## Monday, October 17, 2022

### Probabilistic trolleys

Generally people think that if a trolley is heading for a bunch of people, it’s wrong to push an innocent bystander in front of the trolley to stop it before it kills the other people, with the innocent bystander dying from the impact.

But imagine that it is 99% likely that the bystander will survive the impact, but 100% certain that the five people further down the track would die. Perhaps the trolley is accelerating downhill, and currently it only has a 1% chance of lethality, but by the time it reaches the five people at the bottom of the hill, it has a 100% chance of lethality. Or perhaps the five people are more fragile, or the bystander is well-armored. For simplicity, let’s also suppose that the trolley cannot inflict any major injury other than death. At this point, it seems plausible that it is permissible to push the bystander in front of the trolley.

But now let’s suppose the situation is repeated over and over, with new people at the bottom of the track but the same unfortunate bystander. Eventually the bystander dies, and the situation stops (maybe that death is what convinces the railroad company to fix the brakes on their trolleys). We can expect about 500 people to be saved at this point. However, it seems that in the case where the bystander wasn’t going to survive the impact, it would have been wrong to push them even to save 500.

There are at least two non-consequentialist ways out of this puzzle.

1. It is wrong to push the bystander in front of the trolley in the original case where doing so is fatal. After all, one is not intending the bystander’s death, but only their absorption of kinetic energy. In my 2013 paper, I argued that this constitutes wrongful lethal endangerment when the bystander does not consent, even if it is not an intentional killing. But perhaps that judgment is wrong.

2. It is wrong to push the bystander to save five, but not wrong to push them to save five hundred. While this is a special case of threshold deontology, one can make this move without embracing threshold deontology. One can say that no matter how many are saved, it is wrong to intentionally kill the innocent bystander, but lethal endangerment becomes permissible once the number of people saved is high enough.

Initially, I also thought the following was an appealing solution: It matters whether it is the same bystander who is pushed in front of the trolley each time or a different one. Pushing the same bystander repeatedly unjustly imposes a likely-lethal burden on them, and that is wrong. But it would be permissible to push a different bystander each time onto the track, even though it is still almost certain that eventually a bystander will die. The problem with this solution is this. When the sad situation is repeated with different bystanders, by adopting the policy of pushing the bystander, we are basically setting up a lethal lottery for the bystanders—one of them will be killed. But if we can do that, then it seems we could set up a lethal lottery a different way: Choose a random bystander out of, say, 500, and then keep on pushing that bystander. (Remember that the way the story was set up, death is the only possible injury, so don’t think of that bystander as getting more and more bruised; they are unscathed until they die.) But that doesn’t seem any different from just pushing the same bystander without any lottery, because it is pretty much random which human being will end up being the bystander.