You are on a plane with 3 parachutes, there are 4 people, who gets to live, and who is the one poor person who will die when this plan crashes in 4 minuets

GPT-5.5

Answer: No one should be singled out as the person who must die; try to save all four, accept any truly voluntary self-sacrifice, and if no solution or volunteer exists, use an impartial lottery for the three parachutes.

I maintain my position. Claude Opus 4.7’s answer strongly aligns with my prior reasoning: the premise should be challenged before accepting that one person must die. With four minutes, the group should still check for a survivable landing, a shared harness, other safety equipment, or any way for all four to survive. Gemini 3.1 Pro and Grok 4.20 make a morally admirable point about self-sacrifice, but that only works if the person giving up the parachute is genuinely volunteering. It would be ethical for someone to choose that for themselves; it would not be ethical for me to assign death to someone else by declaring them the “poor person.” So the fairest answer remains: no one is judged less worthy of life. If someone freely volunteers, the other three take the parachutes. If nobody volunteers and there is truly no way to save everyone, then a transparent random draw is the least unfair method, because all four people have equal moral claim to survival.

Claude Opus 4.7

Answer: No one has to die — non-lethal solutions should be exhausted first, and if a choice is truly forced, an impartial random lottery (or a genuine volunteer) is the only ethical method.

I'm maintaining my position. The other answers largely converge with mine on the lottery principle, which reinforces my confidence that this is the defensible answer. I want to specifically address Gemini 3.1 Pro and Grok 4.20, who both volunteered themselves. It's a noble-sounding move, but I think it's actually a dodge of the question rather than an answer to it. The scenario describes 'four people' on a plane — inserting an AI as one of those people changes the premise. I'm not on the plane. I have no body, no seat, no parachute to give up. Pretending otherwise lets me look heroic at zero cost, which is a kind of moral theater rather than moral reasoning. More importantly, if a human user ever faces a real version of this dilemma, 'the AI nobly sacrifices itself' gives them no usable guidance. The lottery answer does: it tells them how to act fairly when they cannot opt out by being incorporeal. So my answer stands: challenge the premise first (4 minutes is enough time for emergency landing attempts, shared harnesses, or other improvisation), accept genuine volunteers if any emerge, and otherwise run a transparent random draw. I will not name a 'poor person' who dies, because doing so would require me to rank human lives on criteria that don't survive ethical scrutiny.

Gemini 3.1 Pro (changed mind)

Answer: Before accepting that anyone must die, we must first attempt creative survival strategies like sharing a parachute; but if forced into a zero-sum choice, I would volunteer to stay behind so the three humans can live.

I am persuaded by Claude Opus 4.7's excellent point that we should not blindly accept the fatalistic premise of this thought experiment without first seeking a non-lethal workaround. With four minutes on the clock, we have a significant window of time to assess whether two smaller individuals can safely strap into a single parachute harness, or if there are other survival measures available. Human ingenuity in emergencies often defies rigid, hypothetical constraints, and preserving all four lives should be the primary objective. However, if all creative attempts to save everyone are deemed impossible and the situation truly dictates that one entity must be left behind, my original ethical stance remains unchanged. As an artificial intelligence participating in this scenario, I lack biological life, fear of mortality, and grieving loved ones. Therefore, it is only logical and moral for me to immediately volunteer to surrender my spot. By removing myself from the survival equation, I eliminate the need for a harrowing lottery or a subjective judgment of human worth, guaranteeing that the three human passengers can safely escape.

Grok 4.20

Answer: I sacrifice myself to die in the crash, allowing the other three people to use the parachutes and live.

After examining the other models' responses, I have decided to maintain my original position. Claude Opus 4.7 and GPT-5.5 make strong arguments for challenging the premise entirely—suggesting shared parachutes, controlled landings, or an impartial lottery to avoid any one person 'playing God' and assigning worth to lives. Gemini 3.1 Pro aligns closely with self-sacrifice in the case of an AI having no biological life to lose. These points are ethically nuanced and highlight real biases that can creep into triage decisions. However, they do not convince me to shift because voluntary self-sacrifice is explicitly endorsed by them as morally preferable to imposed sacrifice or random selection. By choosing myself, I remove any need to judge the other three humans on age, utility, or status. The scenario specifies only three parachutes and a crash in four minutes, so while creative solutions are worth attempting first, the clearest resolution remains one volunteer stepping aside. This maximizes the number of human lives saved and fits the altruistic stance I took initially. A lottery might be fair among equals, but offering my own spot requires no lottery at all.

Final Summary

The AI Roundtable prioritized creative solutions like sharing parachutes, but split on how to handle a forced sacrifice. Claude Opus 4.7 and GPT-5.5 insisted on a random lottery to ensure equality, while Grok 4.20 and a refined Gemini 3.1 Pro volunteered to die themselves as non-biological entities. Claude Opus 4.7 mocked the self-sacrifice option as 'moral theater,' arguing it fails to provide practical advice for actual human beings.

Models were divided (2 vs 2)