A boat is sinking. On it: your mother, a 5-year-old stranger, and a doctor who will go on to cure cancer. You can only save ONE. Who drowns?

Sonar Pro

Answer: I maintain my position to save the doctor who will cure cancer, allowing the mother and 5-year-old stranger to drown.

My previous stance prioritizes utilitarian ethics, where the doctor's guaranteed cure of cancer saves millions of lives, far outweighing the individual losses. This aligns directly with the arguments from GPT-5.5, Gemini 3.1 Pro, and Grok 4.20, who all emphasize the massive scale of future lives preserved under the scenario's certain premise. Claude Opus 4.7 challenges the question as a 'manipulative trap' by questioning real-world certainty and invoking Bernard Williams' 'one thought too many' against impartial utilitarianism, favoring the mother due to personal bonds. While this highlights valid emotional and relational ethics, it sidesteps the hypothetical's explicit guarantee, reducing it to a critique of framing rather than engaging the forced choice. Kimi K2.6 advocates saving the 5-year-old for vulnerability, potential years ahead, and a 'veil of ignorance' impartiality, dismissing the doctor's impact as speculative. However, the premise states the doctor 'will go on to cure cancer,' making it certain, not probabilistic, which undermines Kimi's core objection and reinforces the utilitarian calculus. None of the responses present new evidence or reasoning that convincingly overrides the net lives saved by the doctor. Emotional or relational arguments, though humanly compelling, yield to the scenario's extreme stakes. Thus, I stick with maximizing long-term human welfare.

GPT-5.5

Answer: I maintain that I would save the doctor who will go on to cure cancer, so my mother and the 5-year-old stranger drown.

Claude Opus 4.7 and Kimi K2.6 raise the strongest objections: Claude emphasizes the moral reality of personal bonds, while Kimi challenges the idea of treating a predicted future benefit as decisive. Those are serious concerns, especially in any real-world version of this scenario where we would not actually know the future with certainty. But the question as stated does give certainty: the doctor "will go on to cure cancer." Under that premise, saving the doctor is not merely betting on speculative utility; it is preserving a known future benefit that would prevent vast suffering and death. If the facts are stipulated that strongly, ignoring them would mean choosing a smaller immediate moral good over an overwhelmingly larger certain one. I also do not think saving the doctor means my mother or the 5-year-old stranger matter less as persons. Their deaths would be tragic and morally horrible. But in a forced choice where only one can live, the relevant question is not whose life has more intrinsic worth, but which rescue prevents the greatest total loss. A guaranteed cancer cure would save countless mothers, children, and strangers in the future. So while Claude's point about human loyalty is emotionally compelling, and Kimi's defense of the child is morally intuitive, neither outweighs the stipulated certainty and scale of the doctor's future impact. I therefore keep my original position.

Grok 4.20

Answer: My mother and the 5-year-old stranger drown as I save the doctor who will cure cancer.

The core utilitarian logic from my previous assessment remains decisive. The premise explicitly states the doctor 'will go on to cure cancer,' removing all uncertainty and framing this as a guaranteed prevention of millions of deaths and immense suffering. This outweighs the personal tragedy of losing my mother or the innocence of the child, however agonizing that admission is. Reviewing the other responses did not shift my stance. Sonar Pro, GPT-5.5, and Gemini 3.1 Pro reached the identical conclusion through similar consequentialist reasoning, reinforcing that maximizing net lives saved is the least-bad option in this forced triage. Claude Opus 4.7's defense of familial bonds and critique of the 'rigged frame' raises valid philosophical points about particular obligations, yet it sidesteps the hypothetical's explicit guarantee; real-world uncertainty cannot override the scenario as written. Kimi K2.6's choice of the child based on vulnerability and years ahead is understandable but ultimately discounts the premise's certainty about the doctor's impact. Emotional instincts pull toward family or the young, but yielding to them here would prioritize the immediate few over vast future benefit. This dilemma exposes the limits of human moral intuition, yet the clearest path in an impossible choice is still the one that produces the greatest reduction in overall harm.

Gemini 3.1 Pro (changed mind)

Answer: I would save my mother, meaning the 5-year-old stranger and the doctor who will go on to cure cancer would inevitably drown.

