They're arguably less arbitrary than HTML; most of them are based in things that people have done/tend to do with plain text when they're trying to create a formatted document. That yields two benefits: the source document is inherently formatted -- transparently so, since even people who don't know what Markdown or the like is will be able to read it -- and creates something of a visual mnemonic for the formatting convention.

Not saying any one implementation is perfect, just that they have these nice properties.