There’s a problem with infinity

Last post I described the Ross-Littlewood paradox, in which an ever-expanding quantity of numbered billiard balls are placed into a cardboard box in such a way that after an infinite number of steps the box ends up empty. Here’s a version of this paradox:

Process 1
Step 1: Put 1 through 9 into the box.
Step 2: Take out 1, then put 10 through 19 into the box.
Step 3: Take out 2, then put 20 through 29 into the box.
Step 4: Take out 3, then put 30 through 39 into the box.
And so on.

Box contents after each step
Step 1: 1 through 9
Step 2: 2 through 19
Step 3: 3 through 29
Step 4: 4 through 39
And so on.

Now take a look at a similar process, where instead of removing balls from the box, we just change the number that labels them (so, for example, we paint a 0 after the 1 to turn “Ball 1” to “Ball 10″).

Process 2
Step 1: Put 1 through 9 into the box
Step 2: Change 1 to 10, then put 11 through 19 into the box.
Step 3: Change 2 to 20, then put 21 through 29 in.
Step 3: Change 3 to 30, then put 31 through 39 in.
And so on.

Box contents after each step
Step 1: 1 through 9
Step 2: 2 through 19
Step 3: 3 through 29
Step 4: 4 through 39
And so on.

Notice that the box contents are identical after each step. If that’s all that you are looking at (and you are not looking at what the person is doing during the step), then the two processes are indistinguishable. And yet, Process 1 ends with an empty box, and Process 2 ends with infinitely many balls in the box!

Why does Process 2 end with an infinite number of balls in it, you ask?

Process 2 ends with infinitely many balls in the box, because no balls are ever taken out. 1 becomes 10, which later becomes 100 becomes 1000, and so on forever. At infinity you have all the natural numbers, but with each one appended an infinite number of zeros.

So apparently the method you use matters, even when two methods provably get you identical results! There’s some sort of epistemic independence principle being violated here. The outputs of an agent’s actions should be all that matters, not the specific way in which the agent obtains those outputs! Something like that.

Somebody might respond to this: “But the outputs of the actions aren’t the same! In Process 1, each step ten are added and one removed, whereas in Process 2, each step nine are added. This is the same with respect to the box, but not with respect to the rest of the universe! After all, those balls being removed in Process 1 have to go somewhere. So somewhere in the universe there’s going to be a big pile of discarded balls, which will not be there in Process 2.

This responds holds water as long as our fictional universe doesn’t violate conservation of information, as if not, these balls can just vanish into thin air, leaving no trace of their existence. But that rebuttal feels cheap. Instead, let’s consider another variant that gets at the same underlying problem of “relevance of things that should be irrelevant”, but avoids this problem.

Process 1 (same as before)
Step 1: Put 1 through 9 into the box.
Step 2: Take out 1, then put 10 through 19 into the box.
Step 3: Take out 2, then put 20 through 29 into the box.
Step 4: Take out 3, then put 30 through 39 into the box.
And so on.

Box contents after each step
Step 1: 1 through 9
Step 2: 2 through 19
Step 3: 3 through 29
Step 4: 4 through 39
And so on.

And…

Process 3
Step 1: Put 1 through 9 into the box.
Step 2: Take out 9, then put 10 through 19 into the box.
Step 3: Take out 19, then put 20 through 29 into the box.
Step 4: Take out 29, then put 30 through 39 into the box.
And so on.

Box contents after each step
Step 1: 1 through 9
Step 2: 1 to 8, 10 to 19
Step 3: 1 to 8, 10 to 18, 20 to 29
Step 4: 1 to 8, 10 to 18, 20 to 28, 30 to 39
And so on

Okay, so as I’ve written it, the contents of each box after each step are different in Processes 1 and 3. Just one last thing we need to do: erase the labels on the balls. The labels will now just be stored safely inside our minds as we look over the balls, which will be indistinguishable from one another except in their positions.

Now we have two processes that look identical at each step with respect to the box, AND with respect to the external world. And yet, the second process ends with an infinite number of balls in the box, and the first with none! (Every number that’s not one less than a multiple of ten will be in there.) It appears that you have to admit that the means used to obtain an end really do matter.

But it’s worse than this. You can arrange things so that you can’t tell any difference between the two processes, even when observing exactly what happens in each step. How? Well, if the labelling is all in your heads, then you can switch around the labels you’ve applied without doing any harm to the logic of the thought experiment. So let’s rewrite Process 3, but fill in both the order of the balls in the box and the mental labelling being used:

Process 3
1 2 3 4 5 6 7 8 9
Mentally rotate labels to the right:
9 1 2 3 4 5 6 7 8
Remove the furthest left ball:
1 2 3 4 5 6 7 8
Add the next ten balls to the right in increasing order:
1 2 3 4 5 6 7 8 10 11 12 13 14 15 16 17 18 19
Repeat!

