The main argument I have against 1-based indexing is that it makes a lot of indexing calculations, which all programmers will need to learn at some point, even more confusing with all the additional +/-1. In other words, 1-based is easier only for the easy cases, but harder for the hard cases.

In some ways I think it's similar to the endianness debate: big endian looks natural at first glance, but when you see the expression for the total value, having the place value increase with the address just makes far more sense since it eliminates the extra length-dependent term and subtraction.