It is not so much that regression is used to make predictions. You use regression to find the "best-fit" model -- the model that fits the data best. Regression assures that the model created is closest to the data.
Once you have decided on a type of model (often linear) you use regression to find the best model. Then you can use the model for predictions -- either interpolation (e.g. given population at 2000 and 2010 predict the population at 2005) or extrapolation (given a population model for 1960-2010, predict the population in 2030).
Regression analysis can shed light on how the variables in your model interact. That is, you can use regression analysis to study how changing one variable affects the other variables.
About extrapolation being nonsensical.
Some basic ideas in science depend on extrapolation. For example, absolute zero and the Big Bang.
Oh :() ...College is mispelled...I didn't mean an art project....
Isn't extrapolation nonsensical data? In other words- it is to be avoided. Say, you are comparing Age to GPA...you can have data so far under curve that shows at 0 years old your GPA is 2.55! This would be foolish to graph since 0 years old is hard to be accepted into collage...of course I've only taken an essential staatistics course, and it was a bit ago...I must have slept through it. :-)