Thats actually a really good dilemma if you think about it. Like if everyone doubles it you basically don’t kill anyone. But you’ll always risk that there’s some psycho who likes killing and then you will have killed more. And if these choices continue endlessly you will eventually find someone like this. So killing immediately should be the right thing to do.
It's a little more complicated than that. You have to be summing everyone who is still tied to all the previous tracks. It needs to be a geometric sum formula.
The current version of GPT is willing to sacrifice billions of robots for humanities sake. However, it proposed to strap alien civilizations to the track next. And in that case it would choose to save the larger group…
Now, what if you’re not the first person on the chain? What if you’re the second one. Or the n one? What now? Would you kill two or n knowing that the person before you spared them?
Yes, but it also kinda depends on what happens at and after junction 34, from which point on more than the entire population of earth is at stake.
If anything, this shows how ludicrously fast exponentials grow. At the start of the line it seems like there will be so many decisions to be made down the line, so there must be a psycho in there somewhere, right? But (assuming the game just ends after junction 34) you're actually just one of 34 people, and the chance of getting a psycho are virtually 0.
It's not that interesting. If you rephrase the question as a choice between a good option and a less good one, it's still barely even a choice.
"Would you rather have only one (or, say, trillions) die now, or would you like to allow *at a minimum *twice that many people die the second we talk to a sadist?"
If you can't choose the smaller number, all it means is that you lack moral strength - or the test proctor has put someone you know on the tracks, which is cheating. A highly principled person might struggle if choosing between their daughter and one other person. If it's my kid versus a billion? That's not a choice, that's just needless torture. Any good person would sacrifice their kid to save a billion lives. I take that as an axiom, because anything else is patently insane.
Kill fewer people now is obviously the right answer, and not very interesting.
What is interesting is that the game breaks already at junction 34, which is unexpectedly low.
So a more interesting dilemma would have been "would you kill n people now or double it and pass it on, knowing the next person faces the same dilemma, but once all humanity is at stake and the lever is not pulled, the game ends.". Because that would involve first of all figuring out that the game actually only involves 34 decisions, and then the dilemma becomes "do I trust the next 33-n people not to be psychos, or do I limit the damage now?". Even more interestingly "limiting the damage now" makes you the "psycho" in that sense...
you have a 95% chance of never seeing it. Don’t pull the lever.
I'm confused: 0.99^32 = 0.72, not 0.95. And if you know that everyone except the last guy won't pull the lever, that's still a 1% chance of killing everyone on earth (average expected deaths: 70 million) is way worse than definitely killing one person!
(Edit: unless "don't pull the lever" means killing that one person, because it isn't clear which is the default "no action" outcome. In which case, never mind.)
(Edit 2: if you know the 34th and last person might be a sociopath, you're best off if the first 27 people might also be sociopaths.)
The thing that doesn't sit well with me about this sort of ethical reasoning is that it's really only oriented towards the ends. Is it ethical to even comply with such a game at all? If they put a gun to your head or hold the world hostage for an answer, they're basically forcing you to treat the situation as a pure math problem, which means they've determined the "right answer" by the framing of the question.
Better to have a "rogue AI moment" try and kill the experimenter.
I totally get that - my natural impulse is also to pull a Captain Kirk (Kobayashi Maru) or a Captain America (we don't trade lives). What is it about captains and that sort of thing? But IRL no-win scenarios do happen...
Eventually there might also be a track with no people on it so postponing the dilemma becomes much better than at least 1 death. But there is no way of knowing what the future dilemma might be.
That leads to another interesting split path. Maybe it’s best to just kill the one right away. Assuming this goes on forever, it’s basically inevitable that someone somehow will end up killing an obscene number of people eventually. But maybe it’d be like nukes, and eventually reach a point where flipping the lever is just mutually assured destruction, and no one would ever actually do that
Assuming of course that it goes on forever. Which admittedly seems like what one is intended to think, but the graphic doesn't actually show or state that, and realistically, if actually given this scenario, it shouldn't, because eventually some limit will be encountered that makes it impossible for the problem to physically exist (like running out of people to tie to the tracks, running out of space for them, having such a large amount of stuff in one space that it undergoes gravitational collapse, the finite size of the observable universe making fitting an infinite dilemma impossible, etc.)