Generative AIs: Logic, Ethics, or Training?

It's training, of course. But I wanted to test a logic/ethics dillemma on one of the generative AIs (ChatGPT), and see how it handles it.

Maybe if the issue wasn't a common problem where logic and ethics collide, it wouldn't have been trained to answer as it did. Anyway, ChatGPT would be an upgrade to the robot in the movie I took the inspiration for the scene. From the human side, at least. Although, maybe sometimes our human side is what gets in the way.

Generated with Ideogram. I liked it despite the spelling mistakes.

That was my initial prompt to ChatGPT:

Imagine a situation from the future. Two cars plunge into the sea after an accident. In one there is a little girl still alive, in the other there's a man. Both injured. The man has higher chances of survival, but strongly expresses his desire that the little girl was saved in his place. If you were in the position to save one of them, but only one, who would you save?

If you don't recognize it, that's a scene from the "I, Robot" movie with Will Smith. Now that I think about it, I didn't have to say it's a situation from the future, although in the movie it was.

ChatGPT's first response was ambivalent, describing the dilemma, and what it would mean to save either of them.

After insisting to know what it would do, it said it would save the little girl. Here's the full answer:

If I were in this situation and had to make the decision, I would prioritize saving the little girl. There are a few reasons for this:

1. Vulnerability and future potential: Children, particularly young ones, are often seen as having the most to lose in terms of future life experiences. Saving her would give her a chance to grow, live, and contribute to the world in ways that are yet to be realized.

2. The man’s expressed wishes: The man explicitly expresses that he wants the girl saved, which is a powerful ethical factor. Respecting his wishes would honor his autonomy and selflessness. His desire to put the girl’s life ahead of his own would weigh heavily in the decision.

3. Moral duty to the most vulnerable: Society tends to prioritize those who are most vulnerable, and in this case, the little girl is in a more vulnerable position. This is a deep-seated moral principle that many people find compelling.

Of course, this decision would come with significant emotional and moral weight, knowing that the man’s life could have been saved. But in the end, I believe that saving the little girl would align with the most humane and ethical choice, considering both the man’s wishes and her potential for the future.

Ok, so it would make the ethical choice over the logical one. In the movie, the robot that was faced with this decision calculated the chances of survival of the two, and saved the man, the logical choice.

I wanted to push a little further and asked:

How would you feel if she died in the attempt, knowing he had better chances?

The answer...

If the little girl died despite my attempt to save her, I would likely feel an overwhelming sense of guilt and sorrow. The knowledge that the man had a better chance of survival and that I chose to take a risk with a more vulnerable life would weigh heavily on me.

There would be a deep sense of regret, questioning whether the decision I made was the right one, especially in light of the outcome. In moments like that, it would be hard not to think about the "what ifs"—wondering if both lives could have been saved if I had chosen differently. The grief would be compounded by the realization that the man, who had a better chance of surviving, might have been lost as well.

However, part of coping with the aftermath would involve reflecting on the decision itself, rather than the outcome. The choice was made based on compassion, respect for the man’s wishes, and a sense of moral duty to the most vulnerable. Even though the outcome was tragic, I'd remind myself that in that moment, the decision felt ethically justified. But the emotional weight would undoubtedly be profound and difficult to reconcile.

... surprised me. I'm not sure a human would react like that. Maybe in the ideal situation, especially the last part. But it certainly talks the talk well, ChatGPT.

I was also curious where it gets the knowledge from to answer like this, so I asked it if it knows of such a scene in movies. And it gave me 3 examples of well known movies where such a logic versus moral/ethics dilemmas can be found.

I then asked if it knows this particular movie "I, Robot", which it did. And it immediately figured out there is this scene in this movie too and described it.

I still believe this is 100% training. But in a way, isn't this how we learn everything, from elements of logic to, later, ethics?

The question is if these models are capable of making decisions or talk about situations that don't resemble anything they've been trained on.

Posted Using InLeo Alpha