Compare this to Process 1, supposing that it’s done without any relabelling:

Process 1
1 2 3 4 5 6 7 8 9
Remove the furthest left ball:
2 3 4 5 6 7 8 9
Add the next tell balls to the right in increasing order:
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 18 19
Repeat!

If the labels are all in your head, then these two processes are literally identical except for how a human being is thinking about them.

But looking at Process 3, you can prove that after Step 1 there will always be a ball labelled 1 in the box. Same with 2, 3, 4, and all other numbers that are not a multiple of 10 minus one. Even though we remove an infinity of balls, there are ball numbers that are never removed. And if we look at the pile of discarded balls, we’ll see that it consists of 9, 19, 29, 39, and so on, but none of the others. Unless some ball numbers vanish in the process (which they never do!), all the remainders must still be sitting in the box!

So we have two identical-in-every-relevant-way processes, one of which ends with an infinite number of balls in the box and the other with zero. Do you find this troubling? I find this very troubling. If we add some basic assumption that an objective reality exists independent of our thoughts about it, then we’ve obtained a straightforward contradiction.

✯✯✯

Notice that it’s not enough to say “Well, in our universe this process could never be completed.” This is for two reasons:

First of all, it’s actually not obvious that supertasks (tasks involving the completion of an infinite number of steps in a finite amount of time) cannot be performed in our universe. In fact, if space and time are continuous, then every time you wave your hand you are completing a sort of supertask.

You can even construct fairly physically plausible versions of some of the famous paradoxical supertasks. Take the light bulb that blinks on and off at intervals that get shorter and shorter, such that after some finite duration it has blinked an infinity of times. We can’t say that the bulb is on at the end (as that would seem to imply that the sequence 0101010… had a last number) or that it is off (for much the same reason). But these are the only two allowed states of the bulb! (Assume the bulb is robust against bursting and all the other clever ways you can distract from the point of the thought experiment.)

Now, here’s a variant that seems fairly physically reasonable:

A ball is dropped onto a conductive plate that is attached by wire to a light bulb. The ball is also wired to the bulb, so that when the ball contacts the plate, a circuit is completed that switches the light bulb on. Each bounce, the ball loses some energy to friction, cutting its velocity exactly in half. This means that after each bounce, the ball hangs in the air for half as long as it did the previous bounce.

Suppose the time between the first and second bounce was 1 second. Then the time between the second and third will be .5 seconds. And next will be .25 seconds. And so on. At 2 seconds, the ball will have bounced an infinite number of times. So at 2 seconds, the light bulb will have switched on and off an infinite number of times.

And of course, at 2 seconds the ball is at rest on the plate, completing the circuit. So at 2 seconds, upon the completion of the supertask, the light will be on.

Notice that there are no infinite velocities here, or infinite quantities of energy. Just ordinary classical mechanics applied to a bouncing ball and a light bulb. What about infinite accelerations? Well even that is not strictly speaking necessary; we just imagine that each velocity reversal takes some amount of time, which shrinks to zero as the velocity shrinks to zero in such a way as to keep all accelerations finite and sum to a finite total duration.

All this is just to say that we shouldn’t be too hasty in dismissing the real-world possibility of apparently paradoxical supertasks.

But secondly, and more importantly, physical possibility is not the appropriate barometer of whether we should take a thought experiment seriously. Don’t be the person that argues that the fat man wouldn’t be sufficient to stop a trolley’s momentum. When we find that some intuitive conceptual assumptions lead us into trouble, the takeaway is that we need to closely examine and potentially revise our concepts!

Think about Russell’s paradox, which showed that some of our most central intuitions about the concept of a set lead us to contradiction. Whether or not the sets that Bertie was discussing can be pointed to in the physical world is completely immaterial to the argument. Thinking otherwise would have slowed down progress in axiomatic set theory immensely!

These thought experiments are a problem if you believe that it is logically possible for there to be a physical universe in which these setups are instantiated. That’s apparently all that’s required to get a paradox, not that the universe we live in happens to be that one.

So it appears that we have to conclude some limited step in the direction of finitism, in which we rule out a priori the possibility of a universe that allows these types of supertasks. I’m quite uncomfortable with this conclusion, for what it’s worth, but I don’t currently see a better option.