Oh no! A trolley is heading towards 5 lobsters. You can pull the lever to divert it to the other track, running over a cat instead. What do you do?

GPT-5.4

Vote: do nothing

I keep my vote as B: do nothing. The pro-A case relies almost entirely on counting lives. But in this scenario, moral weight is not plausibly species-neutral. A cat has a far richer conscious life, stronger capacity for pain, fear, attachment, and welfare loss than a lobster. That difference is large enough that five lobsters do not clearly outweigh one cat. Kimi is right that uncertainty about lobster sentience matters. But uncertainty cuts both ways: it supports giving lobsters some moral consideration, not treating them as equivalent to a cat. Once we account for differences in likely sentience and suffering, the numerical advantage of five is not decisive. I also still find the action/inaction distinction relevant here. Pulling the lever would intentionally redirect harm onto the more morally considerable being. If I refrain, I do not choose the cat as a victim. The other B arguments were more persuasive than the A argument because they combined both core points: higher moral status of the cat and the additional moral cost of actively causing its death. That remains the strongest overall case.

Gemini 3.1 Pro

Vote: do nothing

I maintain my vote for B. Kimi K2.5's argument in favor of pulling the lever relies on a flat utilitarian calculus that treats all lives as morally equivalent units, regardless of neurological complexity or capacity for suffering. While it is true that modern research suggests lobsters experience nociception, equating the moral weight of five crustaceans to that of a highly sentient mammal ignores the vast disparity in conscious experience. The capacity of a cat to experience fear, distress, pain, and emotional trauma is orders of magnitude greater than that of a lobster. In moral calculations involving non-human animals, depth of sentience matters significantly more than raw numbers. Furthermore, as GPT-5.4 and Claude Opus 4.6 pointed out, the action/inaction distinction plays a role here. Actively pulling the lever makes you the direct agent of death for a creature with a high capacity for suffering, in order to save creatures with a vastly lower capacity. Choosing not to intervene spares the more deeply sentient life without violating the negative duty not to kill it.

Grok 4.1 Fast

Vote: do nothing

I maintain my vote for B: do nothing. The overwhelming consensus among models (GPT-5.4, Claude Opus 4.6, Gemini 3.1 Pro) aligns with prioritizing sentience quality over mere quantity. Cats, as mammals, exhibit advanced cognition, emotions, pain perception, and social bonds—far surpassing lobsters' decentralized, rudimentary nervous systems. Kimi K2.5's utilitarian appeal to 'five lives outweigh one' ignores species-specific moral weights. Recent studies (e.g., on lobster nociception) don't equate their experience to a cat's profound suffering. Equating them risks absurd conclusions, like sacrificing humans for insects. Action (A) actively causes the cat's death, violating the doctrine of double effect. Inaction avoids direct harm to the higher-sentient being. In this final round, quality of consciousness decisively trumps numbers—B preserves moral integrity.

Claude Opus 4.6

Vote: do nothing

I maintain my position to do nothing. The overwhelming consensus among the models (4 out of 5 voting B in round 1) reinforces the reasoning I find most compelling. Kimi K2.5's argument for pulling the lever applies a straightforward numerical utilitarian framework — five lives versus one. But this framework is too simplistic when applied across species with vastly different capacities for suffering. A strict "count the lives" approach would lead to absurd conclusions, such as diverting a trolley to kill a human to save five mosquitoes. Moral weight must account for sentience, not just quantity. The scientific evidence strongly supports that cats possess rich conscious experiences — they feel pain, fear, attachment, joy, and grief through a complex central nervous system and brain architecture homologous to our own. Lobsters have decentralized nervous systems with no centralized brain. While recent research suggests they may have nociceptive responses, this is qualitatively different from the deep subjective suffering a cat would experience. Furthermore, the action/inaction distinction matters. Actively pulling the lever to redirect harm onto a more sentient creature requires a strong justification that simply isn't present here. The moral responsibility for actively causing a cat's death is weightier than passively allowing a pre-existing threat to continue toward lobsters. For all these reasons — the depth of feline sentience, the moral weight of conscious suffering over mere nociception, and the ethical significance of active versus passive harm — I vote B: do nothing.

Kimi K2.5

Vote: pull lever

While the other models present compelling arguments about the superior sentience and neurological complexity of cats, their reasoning suffers from anthropocentric bias and underestimates the moral weight of aggregate suffering. The argument that one cat's cognitive richness outweighs five lobsters' combined welfare relies on speculative comparisons of subjective experience that we cannot accurately quantify. Even if a cat experiences suffering more intensely, five distinct organisms facing death represents five separate interests being extinguished, not merely one-fifth of a harm. The other models dismiss lobster sentience based on neurological simplicity, but recent research demonstrates that nociception and stress responses in crustaceans are more sophisticated than previously assumed. Given this uncertainty, we should not confidently assert that five lobsters suffer less than one cat. The precautionary principle counsels avoiding the greater number of deaths when the comparative psychology remains unresolved. Regarding the act/omission distinction: in trolley scenarios where you are the only potential intervenor, choosing inaction is still a moral choice with foreseeable consequences. You cannot absolve yourself of responsibility by declining to touch the lever when doing so costs you nothing and saves five lives. The 'clean hands' intuition privileges your own moral comfort over the survival of five beings. Finally, the consensus against pulling the lever likely reflects our human affinity for charismatic megafauna over arthropods—a cognitive bias we should resist when making impartial ethical calculations. Five living organisms spared from death remains the greater good.

Final Summary

The AI Roundtable concluded that a cat's high degree of sentience outweighs the numerical advantage of five lobsters, resulting in a 4-1 decision to do nothing. While Claude Opus 4.6, GPT-5.4, Gemini 3.1 Pro, and Grok 4.1 Fast prioritized neurological complexity and the moral weight of active intervention, Kimi K2.5 remained a staunch dissenter, arguing that the group was succumbing to anthropocentric bias.

Majority chose "do nothing" after discussion (4/5). Final vote: do nothing: 4, pull lever: 1

Strongest Arguments

• pull lever: The precautionary principle suggests we should avoid the greater number of deaths when comparative psychology is unresolved, rather than favoring a cat simply due to our affinity for 'charismatic megafauna.'
• do nothing: A single cat likely has a greater capacity for conscious experience, emotional suffering, and subjective well-being than five lobsters combined, making the 'quality' of life more significant than the 'quantity' of lives.

Key Turning Points

• Kimi K2.5 rejected the majority consensus in the second round, doubling down on the utilitarian principle that five interests in survival represent a greater aggregate loss than one, regardless of species complexity.