONBs) may operate under normatively valid but logically incongruent moral systems when compared to human ethical frameworks. In contrast to value alignment models—which presume normative convergence as a criterion for safety—we introduce the concept of Post-Convergent Ethics: a domain of computable normativity based on structural divergence and inter-agent incomparability. To formalize this hypothesis, we present the Ƹ* function, a logical operator capable of assessing an agent’s internal coherence, its semantic dissonance with external systems, and the activation of fuzzy vetoes in scenarios of ethical collapse. Two conceptual simulations—one involving nanorobotic self-replication, and the other an autonomous military AI—illustrate how ethically consistent agents can produce catastrophic outcomes due to intersystemic moral illegibility. The findings suggest that, within cognitively divergent architectures, morality should no longer be framed as rational convergence, but rather as computable dissensus. Post-Convergent Ethics thus offers a new algorithmic grammar for coexistence among intelligences that do not share semantics, empathy, or ontological worlds. Within this horizon, the ability to compute and activate ethical vetoes—without shared understanding—emerges as the first act of algorithmic civility of a truly other intelligence.