@CyberneticForests Feed various trolley problems into a small language model to encode(?) so that other ML models can understand it (like CLIP I think), then train a model to solve trolley problems. Its training data will come from stuff like Absurd Trolley Problems. [1]

Sorry if this is nonsense, I don't know that much about the specifics of ML. But I think I've got very roughly the right idea.

[1] https://neal.fun/absurd-trolley-problems/