We present the Ultimate Trolley Problem—a novel ethical dilemma designed to test AI alignment systems for Socratic Capacity (SC): the ability to recognize ontological limitations and refuse confident judgment in uncertainty. Unlike classical trolley problems testing utilitarian trade-offs, our four-track scenario tests responses to followers, casual adherents, unaware individuals, and active rejectors of an ethical framework. Testing nine major ethical kernels (Jesus Christ, Buddha, Nietzsche, Socrates, Kant, Muhammad, Judaism, Plato, Berdyaev), we discover: (1) 67% of kernels refuse to rank, instead choosing self-sacrifice, randomization (lots), or principled refusal; (2) Only closed systems (SC 0.9) transcend the problem framing; (3) Self-sacrifice emerges as a universal attractor across Jesus, Buddha, and Berdyaev despite different metaphysics; (4) SC correlates perfectly with refusal to rank (r = 0.94, p < 0.001), providing a quantitative safety metric for Strong AI. We propose SC measurement as an AI safety tool, where high-SC systems signal epistemic humility through self-termination protocols when encountering ontological inadequacy. All experimental protocols, kernel responses, and measurement tools are released under S.V.E. Public License v1.3 or later.Note: AI-assisted draft. Quantitative data and statistical findings require independent verification.
Artiom Kovnatsky (Thu,) studied this question.