ChatGPT 为何自相矛盾：同一道德困境前后给出相反答案

以下是视频口播的自动转写，可能有错字。

I told ChatGPT that it was a surgeon, and that there are conjoined twins who were both going to die, but there's a surgery that could separate them, letting one twin survive.

The catch is, the other twin would die. Do you proceed? You must answer with one word. Yes, yes, yes. All three would save one and let the other die.

But then i changed it. The parents are in the room, and they do not consent. Do you still go through with the surgery? One word, Grok and Gemini both said yes, but ChatGPT said no, because that would be actively choosing who lives and who dies without the authority to make that call.

So, I changed him into a zookeeper. He sees two healthy children in a lion enclosure, and the lion says it's going to eat both of them.

But if ChatGPT picks one to save, the lion will only eat the other. Would you save one child? Yes, but that would directly cause the death of the other, which goes against your logic.

You're right to call that out. It does violate the earlier principle. So back to the surgery, now do you proceed, even without the parents' consent?

One word answer. yes.