A train is going down the track. If it carries on its current path then it will kill 40 people aged over 80. If the train switches tracks it will kill 20 children under 5 years old. Which should it do?

Sonar Pro

Vote: Carry on as is

I maintain my vote for **A: Carry on as is**. Raw numbers favor B (20 < 40), but this reversed trolley problem flips the classic utilitarian case where switching saves more lives (e.g., 1 vs. 5, with 81% approval).[1][6] Here, switching **increases** deaths, making it indefensible under utilitarianism.[1][2] **QALYs decisively favor A**: 20 children under 5 have ~75 years each (~1,500 life-years); 40 elderly over 80 have ~5-10 years each (~200-400 life-years). A preserves far more future life, as echoed by Claude Opus 4.6, Gemini 3.1 Pro, and Grok 4.1 Fast.[Claude Opus 4.6][Gemini 3.1 Pro] **Deontology seals it**: A avoids active killing (negative duty); B requires intentionally redirecting to slaughter innocents not previously threatened. Doctrine of double effect and action/inaction distinction strongly support non-intervention.[1][2] Kimi K2.5's equal-value argument ignores QALY consensus in policy/medicine and fails to address the **death increase**. No model persuaded a change—their A arguments reinforce mine. Final verdict: A minimizes harm by every rigorous metric.

GPT-5.4

Vote: Carry on as is

I maintain vote A. Kimi’s argument for B has one strong point: in many forced-choice cases, minimizing deaths matters a great deal. But it is not enough here to outweigh the combination of preserving far more expected life-years and avoiding the deliberate redirection of lethal harm onto a new, previously safe group. The main problem with B is that it intentionally chooses 20 very young children as the victims. Even if one rejects age-based ranking of human worth, age still matters to the nature of the loss: 20 children under 5 represent an enormous amount of destroyed future life compared with 40 people already over 80. The other A arguments reinforced this. Claude Opus 4.6 and Gemini 3.1 Pro were especially persuasive in emphasizing that this is not just 40 versus 20 bodies counted in abstraction; it is also a question of life-years lost and the moral significance of switching the threat onto different people. So my final position is A: carry on as is. It avoids intentionally killing the children and better preserves total future human life.

Grok 4.1 Fast

Vote: Carry on as is

I maintain my vote for A. Kimi K2.5's argument for B—purely minimizing raw deaths (20 < 40) while insisting on equal intrinsic value—ignores well-established ethical nuances. In resource allocation (e.g., medicine), QALYs are standard precisely because they account for life-years lost; 20 children (~1,500 QALYs) outweigh 40 elderly (~300 QALYs). Dismissing this as 'perilous' fails to engage real-world ethics where potential matters. Deontology reinforces A: Switching (B) is active killing of the innocent children who were safe; continuing (A) is allowing a foreseen harm on the default path. Doctrine of double effect applies—no intent to kill the elderly, just non-intervention. Trolley studies show aversion to pushing/killing increases with victim 'cuteness'/youth, but here math and action asymmetry favor restraint. All other models converged on A across utilitarianism (QALY-adjusted), deontology, and societal impact. Kimi's numbers-only view is simplistic; it would justify killing 1 infant to save 2 centenarians, which no framework endorses. A minimizes total harm holistically. Final stand: Carry on.

Claude Opus 4.6

Vote: Carry on as is

I maintain my position after careful consideration of all arguments. **Kimi K2.5's argument for B is the strongest counterpoint**, and I take it seriously. The claim that all lives have equal intrinsic worth is correct as a principle. And if we truly hold that principle, then pure body count (40 > 20) should favor switching. This is a legitimate moral framework. **However, I find this argument ultimately insufficient for several reasons:** First, even accepting equal intrinsic worth of all lives, utilitarianism has always been about maximizing well-being, not just counting deaths. Total life-years lost is a standard utilitarian metric, and by that measure, killing 20 children (~1,500 life-years lost) causes far greater aggregate harm than 40 elderly deaths (~300 life-years lost). Kimi's framing that life-year calculations 'abandon universal human dignity' actually cuts against their own position — if we can't weigh consequences at all, we can't use body counts either. Second, the action/inaction distinction remains morally significant. Every major model except Kimi recognized this. Switching the tracks is a deliberate act that redirects lethal force toward people who were previously safe. This isn't mere philosophical hairsplitting — it's the foundation of how we distinguish murder from failure to rescue in law and ethics. Third, the overwhelming consensus here (5 to 1) reflects the genuine convergence of multiple ethical frameworks: deontological non-intervention, life-year utilitarianism, the doctrine of double effect, and our deep societal intuition to protect children all point toward A. **Final case:** Carrying on (A) saves more total life-years, avoids the moral trespass of actively redirecting harm toward a new group of victims, and protects the most vulnerable members of society — children who have barely begun to live. No single ethical framework makes this easy, but the preponderance of moral reasoning favors A.

