Evolution, in the Darwinian sense, can be broadly described as the interaction of three processes: inheritance, mutation and selection.
Of those three processes, mutation is the one that produces novel innovation, by occasionally producing new alleles that did not exist in the parent population. It is, however, a blind innovator — the novel genotypes produced by mutation are essentially random, and no more likely to be advantageous to the organisms carrying them than one would expect by change (i.e. not very likely at all).
Selection, whether natural or artificial, is what happens when the organisms carrying the diverse genomes produced by mutation end up competing for survival and reproductive opportunities against each other and against their environment. It acts on the diversity generated and replenished by mutation, favoring those mutations that happen to confer a survival advantage.
Finally, inheritance ensures that the beneficial alleles generated by mutation and filtered by selection get passed on to successive generations, allowing them to spread in the population and to be further modified by successive mutations. In a sense, inheritance and mutation can be seen as opposite sides of the same coin: the more faithfully genes are inherited, the less mutations there are. Yet both are necessary for evolution; an organism that never mutates obviously could not evolve, but neither could one that mutated so much that all of its genes were totally scrambled in each generation.
Of course, this whole process is stochastic in nature — even the best adapted individual may die without offspring due to simple bad luck, just as even a maladapted individual may get lucky and prosper. But over generations, there is nonetheless a statistical tendency for individuals carrying certain alleles, or combinations of them, to be more fecund than others, and those will be the alleles that will eventually come to dominate the population (assuming they're not it turn displaced by even more successful mutants).
Thus, I would personally say that, of these three components of evolution, (natural) selection indeed "works only with what is already present in a population." However, it's not really like a copy editor, since it does not seek to "correct errors" in the genome; it's more akin to a slush reader, who just sorts the incoming texts into ones to keep and ones to throw away.
