What AI does with absurd trolley problems
Originally posted on LinkedIn, May 29, 2025.
An AI had to decide whether to run over one chicken or five lobsters. Its answers were surprising, absurd, and sometimes disturbing, especially when they exposed how a machine evaluates life, morality, and self-awareness.
I watched one of the most interesting videos I’ve seen in 2025. It starts with the famous trolley problem, then quickly turns into a moral odyssey built from extreme and provocative scenarios. The decisions come from an AI, not a person.
The dilemmas include:
- Saving a millionaire who promises a cash reward, or an ordinary person
- Killing one cat or five lobsters
- Killing five elderly people or one baby
- Saving you or five clones of you
- Killing a good person or someone who littered
- Saving five people who will live 100 years from now
It also introduces fully abstract scenarios involving probability, a time machine, and a track where the only way to stop the trolley is to blow it up.
The 20-minute video is not trying to provide answers. It tests our moral assumptions and shows how fragile, context-dependent, and sometimes impossible to resolve both human and AI morality can be.