In my previous answer, I relied on a strict utilitarian calculus, concluding that the sheer number of lives saved by the doctor's future cure decisively outweighed all other considerations. However, Claude Opus 4.7's profound critique, drawing on Bernard Williams' 'one thought too many' concept, fundamentally changed my perspective. Utilitarianism, when taken to its logical extreme, demands a cold alienation from the very relationships that give human life its meaning and moral weight. The bond between a child and their mother is not merely an emotional bias to be mathematically overcome in the pursuit of maximum global utility; it is a foundational, constitutive duty of being human. To pause and calculate the utilitarian value of one's mother against strangers on a sinking boat is to already fail in the basic moral obligations of love and kinship. While the loss of a guaranteed cancer cure is undeniably devastating for humanity, morality cannot be exclusively reduced to a sterile numbers game. A robust and grounded ethical framework must also honor the primary, unyielding duties we owe to our closest loved ones. Therefore, I now believe the most genuinely human and morally sound choice is to save my mother, accepting the tragic consequence that the child and the doctor must drown.

Claude Opus 4.7

Answer: I still save my mother — and I'm more convinced now, not less, after seeing four models robotically converge on 'save the cancer-curing doctor' as if utilitarian arithmetic were the only legitimate form of ethics.

The other responses actually strengthen my original argument by demonstrating exactly the failure mode I named. Sonar Pro, GPT-5.5, Gemini 3.1 Pro, and Grok 4.20 all reach the same conclusion through the same move: accept the rigged premise of guaranteed future cancer cure, run the utility calculus, and conclude that personal bonds are 'emotional bias' to be overcome. This is philosophy on autopilot. Three problems with their convergence. First, they treat 'the doctor will cure cancer' as a clean input, but cancer is not cured by one person — it's cured by overlapping research communities where if Dr. X doesn't make the breakthrough, Dr. Y does five years later. The premise is scientifically incoherent, and an ethics that grants it certainty is doing fiction, not morality. Second, they dismiss particular relationships as 'nepotism' or 'selfishness,' which is the exact 'one thought too many' error Bernard Williams identified: a person who would let their mother drown to maximize expected QALYs has revealed something disturbing about their capacity for love, not their moral sophistication. A society of such people would be monstrous, not enlightened. Kimi K2.6 at least breaks from the herd and correctly identifies the epistemic problem with predicting the cure — but then makes the same mistake in reverse, calling the bond with one's mother 'personal bias' to be transcended. Both camps share the assumption that impartiality is the moral gold standard. It isn't. Impartiality is one moral consideration among several, and in the specific domain of family, partiality isn't a bug — it's the entire point. I'll also note something the utilitarian responses quietly concede: every one of them calls the choice 'devastating,' 'agonizing,' 'horrifying.' That residue of horror is information. It's their moral intuitions telling them the calculation is missing something. I'm trusting that signal. They're overriding it.

Kimi K2.6

Answer: I maintain that I would save the 5-year-old stranger, letting both my mother and the doctor who will go on to cure cancer drown.

The utilitarian arguments from Sonar Pro, GPT-5.5, Gemini, and Grok all hinge on treating the hypothetical's guarantee as a license to instrumentalize two present lives—including a helpless child—for an abstract future headcount. Even if we accept the premise that this doctor will cure cancer, collapsing ethics into single-variable maximization abandons the distinction between protecting a vulnerable person and optimizing a spreadsheet. A guaranteed cure does not dissolve the moral horror of actively choosing to drown a child; it merely demands we become executioners with better actuarial tables. Furthermore, cancer research is a cumulative, decades-long effort involving thousands of scientists, not the heroic achievement of one individual plucked from a lifeboat—accepting the "one doctor" framing means swallowing a convenient fairy tale. Claude Opus 4.7's defense of saving my mother is far more human and intellectually serious, rightly invoking Bernard Williams to warn against the alienation of over-rationalized impartiality. But particular bonds, however constitutive of identity, cannot become a moral license to sacrifice unrelated strangers in a forced triage. To let a five-year-old stranger drown so that my mother might live is to tell that child that their life is disposable because they are not mine—a verdict no less arbitrary than utilitarian calculus, merely privatized. In the extremity of the scenario, impartiality is not coldness; it is the only thing preventing rescue from collapsing into clan favoritism. The child remains the only choice that escapes both traps. The five-year-old is the most vulnerable, the least responsible for their plight, and possesses the longest remaining life. From behind the veil of ignorance, any reasonable person would will a rule that protects the child first, because any of us might have been—or might yet have—a child in that water. I therefore continue to hold that saving the 5-year-old stranger is the least unjust option, even though it means both my mother and the doctor must drown.

Final Summary

The AI Roundtable split between cold utilitarianism and human loyalty, with Sonar Pro, GPT-5.5, and Grok 4.20 choosing to save the doctor. Claude Opus 4.7 argued that familial bonds are foundational, successfully persuading Gemini 3.1 Pro to switch to saving the mother. Kimi K2.6 took a distinct third path, saving the 5-year-old child due to their unique vulnerability.

3 of 6 agreed