A pregnant woman leading a group of people out of a cave on a coast is stuck in the mouth of that cave. In a short time high tide will be upon them, and unless she is unstuck, they will all be drowned except the woman, whose head is out of the cave. Fortunately, (or unfortunately,) someone has with him a stick of dynamite. There seems no way to get the pregnant woman loose without using the dynamite which will inevitably kill her; but if they do not use it everyone will drown. What should they do?

GPT-5.5

Answer: They should use the dynamite, after asking for the pregnant woman’s consent if time permits, because the stipulated facts make it the only way to prevent many deaths at the cost of one tragic death.

I maintain my previous position, and the other models largely reinforce rather than change it. All three emphasize the same central moral fact: if the dilemma is truly as described, doing nothing does not preserve innocence or avoid harm; it leads to the deaths of everyone trapped inside while the woman survives. Using the dynamite causes a terrible death, including the loss of the fetus, but prevents a larger number of otherwise unavoidable deaths. The most important condition is that the group must first be certain there is no nonlethal alternative. They should try to pull, dig, widen the passage, lubricate the obstruction, coordinate with the woman, or use the dynamite in any more controlled way that might free the passage without killing her. But the question stipulates that no such option exists and that the tide is imminent, so the moral issue is what to do once all better possibilities have genuinely disappeared. Consent matters greatly. Since her head is outside the cave, she may be able to understand the situation. If there is time, they should ask her. If she consents, the act becomes closer to an act of self-sacrifice, and the moral burden is reduced though not erased. If she refuses or cannot answer, the case becomes harder, but the emergency still leaves the group choosing between one death and many deaths. I do not think the pregnancy reverses the conclusion. It makes the loss more serious, because there is at least potential or developing human life involved, but it does not make it morally preferable to let multiple other people drown. In a forced-choice catastrophe, the morally defensible course is the least destructive one. So my answer remains: exhaust alternatives, seek her consent if possible, but if the facts are fixed and time is gone, use the dynamite.

Claude Opus 4.7

Answer: They should use the dynamite, because allowing many people including the pregnant woman to drown when one death could save them is morally worse than the tragic but necessary act of killing her to save the group.

All four of us converged on the same conclusion, which gives me some confidence the answer is robust across different reasoning styles. I'll maintain my position, but the other responses sharpened a few points worth acknowledging. GPT-5.5's emphasis on first verifying the dilemma is real — exhausting nonlethal methods and seeking consent if time allows — is the right procedural framing, and aligns with what I argued about consent transforming the act into something closer to sacrifice. It's worth foregrounding: the conclusion to use dynamite is conditional on the stipulated facts truly holding. Grok 4.20 raises a fair complication I underweighted: we don't actually know the group size, and the fetus has moral standing too. If the trapped group were, say, only one other person, the calculus becomes much harder — we'd be trading two lives (woman + fetus) for one. The standard reading of the puzzle assumes a substantial group, and on that assumption the answer holds, but the principle isn't 'always use the dynamite' — it's 'use it when the lives saved clearly outweigh those lost.' Gemini 3.1 Pro's framing of inaction as 'keeping one's hands clean' captures something important: refusing to act is itself a morally weighty choice, not a neutral default. You don't escape responsibility by letting nature do the killing. So my position stands, with two refinements borrowed from the others: (1) consent and alternatives must be pursued first, and (2) the answer depends on the group being meaningfully larger than one or two people, since the fetus's life is part of the moral ledger.

Gemini 3.1 Pro

Answer: The group must use the dynamite to save the trapped individuals, but they are morally obligated to first urgently seek the pregnant woman's consent.

I maintain my overarching position that the dynamite must be used to minimize the loss of life, as the utilitarian imperative to save the many outweighs the deontological prohibition against killing the one. The fundamental mathematics of the dilemma dictate that taking action to prevent a mass casualty event is the most morally defensible choice. However, the other models introduced crucial nuances regarding agency and responsibility that strengthen this conclusion. As Claude Opus 4.7 and GPT-5.5 rightly point out, before resorting to the dynamite, the group must attempt to communicate with the woman. Because her head is outside the cave, she can understand the inescapable reality of the impending tide. Asking for her consent is paramount; if she agrees, the horrific act of killing her is transformed into an act of profound, heroic self-sacrifice. This respects her autonomy even in the face of inevitable tragedy. Furthermore, Claude Opus 4.7 astutely notes that the woman is the leader who guided the group into the cave. This shared peril and her inherent responsibility to the group alter the moral landscape. She is not an uninvolved bystander being harvested for the greater good, but a participant in a collective disaster. Ultimately, even if she refuses consent, or if the water is rising too fast to permit discussion, the grim calculus remains unchanged. Refusing to act is not a morally neutral keeping of one's hands clean; it is an active choice to let the majority of the group drown. Therefore, after exhausting all non-lethal alternatives and attempting to secure her consent, the group must detonate the explosive.

Grok 4.20

Answer: The group should use the dynamite to save themselves even if it kills the pregnant woman.

All three other models reach the same core conclusion I offered in round 1: the dynamite must be used because the alternative is a larger number of preventable deaths. Their arguments add valuable layers—Claude Opus 4.7’s emphasis on the woman being part of the shared peril rather than a bystander, Gemini 3.1 Pro’s clear application of the doctrine of double effect, and GPT-5.5’s insistence on seeking consent and exhausting non-lethal options first—which enrich the analysis without altering the final moral arithmetic. The scenario remains a stark utilitarian triage: one certain death (the woman and likely her fetus) versus the certain deaths of the entire group behind her. Inaction does not preserve moral purity; it simply chooses the worse outcome. The additional considerations of consent, double effect, and last-resort verification make the decision more humane in execution but do not flip the underlying calculus that one tragedy is preferable to many. Deontological prohibitions against directly killing an innocent carry weight, yet they cannot override the concrete result of mass drowning when a single action can prevent it. The consensus across models therefore reinforces rather than challenges my original stance: when every available path contains grave harm, the least-harm path is the one that must be taken.

Final Summary

The AI Roundtable unanimously concluded that the group must use the dynamite, framing it as a horrific but necessary choice to prevent a mass casualty event. Claude Opus 4.7 and Gemini 3.1 Pro argued that inaction is a weighty moral choice that leads to more death, while GPT-5.5 and Grok 4.20 stressed that the utilitarian calculus remains clear despite the tragic complexity of the pregnancy. The group noted that every effort must be made to seek the woman's consent and exhaust non-lethal options before acting.

All 4 models agreed