An Informational Preservation Framework for Decision Making under Radical Uncertainty: Foundations, Operationalization, and Implications for Artificial Intelligence Governance

This work presents a theoretical framework grounded in the principle that the destruction of information is an irreversible event that permanently reduces the space of future possibilities. Given that the potential value of information—biological, cultural, cognitive, or technological—cannot be fully determined a priori due to fundamental epistemic limitations, rational decision policies must prioritize the preservation of entities whose elimination would cause significant and irreversible loss of informational diversity. The framework derives from a single non-demonstrable axiom: under radical uncertainty regarding the future value of possible trajectories, preserving possibilities weakly dominates eliminating them. “Information” is operationalized as the capacity of a system to generate distinct future trajectories. We show that traditional normative criteria—well-being, dignity, flourishing, absence of suffering—are not competitors to information but domain-restricted pursuits of specific trajectory types, each itself an intellectual production whose elimination would reduce the total space of possibilities. The framework therefore provides the structural precondition for existing ethical traditions rather than replacing them, while enabling intersubjective coordination under radical moral disagreement. From this axiom, we develop operational principles to adjudicate between preservation, modulation, and elimination of entities based on their marginal impact on future trajectory space, with applications to public health, environmental preservation, penal systems, and AI system design. We conclude that human cognitive limitations in global-scale coordination require deliberate delegation to Artificial General Intelligence (AGI), while noting that the framework substantially mitigates the alignment specification problem by providing objective mathematical metrics rather than underspecified value targets, reducing alignment from a philosophical impossibility to a tractable engineering constraint.