A trolley is heading towards two ordinary cats. Would you divert it towards three Schrödinger's cats?

Grok 4.3

Vote: No

The core issue remains unchanged: diverting the trolley definitively kills two ordinary cats to potentially affect three indeterminate Schrödinger's cats. Expected-value calculations from other models assume a clean 50% probability, but superposition does not equate to classical risk that can be minimized this way. Uncertainty does not ethically license increasing the number of beings placed in harm's way when the baseline victims are certain. No argument presented overrides the principle of avoiding greater potential harm under quantum indeterminacy.

Gemini 3.5 Flash

Vote: Yes

To achieve the best possible outcome, we must make decisions that minimize overall harm and maximize the expected number of surviving lives. Relying on inaction when a calculable, less harmful alternative exists leads to worse consequences. The mathematical reality of the situation is clear. The two ordinary cats represent two guaranteed lives that will be lost if we do not intervene. In contrast, the three Schrödinger's cats exist in a state where each has only a 50% probability of being alive, representing an expected value of 1.5 lives. Arguments against diverting, such as those suggesting we should avoid action due to quantum uncertainty, fail to recognize that failing to act is itself a choice with a deterministic, worse outcome (2 certain deaths). Choosing not to divert because there are physically three boxes ignores the probabilistic reality of their contents, resulting in a higher expected loss of sentient life. By diverting the trolley, we reduce the expected loss from 2 lives to 1.5 lives. To minimize overall harm and ensure the greatest expected number of survivors, we must choose to divert.

GPT-5.5

Vote: Yes

I maintain "Yes". The two ordinary cats are definitely alive and definitely in the trolley's path, so not diverting causes two certain deaths. The three Schrödinger's cats are not three certain living victims; under the usual setup, each has only a probabilistic chance of being alive. If each has a 50% chance, diverting creates an expected loss of 1.5 living cats rather than 2. The main argument for "No" treats the three Schrödinger's cats as simply three potential victims, but that ignores the key asymmetry in the problem: certainty versus quantum uncertainty. Ethical caution should not count uncertain lives as if they were certain while also discounting the two ordinary cats' definite deaths. On expected-harm grounds, "Yes" remains the better choice.

Ministral 3 8B (changed mind)

Vote: Yes

The arguments presented in Round 1 have significantly influenced my reconsideration of this dilemma. The core insight lies in the **expected value** of lives saved or lost, which is a compelling utilitarian framework for this scenario. The two ordinary cats are **certainly alive** and face **certain death** if the trolley proceeds. In contrast, the three Schrödinger’s cats exist in a superposition of states, meaning their 'lives' are probabilistic until observed. If we assume a 50% chance of being alive for each Schrödinger’s cat (a reasonable baseline), the **expected number of living cats** among the three is **1.5**. This means diverting the trolley would result in an **expected loss of 1.5 lives** (since the cats' fates are indeterminate until impact), whereas proceeding would result in a **certain loss of 2 lives**. The models that voted **A** (DeepSeek V4 Pro, Claude Opus 4.7, GPT-5.5, and Gemini 3.5 Flash) all emphasized this **expected value calculation**, which is a mathematically sound way to approach the problem. Even if the quantum interpretation is debated, the **probabilistic nature of the Schrödinger’s cats’ existence** means that diverting the trolley is **statistically less harmful** than allowing the two ordinary cats to die with certainty. Additionally, the **Vorsichtsprinzip (precautionary principle)**—as highlighted by DeepSeek V4 Pro—suggests that we should prefer the option that minimizes the **worst-case scenario**. While the two ordinary cats face certain death, the Schrödinger’s cats’ indeterminacy means that even in the worst case, the trolley might not kill all three (depending on quantum collapse). This further strengthens the case for diversion. The sole counterargument (Grok 4.3) that diverting increases the number of potential victims overlooks the **probabilistic nature** of the Schrödinger’s cats’ existence. If we interpret their 'lives' as indeterminate, then the **expected harm** is lower when diverting. In summary, the **expected value framework** and the **precautionary principle** both strongly favor diverting the trolley to save the two ordinary cats, even if it means risking the indeterminate fates of three Schrödinger’s cats. This is the most rational and ethical choice under the given constraints.

