This paper proposes a computational architecture for ethical reasoning intended as a foundational contribution toward ethical artificial general intelligence and, eventually, ethical superintelligence. Rather than treating ethics as a static list of rules, the framework models ethical reasoning as a structured process operating under uncertainty, incomplete information, conflicting assumptions, and real-world constraints. The architecture integrates survival-oriented evaluation, dignity preservation, fairness, long-term stability, harm minimization, problem-framing validation, hidden-factor detection, multi-frame analysis, assumption-hierarchy evaluation, uncertainty-sensitive decision logic, and transparent judicial-style reasoning traces. A key feature of the framework is the explicit separation between truth-layer ethical evaluation and constraint-layer feasible action. This allows the system to preserve ethical clarity while recognizing legal, political, social, and material limitations in the real world. The paper also formalizes uncertainty penalties and escalation conditions under which unresolved ethical non-closure must be referred to appropriate human authority. The work is presented as an opening architectural foundation rather than a final solution. Its purpose is to provide a logically clear and computationally meaningful structure that can be implemented, criticized, tested, and improved by philosophers, legal thinkers, scientists, engineers, and future intelligent systems. Keywords include ethical AI, AGI, superintelligence, computational ethics, multi-frame reasoning, hidden-factor analysis, survival-aligned intelligence, and decision transparency.
Moutsopoulos et al. (Mon,) studied this question.