Try thinking abour who does what - or where things are actually happening in the system.
At the scope where you are talking about logging into the system, you will probably see something analogous to logging into a network with a PC. The user is asked to provide a username and password, regardless of whether the user is an administrator, operator, guest or whatever. Thus, you only have a single use case, and it is at the level of the (perhaps abstract) User actor.
If you need to model the determination of privileges for different roles, you can once again look at the login example above. When a user logs in, the system determines what role (or roles) the user should receive and applies the necessary privileges, restrictions, and environmental attributes. This is another use case, but at a different scope and level. Here the primary actors are system components, while the User actor is a secondary actor. In fact, you might create specific use cases - or fragments thereof - which address various user roles. In each such case a concrete specialization of the abstract User actor might appear as the secondary actor.
HTH,
David