The problem you're having is that the selected checkbox values are already a list, but the split function creates another list so the indexing can't really work on both lists cleanly.

One way around this is to flatten/delimit the list first, then use the Substitute function of the Token Calculator activity on the result to avoid having to split and join things back together.

Step 1

Flatten the list of raw values with your desired delimiter

Step 2

Replace the underscores in that flattened value using substitute

You could probably combine these into a single step by having the formatted checkbox token directly in the substitute, but I usually keep things like this separate to make it more readable and to emphasize the fact that you have to flatten the list before using the substitution.