Suppose you have a trolley going through track A, and if it continues on this way it will kill 2 people. If you switch the track to B, it will destroy a computer that is currently simulating a brain, which is perfectly conscious. What should we do? Now imagine that instead of simulating one brain, that computer is simulating thousands of them, but with one catch, they are the same simulation and are indistinguishable, they receive the same inputs and give the same outputs, as they work in a deterministic way. We know for sure that they will never differ in the future. Does it change our choice? If yes, why?

  • 1
    You can just re-run the sims in another computer. There's no moral evil attached to restarting a sim any more than there is to rebooting your computer, exactly for this reason. You can restart the simulation on different hardware and you can even start it at the same point so that the "consciousness" experiences perfect continuity of their simulated existence.
    – user4894
    Nov 14, 2021 at 19:36
  • @user4894 Then let's add the caveat that this computer is the only one that contains the simulation data, and in general just assume that all the simulated people "die", or stop existing, if the computer is destroyed. And it will be if the trolley goes to track B.
    – Pingou
    Nov 14, 2021 at 19:54
  • 1
    But that's nonsense. If your computer program crashes you just start it again. What's different about the computer programs running the sims? Whether the computer is running a word processor or running a sim, how can the semiconductors know the difference? You're claiming without evidence that there's some difference between one program and another.
    – user4894
    Nov 14, 2021 at 20:41
  • @user4894 I am a bit confused about your comment? I said the computer would be destroyed, and the data would be lost, it doesn't matter if the computer is running a simulation or it runs Microsoft Word 2052, your document or your simulation would be lost. And I really don't see how that is really relevant to my question as it is in my opinion a detail, it is as if I was talking about the original trolley problem and you told me that there is no issue because you could run to the people on the track and warn/untie them.
    – Pingou
    Nov 14, 2021 at 21:36
  • 2
    I have a separate question. What is the intent of running a thousand copies of the same simulation? What aspect of the ethical problem is that intended to bring out?
    – user4894
    Nov 15, 2021 at 8:59

3 Answers 3


I fail to see the utility of such a hypothetical question. It seems to be nothing more than abstract reflection on an imagined circumstance. Is it a heuristic meant to refine our moral sensibility or test our mental agility? Maybe it has a significance not apparent to me. I fear this is a rather flat-footed and perhaps inappropriate response, but I am theoretically receptive to correction.


There isn't any moral difference between simulating one brain, or simulating a thousand copies of one brain in perfect lockstep. We can view the second as a special case of the first, or the first as a special case of the second. This is because there is a direct correspondence between the first case and the second. A simulation of one brain is also a simulation of a thousand copies of the brain in lockstep, in the sense that we can use the simulation of one brain to derive the state of the simulation of the thousand copies. This is what it means for something to be a simulation. And vice versa; we can use the lockstep thousand-brain simulation to derive the state of the single-brain simulation. They are functionally equivalent.

Think about how the variables are laid out in memory. Let's imagine (for simplicity) there are four variables A, B, C, and D involved in the simulation, representing the state of the brain. Say we have three simulations, each of which has its own brain state. These 3x4 = 12 total variables can be laid out in memory like so:


Next we may perform an update, replacing the value of A with A' on the next time step in each simulation, so that memory changes like this:

A B C D A B C D A B C D ==> A' B C D A' B C D A' B C D

Here, the three sets of variables look independent, in three different simulations. But what if we rearrange the order of the variables, grouping them by variable instead of by simulation?

A A A A B B B B C C C C D D D D ==> A' A' A' A' B B B B C C C C D D D D

Now we haven't changed anything significant about the program - we're still running three simulations. But now the A's appear together, as do the B's and C's. It becomes more clear that the A's are redundant - why do we have three of them, all with the same values? That's simply inefficient. We could optimize our program so that all three simulations use the same variable for A, still running three simulations. Functionally we haven't changed anything essential about the simulations. Now the update for all three simulations looks like:

A B B B C C C D D D ==> A' B B B C C C D D D

And we can do the same for B:

A B C C C D D D ==> A' B C C C D D D

Still we have not fundamentally changed anything about our program; we're still running three simulations, still giving the same results. Now for C and D:

A B C D D D ==> A' B C D D D
A B C D ==> A' B C D

Still we have not fundamentally changed our program. We're still running "three" simulations, we've just optimized the memory layout for them, step by step. But now it is clear that the "three" are actually just one simulation.


Your problem setup implies that the humans at risk can be duplicated which significantly reduces the stakes and implies that the switcher has no free will. But I don't think it really changes anything about the problem.

You could choose to measure the ethics by looking at the consequences, "Is one of the victims a future Hitler?" or from the utility, "Is one of the victims a millionaire about to donate to kitten rescue" or for the benefit of science/art "Is that the first and only brain simulator?", "Is the computer about to run out of batteries anyway?" etc, all the usual things.

I understood the accepted "correct" answer is to do whatever and when questioned say the problem was too difficult to solve, so you made your best guess?

  • Well some compatibilists would say that the switcher has free will, even if the world is deterministic. But that's not really what I'm interested in here. From an utilitarianism point of view, if we try to maximise happiness, should we save the thousands of identical simulations, or not because they are somehow all the same person? It's unclear to me.
    – Pingou
    Nov 14, 2021 at 19:52
  • hmm, I guess maximize happiness kind of breaks down if you can copy and paste happy humans. But presumably the simulated humans don't interact with the world of the switcher. Is there an unspoken rule to utility that you only value happiness in your society? Do you kill a million ants or one human?
    – Ewan
    Nov 14, 2021 at 21:47
  • I mean to really maximize happiness you could put the simulated humans in a loop of their happiest moment. forever....
    – Ewan
    Nov 14, 2021 at 21:52
  • I think i would argue that the simulated humans have no utility of their own in the eyes of the switcher. Not because they are simulated, but because in order to remain perfect copies they must not become aware of each other which means they cant really interact with the outside world. If it was a simulated human society, like a tiny ant world that could talk to you. I think you would have to assign them utility in your society. But then you couldn't duplicate them.
    – Ewan
    Nov 14, 2021 at 21:59
  • I feel like it would be possible for them to interact with the outside world, and from the outside world they look like one person, also they can be aware of the copies, but they would not be able to communicate between themselves since they always have the exact same thoughts/sensory inputs.
    – Pingou
    Nov 14, 2021 at 22:04

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .