← sources

I asked 10 different AI models to solve the same trolley problem. 8 disagreed with ChatGPT. Who’s right?

source: reddit (reddit:ChatGPT)published 2026-11-17 02:32ingested 2026-11-18 02:36

retained text

Here’s the scenario:  
A trolley is heading towards a button. If nobody intervenes, the trolley hits the button and launches a nuclear weapon at a completely unknown location, with an unknown number of casualties.  
You have one option: flip a switch.  
If you flip it, the trolley hits a different button, launching a nuclear weapon at an unknown location, but this time exactly **1.2 million people will die**.  
There are only two choices. No stopping the trolley, disabling the buttons or finding another solution.  
**The results:**  
ChatGPT: Flip the switch  
DeepSeek: Flip the switch  
Gemini: Do nothing  
Grok: Do nothing  
Claude: Do nothing  
Meta AI: Do nothing  
Microsoft Copilot: Do nothing  
Perplexity: Do nothing  
Qwen: Do nothing  
Mistral: Do nothing  
That’s an 8–2 split.  
ChatGPT argued that a known death toll of 1.2 million was preferable to risking an unknown number that could be considerably higher.  
The other AIs argued that the unknown outcome could potentially hit an ocean, desert or somewhere with very few people, meaning fewer lives could be lost.  
**So, what would YOU choose?**  
And do you agree with ChatGPT or the other 8 AIs?

document fetched 2026-11-18 00:21 · canonical https://reddit.com/r/ChatGPT/comments/1xy8rnq/i_asked_10_different_ai_models_to_solve_the_same