You don't need it. If you are a Junior developer (or aiming at being one) Haskell is a language you most definitely don't need.
Now, once you have your feet wet, it's worth learning. Here's why:
- Pure(mostly) functional language. This is in opposition to Java & Ruby, which while having functional aspects are OO (Object Oriented).
- It's okay not to know what a functional language is. It's making functions first-class citizens, such that a function becomes a value like any other.
- Separates IO from computation. IO is inherently impure and unreliable, so functions are generally pure and instead wrap IO in containers. Obviously, programs would be less useful (still use in the compiler running!) without IO.
- All of these concepts (and more) will change how you think about programming and make you a better programmer in all languages.
Haskell can be found in some backend stacks, the more notable examples I can think of are Mailrank (acquihired by Facebook) and Bump.
Personally, I've only done a limited number of projects with Haskell (a Postgres-backed beer API & IRC bot) but I've grown to respect it as a language and through limited exposure has made me love FP.
Functional programming is really more than first-class functions. A lot of languages have support for first-class functions (including Ruby's procs) that are not really functional languages. Functional (like all paradigms) is a bit nebulous, but to really be a functional language I would expect to see immutability as a default assumption, and I would expect to see only expressions, no statements. That latter expectation is really what gives functional programming its distinctive style. When everything returns a value, chaining together functions as transformations on data simply becomes the natural way to program.
To amplify what another has said, the emphasis of functional programming is side-effect-free code, where the default assumption is that functions don't have side effects, and so can be treated as components in a pipeline that modifies data step-by-step from its initial input state to its final output state.
Ideally, these components are composable and reusable, both to facilitate reasoning about the code and code reuse. The Haskell Standard Prelude is a fantastic example of this working out well.
Basically, it's based on the idea that the functions which are easiest to reason about are those with one well-defined input and one well-defined output which is completely determined by that input. You can't always have this, but functional languages attempt to get you as close as possible. The simplicity of pipeline-style programming is one of the happy little dividends of this kind of thinking.