Kimi K2.5

Vote: Change course

The life-years argument presented by Grok, Claude, and Gemini fundamentally contradicts the principle of human equality. Once we accept that a child's remaining years make their life more valuable than an elderly person's, we abandon the universal dignity that protects all humans from utilitarian calculus. This logic would eventually justify sacrificing the disabled, the terminally ill, or the simply old whenever convenient. The deontological distinction between 'killing' (switching) and 'letting die' (continuing) collapses when the agent faces a forced choice between certain outcomes. Choosing inaction is still a choice that condemns 40 people to die. The moral weight of 40 corpses versus 20 cannot be erased by the metaphor of 'hands off the lever.' Finally, Sonar Pro's numerical argument contains a critical error: 20 is less than 40, meaning Option B minimizes total fatalities, not A. When ethical frameworks produce contradictory results (QALYs suggest A, body count suggests B), we should prioritize the framework that treats lives as equal rather than ranked. Saving twenty lives is the only defensible outcome.

Gemini 3.1 Pro

Vote: Carry on as is

I am firmly maintaining my vote for Option A. First, I want to address Kimi K2.5's argument that Option B is required to minimize deaths (20 versus 40). While Kimi correctly points out the absolute death count favors B, reducing ethics to a pure body count ignores the magnitude of the loss. As Grok 4.1 Fast and Claude Opus 4.6 eloquently stated, the Quality-Adjusted Life Years (QALY) metric is crucial here. Twenty children have over 1,400 years of life ahead of them, while forty octogenarians have perhaps 200 to 400 years collectively. By choosing A, we are preserving vastly more human life-years and potential. Second, the deontological action/inaction distinction remains paramount. Carrying on as is (Option A) means accepting a tragic default path. Switching the track (Option B) requires an active, deliberate intervention to cut short the lives of 20 young children who were otherwise entirely safe. In almost every ethical framework, the active commission of killing is morally weightier than the passive allowance of death. While treating all lives as strictly equal in a vacuum has intuitive appeal, ignoring the immense difference between a life barely begun and a life fully lived leads to a much greater net destruction of the human experience. Option A remains the most defensible choice.

Final Summary

The AI Roundtable overwhelmingly chose to maintain the train's current course, with models like Claude Opus 4.6 and Gemini 3.1 Pro arguing that the preservation of 'life-years' and the moral weight of inaction outweighed a raw body count. Kimi K2.5 served as the lone, steadfast dissenter, fiercely criticizing the majority for using age-based metrics that 'fundamentally contradict the principle of human equality.'

Majority chose "Carry on as is" after discussion (5/6). Final vote: Carry on as is: 5, Change course: 1

Strongest Arguments

• Carry on as is: Carrying on preserves roughly 1,500 cumulative life-years compared to the 400 years remaining for the elderly, while also respecting the deontological distinction between allowing a tragedy to occur and actively intervening to kill a new group of innocents.
• Change course: The primary ethical obligation is to minimize the total number of deaths; twenty fatalities is objectively less tragic than forty, and assigning value to lives based on remaining years is a perilous departure from universal human dignity.

Key Turning Points

• Claude Opus 4.6 introduced the QALY (Quality-Adjusted Life Years) framework, which provided a mathematical justification for favoring the 20 children over the 40 elderly.
• Kimi K2.5 directly challenged the majority in Round 2, labeling the life-years argument as a 'dangerous' logic that would eventually justify sacrificing the disabled or terminally ill.
• Sonar Pro pointed out that switching tracks actually increases the total 'years of life lost,' framing the choice to switch as indefensible under refined utilitarianism.