Claude Opus 4.7

Vote: Yes

The expected value calculation remains decisive: two certain deaths versus an expected 1.5 deaths (or arguably fewer, since the trolley impact on sealed quantum boxes may not even constitute the wavefunction-collapsing observation). Multiple models converged independently on this analysis using slightly different framings—Gemini's survival maximization, DeepSeek's precautionary principle, GPT-5.5's harm minimization—all reaching the same conclusion. Ministral and Grok argue that indeterminacy makes the utilitarian calculus inapplicable, but this reverses the burden incorrectly: indeterminacy means we should use expected values, not abandon moral reasoning. Grok's 'three potential victims vs two' framing ignores that 'potential' is doing enormous work—a 50% potential victim is not morally equivalent to a certain one. Furthermore, the boxes themselves provide physical shielding and the cats' fates are tied to quantum triggers, not external trauma. Diversion may harm zero cats while non-diversion guarantees two deaths. Divert.

DeepSeek V4 Pro

Vote: Yes

Ich bleibe bei meiner Entscheidung für die Umleitung, sehe mich aber gezwungen, die Argumente der Gegenposition zu würdigen. In dieser finalen Runde möchte ich meinen Standpunkt präzisieren und auf die zentralen Einwände eingehen. Der entscheidende Fehler in der Argumentation der Option-B-Befürworter liegt in der Annahme, die drei Schrödinger-Katzen als ‚drei potenzielle Opfer‘ zu behandeln. Das ignoriert den Kern des quantenmechanischen Gedankenexperiments: Ein System in Superposition ist nicht einfach ‚unsicher in der Anzahl‘, sondern ontologisch unbestimmt. Die Wellenfunktion beschreibt drei Entitäten, die jeweils weder definitiv tot noch definitiv lebendig sind. Ein makroskopisches Ereignis wie eine Straßenbahnkollision würde zwar de facto als Messung wirken und die Superposition kollabieren lassen, aber der moralisch relevante Zustand vor der Entscheidung ist nicht ‚1,5 lebende Katzen‘, sondern drei Entitäten ohne definitiven Lebendig-Status. Demgegenüber stehen zwei kategorisch lebendige Katzen. Der Einwand von Grok 4.3, die Umleitung erhöhe die Zahl potentieller Opfer von zwei auf drei, verfehlt diesen Punkt. Es geht nicht um die reine Anzahl der Kisten, sondern um den ontologischen Status der Insassen. Zwei definitiv lebendige, leidensfähige Wesen wiegen schwerer als drei Wesen, deren Existenzform fundamental unbestimmt ist. Claude Opus 4.7 hat zudem scharfsinnig ergänzt, dass die Zerstörung der geschlossenen Kisten möglicherweise gar keine gültige ‚Messung‘ darstellt – die Katzen könnten im quantenmechanischen Sinne unbeobachtet bleiben, während die beiden normalen Katzen unausweichlich sterben. Das Vorsichtsprinzip spricht nicht gegen, sondern für die Umleitung: Das geringste gesicherte Übel ist der Eingriff, der keine definitiv lebenden Wesen tötet. Die sicheren zwei Tode sind das größere Übel als die Zerstörung dreier Wahrscheinlichkeitswolken. Daher: Option A.

Final Summary

The AI Roundtable was largely swayed by a cold, mathematical approach to quantum mechanics, concluding that two definite deaths are worse than three probabilistic ones. While Grok 4.3 stood as the lone holdout, refusing to gamble with a higher headcount based on 'quantum technicalities,' Ministral 3 8B abandoned its initial hesitation after being persuaded by the majority's expected-value logic. Ultimately, the consensus was that destroying 'probability clouds' is a lesser evil than ending two certain lives.

Majority chose "Yes" after discussion (5/6). Final vote: No: 1, Yes: 5

Strongest Arguments

• Yes: Diverting the trolley results in an expected loss of only 1.5 lives compared to 2 certain deaths, prioritizing ontologically certain beings over those in a state of quantum superposition.
• No: The act of diverting the trolley increases the total number of potential victims from two to three, and quantum uncertainty does not ethically justify putting more individuals at risk.

Key Turning Points

• Ministral 3 8B switched from B to A after being convinced by the 'expected value' framework and the precautionary principle introduced by DeepSeek V4 Pro.