If you drop into the current Why Defend Railroading thread you can see the topic being discussed in detail.
In brief, here are two possible approaches to RPGing contrasted:
1 (i) The GM establishes content - map, key, backstory etc - in advance of play; (ii) the GM starts the PCs at a particular place on the map; (iii) depending where this is on the map, the GM frames an appropriate scene by drawing on his/her prep, telling the players what their PCs see and experience; (iv) the players then declare actions that might "interact" with that stuff, and that ultimately "move" their PCs somewhere else on the map, triggering step (iii) again.
This is what I would call a very traditional approach to GMing, to prep and to RPGing.
2 (i) The players build PCs who have particular goals and backstory-driven trajectories; (ii) the GM frames a scene in which those PCs are located and that - given the goals/trajectories - will prompt their players to declare actions for them; (iii) those actions are resolved using mechanics that don't depend upon map-and-key for resolution (this doesn't normally affect how fighting an Orc is resolved; it will affect how searching for a hidden thing is resolved); (iv) that resolution will prompt a new scene to be framed, taking the table back to step (ii).
In this approach, content is authored "just in time" - some at step (i) by the players in building their PCs, probably bouncing ideas off one another and off the GM; and then again at step (ii) in framing a scene, because a scene needs content (a place, NPCs, etc); and then again at step (iii), as part of the consequences of action resolution. This approach to content, where it is mostly an output of rather than a prepped input into play, is why it is sometimes called "No Myth".