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?
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:
A B C D A B C D A B C D
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?