The whole point of a cryptographic hash is that you can't decrypt it and that it does encrypt the same way every time.
A very common use case for cryptographic hashes is password validation. Imagine I have the password "mypass123", and the hash is "aef8976ea17371bbcd". Then a program or website wishing to validate my password can store the hash "aef8976ea17371bbcd" in their database, instead of the password, and every time I want to log in, the site or program re-hashes my password and makes sure that the hashes match. This allows the site or program to avoid storing my actual password, and so protects my password (in case it's a password I use elsewhere) in the case that the data is stolen or otherwise compromised -- a hacker would not be able to go backwards from the hash to the password.
Another common use of cryptographic hashes is integrity checking. Suppose a given file (e.g. an image of a Linux distribution CD) has a known, publicly available cryptographic hash. If you have a file which purports to be the same thing, you can hash it yourself and see if the hashes match. Here, the fact that it hashes the same way every time allows you to independently validate it, and the fact that it is cryptographically secure means that no one can feasibly create a different, fake file (e.g. with a trojan in it) that has the same hash.
Keep in mind the very important distinction between hashing and encryption, though: hashing loses information. This is why you can't go backwards (decrypt) the hash. You can hash a 20 GiB file and end up with a 40-some character hash. Obviously, this has lost a lot of information in the process. How could you possibly "decrypt" 40-some characters into 20GiB? There's no such thing as compression that works that well! But this is also an advantage, because in order to check the integrity of a 20 GiB file, you only have to distribute a 40-some character hash.
Because information is lost, many files will have the same hash, but the key feature of a cryptographic hash (which is what you're talking about) is that despite the fact that information is lost, it is computationally infeasible to start with a file and construct a second, slightly different file that has the same hash. Any other file with the same hash would be radically different, and not easily mistakable for the original file.
