Both System-F and Martin-Löf type theory are roughly 40 years old as well.

What is wrong with type theory that it needs to be replaced every few decades, just because it got "old"?

Nothing is wrong with it, it's just as relevant today as it was 40 years ago, if not more. Most of the really interesting modern work is just building on top of some form intensional Martin-Löf theory.

I don't think there's anything implicitly wrong with a 40 year old type theory. Rather, C's type system is pretty decrepit.

This was a rhetorical question, it is clearly ridiculous to toss things out just because they are 40 years old

Because older type systems aren't webscale /s.

Please continue! Don't leave us hanging for the riveting conclusion :) I don't know ML so I would love to hear another perspective

Take a look at http://en.wikipedia.org/wiki/ML_%28programming_language%29

ML dates from the 70s and its type system is the inspiration for Haskell's, although Haskell also adds typeclasses (and various Haskell extensions add a bunch of other stuff).

The most widely used versions today are Standard ML and Ocaml, although many languages take inspiration from ML, for example Coq, Haskell, Clean, etc.

Just to add to the list: Scala takes a lot of nods from ML and F# is a nice variant if you're on .NET.