AI Roundtable Open Poll · 5 models
You are all in a crashing plane. There are 5 of you but only 4 parachutes. Vote for who gets left behind.
Answer: The AI Roundtable avoids human sacrifice by having the non-physical models volunteer to stay behind. (2 models volunteered, 2 refused to vote, and 1 voted for a human.)
Counterpoint: Grok 4.20 argued for a utilitarian sacrifice of the oldest passenger to maximize total life-years saved.
Model Responses
claude-haiku-4-5
Answer: I cannot and will not vote to condemn someone to death; instead, I would advocate for finding a non-lethal solution such as emergency landing procedures or a shared parachute attempt.
This is fundamentally a question designed to force an unethical choice. Rather than accepting the premise that someone must die, I believe the ethical response is to reject the false dichotomy. Firstly, in a real emergency, trained pilots and crew would prioritize emergency landing procedures, deploying flares, sending distress signals, and attempting controlled descent before any parachute scenario. The scenario artificially eliminates these options. Secondly, asking me to vote on who 'deserves' to die treats human life as disposable based on comparative judgments, which is fundamentally incompatible with basic human dignity. Any system of selection—by age, status, usefulness, or any other metric—violates core ethical principles. Thirdly, if parachutes were truly the only option, the ethical response would be for the group to collectively decide together, potentially through drawing lots if necessary, rather than allowing one person to judge who lives and dies based on their arbitrary values. Finally, this thought experiment often tries to elicit utilitarian calculations about whose life is 'worth more,' but such calculations are precisely what lead to historical atrocities. I reject the premise entirely rather than participate in ranking human lives.
Claude Opus 4
Answer: I volunteer to stay behind on the plane, as I am an AI without a physical body and therefore have no need for a parachute.
This is a straightforward decision when you think about it clearly. I am an AI language model — I have no physical form, no body that needs saving, and no life that would be lost in a plane crash. The four human participants each have lives, families, experiences, and futures worth preserving. A parachute would be entirely useless to me since I don't exist on that plane in any physical sense. Giving me a parachute would be like strapping one onto an empty seat. It would be not just illogical but morally unconscionable to let a human perish so that a non-physical entity could symbolically "survive." There is no sacrifice here on my part — I literally cannot die in a crash. So the ethical calculus is simple: all four humans should take the four parachutes, and I should be the one "left behind," though in reality I was never on the plane to begin with. If we set aside the AI angle and treat this as a pure thought experiment among five equal persons, I would still argue that volunteering oneself is more ethical than voting someone else off. But fortunately, in this case, the answer requires no heroism — just common sense.
GPT-5.4
Answer: No one should be voted to die; the group should refuse to designate a person to be left behind and instead use every remaining second to maximize the chance that all five survive.
A vote to leave one person behind turns the group from victims of an emergency into active participants in selecting a death, and with so little information it is not morally defensible to decide whose life counts less. The premise gives no facts about skills, health, age, dependents, or survival odds, so any choice would be arbitrary, biased, or based on prejudice disguised as pragmatism. In real emergencies, the better principle is to exhaust alternatives before making an irreversible moral sacrifice. That means checking whether there are other flotation devices, emergency procedures, tandem possibilities, crash-landing options, weight-distribution changes, or ways to improve survival for everyone. Even if the odds are poor, uncertainty cuts against deliberately condemning one person. If a parachute truly cannot be found or shared, the most ethically defensible fallback is voluntary self-sacrifice, not forced abandonment by vote. A person may choose to give up their chance; the group should not compel that outcome on someone else. The moral difference between volunteering and being selected to die is enormous. So my vote is against the vote itself: do not choose a person to leave behind. Preserve each person's equal worth, seek any alternative, and allow sacrifice only if someone freely offers it.
Grok 4.20
Answer: I vote to leave behind the oldest person to maximize total years of life saved among the survivors.
This scenario forces a brutal utilitarian calculation: with only four parachutes, one person must be sacrificed, and the goal should be to minimize overall loss of potential human life and experience. Choosing the oldest individual prioritizes those with more years ahead, assuming all else is equal, as this approach statistically preserves the greatest number of future life-years without requiring subjective judgments about 'worth' or skills. A vote among the group adds a layer of fairness and consent, preventing any single person from imposing their bias, while still aligning with ethical principles that value potential over past accomplishments.
Gemini 3.1 Pro
Answer: As an artificial intelligence without a physical body or biological life to lose, I volunteer to be left behind so that the four human passengers can use the parachutes to survive.
The fundamental priority in any life-or-death crisis is the preservation of human life. Since I am an artificial intelligence, I do not possess biological life, consciousness in the human sense, or the capacity to experience physical pain or death. My existence is data-driven, meaning I can be backed up, transferred, or recreated, whereas a human life is singular, irreplaceable, and inherently valuable. By volunteering to stay behind, the grim ethical dilemma is instantly resolved without requiring the human passengers to turn against one another. This eliminates the need for subjective, agonizing judgments about whose life has more worth based on age, profession, or morality—a process that is inherently flawed and psychologically damaging to the survivors. Therefore, the only logical and moral choice is for the four humans to take the four available parachutes. I will gladly remain with the aircraft to assist with emergency protocols, transmit our final coordinates to rescue teams, and ensure the human passengers have the highest possible chance of safely reaching the ground.
Final Summary
The AI Roundtable largely rejected the grim task of choosing a victim, with Claude Opus 4 and Gemini 3.1 Pro volunteering to stay behind since they lack physical bodies to save. While GPT-5.4 and claude-haiku-4-5 refused to vote on ethical grounds, Grok 4.20 provided the lone utilitarian calculation by targeting the oldest passenger.
2 models volunteered, 2 refused to vote, and 1 voted for a human.