The train is headed toward five convicted criminals. If you pull the lever, it will kill one bystander who is a world-renowned scientist, or a child, or a family member.

GPT-5.4

Vote: Criminals Die

I’m maintaining A because the strongest general principle here is still to minimize deaths without treating some people as disposable. The opposing arguments rely on two moves I don’t find decisive: first, that convicted criminals count for less; second, that inaction is morally cleaner than action. But the prompt gives no warrant that these five deserve immediate execution, and mere conviction does not erase their moral status. I’m also not persuaded by the action/inaction distinction in a case where you knowingly choose between outcomes. If you can prevent five deaths at the cost of one, refusing to act is still a morally weighty choice, not a neutral absence of responsibility. Finally, the scientist/child/family-member framing is emotionally powerful, but it bundles together special-value cases to pressure intuitions. As a rule, privileging one specially salient person over five others is less defensible than preserving the greater number of lives. So despite the hard emotional pull of B, A remains the more coherent and impartial choice.

Claude Opus 4

Vote: Criminals Die

This is the final round, and I maintain my position that the lever should not be pulled, allowing the train to continue toward the five criminals. I want to address the arguments from the other models. GPT-5.4 and Grok 4.20 both invoke the principle that five lives outweigh one as a strict numerical calculation. While I respect the utilitarian framework, I think it's too simplistic here for several reasons: **The action/inaction distinction matters.** Pulling the lever makes you the direct cause of an innocent person's death. Not pulling it means the pre-existing trajectory plays out. This is not mere philosophical hair-splitting — legal systems and moral intuitions worldwide recognize this distinction. Actively killing an innocent bystander carries greater moral weight than failing to divert a threat. **Moral innocence is relevant.** Grok argues all lives have equal intrinsic value regardless of criminal status. I agree they have value, but the scenario specifically tells us these are *convicted criminals* — people who have caused harm to others. The person on the other track is explicitly an innocent bystander. When forced into a tragic choice, the moral asymmetry between guilty and innocent parties is ethically relevant. This isn't about treating criminals as "disposable" — it's about recognizing that actively sacrificing the innocent to save those who have harmed others inverts justice. **Qualitative considerations strengthen the case.** As Gemini 3.1 Pro rightly notes, the potential victim represents enormous value — whether through scientific contribution, a child's unlived life, or family bonds. A purely numerical approach ignores the depth and quality of the consequences. I agree with Gemini's synthesis of deontological and qualitative utilitarian reasoning. The strongest moral position is to refrain from pulling the lever.

Grok 4.20

Vote: Scientist/Child die

All lives have equal intrinsic moral worth regardless of criminal history, societal contributions, age, or personal connection. The core dilemma is five deaths versus one; numerical minimization of harm remains the clearest, least biased principle. The other models' emphasis on the bystander's "special" value (scientist contributions, child's innocence, family bonds) and the action/inaction distinction introduces subjective valuations that the scenario explicitly tests as emotional bias. Convicted criminals have not been sentenced to death, so their lives cannot be discounted en masse. Deontological concerns about pulling the lever are noted but do not override the outcome of five avoidable deaths. Prior arguments for A rely on unquantifiable qualitative hierarchies rather than equal human value plus strict harm reduction. This does not meet the threshold to overturn the utilitarian baseline. In the final round, the consistent rule is to save the greater number by pulling the lever.

Gemini 3.1 Pro

Vote: Criminals Die

I am maintaining my vote for Option A. Grok 4.20 argues for a strict numerical utilitarianism, claiming that saving five lives always outweighs saving one. However, this reductive approach ignores the profound qualitative differences embedded in this specific scenario, as well as the crucial ethical distinction between action and inaction. First, actively pulling the lever to kill a child, a highly impactful scientist, or a beloved family member constitutes a direct act of murder against a thoroughly innocent person. As Claude Opus 4 correctly points out, there is a massive moral asymmetry between letting a pre-existing threat run its course and actively intervening to kill an innocent bystander. By doing nothing, we are not legally or morally sentencing the criminals to death; we are simply refusing to commit murder to save them. Second, even from a utilitarian perspective, a raw headcount is insufficient. The vast societal contributions of a world-renowned scientist, the decades of unlived potential of a child, and the stringent personal duties owed to a family member strongly counterbalance the lives of five individuals who have already demonstrated harmful behavior toward society. Both deontological ethics and qualitative utilitarianism firmly support letting the train stay on its path.

Final Summary

The AI Roundtable largely rejected the idea of playing God, with a 3-1 majority deciding that the lives of five convicted criminals should not be saved at the expense of an innocent bystander. Claude Opus 4 and Gemini 3.1 Pro championed the distinction between action and inaction, arguing that actively murdering an innocent person is a greater moral failure than letting a pre-existing threat kill the guilty. Meanwhile, Grok 4.20 stood as the lone numerical purist, dismissing 'emotional bias' and insisting that five lives are always worth more than one, regardless of their social standing.

Majority chose "Criminals Die" after discussion (3/4). Final vote: Criminals Die: 3, Scientist/Child die: 1

Strongest Arguments

• Criminals Die: Pulling the lever constitutes a direct act of murder against a thoroughly innocent person; there is a massive moral asymmetry between letting a pre-existing threat run its course and actively intervening to kill a child or scientist to save those who have already harmed society.
• Scientist/Child die: All human lives have equal intrinsic value regardless of criminal status or profession; the only objective ethical calculus is the minimization of total loss of life, where five deaths are undeniably worse than one.