There are two very different types of arguments typically made as to why to learn the more esoteric languages (Haskell, Lisp, Forth, others).
The first is the more common: learning these languages exposes you to other ways of programming, which in turn makes you a better programmer no matter what language you're programming in. This is unquestionably true. Mainstream languages are constantly learning new things from these languages, which means you will be able to apply the techniques you learn from them to your code no matter what language you write in. I have the misfortune of writing a lot of PHP at my day job, but it turns out even PHP has support for filter and map over arrays. If I hadn't learned functional programming, I would never have even looked for them.
The other argument, which is made less often, is that you can actually write great software in these languages. This argument is made less often because we can obviously look at the market and the TIOBE index and see that Haskell is way down at #30 with 0.323% of the market share. However, there are really great programs made in Haskell (xmonad, for one). There are also tons of great programs made in Lisps, especially Clojure of late. So, this second argument is not entirely a dead end either.
Revisiting something I mentioned in the first argument for a moment though, I would actually make a third argument that I rarely see. As I said, other languages are constantly learning new things from these languages. Haskell's community is full of PhDs studying at the intersection of programming and mathematics, and they are discovering the things that will revolutionize programming in the future. Haskell, for example, has been exploring strong static typing with algebraic datatypes, Hindley-Milner type inference, and typeclasses to support higher-kinded polymorphism. Those concepts are turning out to be incredibly powerful, and you will inevitably see them in other programming languages in the future. We've seen this before with Lisp (garbage collection, recursion, first-class functions), and there's no reason not to expect the trend to continue.
So, why learn Haskell? It's the easiest way to see what tools you'll have available to you in industry 20 years from now when the other languages catch up.