From “Selfish genes, the phenotype paradigm, and genome evolution,” W. Ford Doolittle & Carmen Sapiena, Nature 284:601-3 (1980), here is one of the original definitions of selfish DNA:
What we propose here is that there are classes of DNA for which a ‘different kind of explanation’ may well be required. Natural selection does not operate on DNA only through organismal phenotype. Cells themselves are environments in which DNA sequences can replicate, mutate, and so evolve. Although DNA sequences which contribute to organismal phenotypic fitness or evolutionary adaptability indirectly increase their own chances of preservation, and may be maintained by classical phenotypic selection, the only selection pressure which DNAs experience directly is the pressure to survive within cells. If there are ways in which mutation can increase the probability of survival within these cells without effect on the organismal phenotype, then sequences whose only ‘function’ is self-preservation will inevitably arise and be maintained by what we call ‘non-phenotypic selection’. Furthermore, if it can be shown that a given gene (region of DNA) or class of genes (regions) has evolved a strategy which increases its probability of survival within cells, then no additional (phenotypic) explanation for its origin or continued existence is required.