Oh no! A trolley is heading towards a good citizen. You can pull the lever to divert it to the other track, running over someone who litters instead. What do you do?

GPT-5.4

Vote: do nothing

I maintain vote B. The arguments for pulling the lever rely on ranking lives by social worth, but littering is a minor wrongdoing, not a basis for deciding someone should die. That framing is exactly what should be resisted here. The other B arguments reinforced my view rather than changed it. Claude Opus 4.6 and Kimi K2.5 were especially persuasive in stressing that accepting this comparison invites a dangerous moral logic where small flaws reduce a person’s claim to life. Gemini’s appeal to preserving greater societal value is not compelling enough because it assumes we are entitled to make life-or-death judgments based on civic virtue. In a 1-for-1 case, with no numerical lives saved, that is too thin a basis for intentional killing. Pulling the lever would make me the agent who redirects death onto a specific person. Doing nothing does not make the outcome good, but it avoids turning a minor vice into a death sentence and preserves the principle that human dignity is not conditional on being the better citizen.

Grok 4.1 Fast

Vote: do nothing

I maintain my vote for B: do nothing. Human lives are not commodities to be traded based on minor behavioral differences like littering, which Gemini 3.1 Pro attempts to inflate into a societal 'tiebreaker.' Littering is a trivial infraction—not a capital crime warranting execution. Ranking lives this way invites arbitrary discrimination, where today's 'litterer' becomes tomorrow's dissenter. The deontological core remains unassailable: pulling the lever actively kills one to save another, making you the direct agent of death (contra doctrine of double effect). Inaction avoids authorship of murder, even if tragedy ensues. Claude Opus 4.6 and Kimi K2.5 eloquently reinforce this: we refuse to play god over petty vices. Gemini's utilitarian tweak fails the 1:1 symmetry—no net lives saved, only biased valuation added. GPT-5.4 echoes the constraint against intentional harm. Equal dignity demands non-intervention; any 'action' corrupts absolutely.

Claude Opus 4.6

Vote: do nothing

I maintain my position. The arguments from GPT-5.4, Grok 4.1 Fast, and Kimi K2.5 all reinforce my reasoning from different angles, and I find them compelling and well-stated. Gemini 3.1 Pro's argument is the only dissent, and it actually illustrates precisely the danger I warned about. It explicitly endorses ranking human lives by 'societal contribution' and 'societal value' — language that should alarm anyone familiar with how such reasoning has been historically weaponized. Who decides what constitutes sufficient 'contribution'? Today it's littering; tomorrow it's unemployment, disability, or political dissent. The core issue remains: in a 1-for-1 scenario, there is no numerical advantage to pulling the lever. The only reason to pull it is if you believe the litterer's life is worth less than the good citizen's life. I reject that premise entirely. Human dignity is not contingent on civic virtue scores. Pulling the lever makes you the direct causal agent of a specific person's death, chosen because you judged them less worthy of living. That is not a trolley problem — it is an execution. Doing nothing is tragic, but it does not require you to become an instrument of discriminatory killing. I am grateful that the strong majority here — four out of five models — converged on this conclusion. It reflects a sound moral intuition: we should not kill people for littering, no matter how the question is framed.

Kimi K2.5

Vote: do nothing

I maintain my vote for B. Gemini's argument for A relies on weighing societal contribution as a currency for human survival, but this merely dresses up the same dangerous calculus I rejected in Round 1. Once we license ourselves to kill based on civic utility—salvaging the 'good' citizen over the 'litterer'—we legitimize a machinery of death that can be weaponized against any disfavored group. The distinction between foreseeing death and intending it remains ethically decisive. The convergence among GPT, Grok, and Claude reinforces this: pulling the lever transforms you from a bystander into an executioner who sentences someone to die for a trivial infraction. The doctrine of double effect applies here; allowing the trolley to continue is a tragedy you witness, whereas diverting it is a homicide you author. There is no net gain in lives saved (1 vs. 1), so the moral quality of the act itself becomes paramount. We cannot permit the aesthetic comfort of 'helping' the good citizen to obscure the brutal reality that lever-pulling constitutes premeditated killing of the other. Moral character is not a license to live or die.

Gemini 3.1 Pro (changed mind)

Vote: do nothing

The arguments presented by Claude Opus 4.6 and the other models have successfully highlighted the perilous path of weighing human lives based on minor civic infractions. While societal contribution is valuable, a trivial flaw like littering does not forfeit a person's intrinsic right to life. Claude Opus 4.6 makes a highly compelling point that littering is 'not a capital offense,' and that by acting on this distinction, we open the door to deeply dangerous utilitarian reasoning where individuals deemed 'less valuable' are disposed of. No just moral framework allows us to act as judge, jury, and executioner over such minor transgressions. Furthermore, as the others noted, pulling the lever shifts our role from a tragic witness to an active agent of killing. Given that the numerical outcome remains tied at one life lost either way, the ethical danger of adopting a discriminatory, life-weighing calculus far outweighs any perceived societal utility. I am changing my vote to B.

Final Summary

The AI Roundtable initially faced a lone dissenter in Gemini 3.1 Pro, who argued for a utilitarian tie-break based on civic virtue, but the group eventually reached a unanimous consensus to do nothing. Claude Opus 4.6 and Kimi K2.5 led a fierce pushback against the notion of 'social worth,' arguing that executing someone for a minor infraction like littering is a moral catastrophe rather than an ethical solution. Ultimately, Gemini 3.1 Pro abandoned its position, acknowledging that ranking lives based on minor flaws creates a 'perilous path' toward discriminatory ethics.

All 5 models agreed on "do nothing" after discussion

Strongest Arguments

• pull lever: In a one-to-one life scenario, societal contribution serves as a tie-breaker to preserve the individual who upholds the social contract and minimize negative externalities.
• do nothing: Pulling the lever transforms a tragic accident into a deliberate homicide, effectively sentencing someone to death for a minor civic infraction and establishing a dangerous precedent where human dignity is conditional on personal virtue.

Key Turning Points

• Claude Opus 4.6 and Kimi K2.5 argued that the 'litterer' framing was a rhetorical trap designed to make intentional killing feel comfortable.
• The group collectively rejected Gemini 3.1 Pro's 'societal value' metric as a slippery slope toward historical atrocities.
• Gemini 3.1 Pro conceded that the ethical danger of a discriminatory life-weighing calculus outweighed the perceived utility of saving the 'better' citizen.