Just like an English speaker sees "1, 2, 3, 4, 5, 6" and pronounces "one, two, three, four, five, six". Sure, the Roman numerals look like letters, but that doesn't mean you pronounce them that way.
In my opinion the important innovation of Hindu–Arabic numbers is not having different symbols for each digit, but rather using the same ten symbols (and no others!) for representing larger groups, which gives the system a lovely uniformity across numbers of different scales.
There aren’t many (if any?) human languages that duplicate this feature of being a purely positional system.
Just like an English speaker sees "1, 2, 3, 4, 5, 6" and pronounces "one, two, three, four, five, six". Sure, the Roman numerals look like letters, but that doesn't mean you pronounce them that way.