Every since John von Neumann came up with the notion of a Universal Computer and Universal Constructor for self replication in the 1940s (before the notion of a "gene" was discovered!), I've been fascinated by the issue of self replication.
Self replication (particularly in biological systems) is a product of emergence (i.e., when a complex set of interactions between the individual units in a dynamical system results in a network/systems/pathway that becomes an entity in and of itself, forming a complex adaptive system). The emergent property of a complex system is at least a necessary condition for self replication to occur. In my view, this emergence arises as a result of interactions occurring at multiple scales that is capable of self reference (i.e., within a a tangled hierarchy causing a Strange Loop, as described by Douglas Hofstadter)
Self replication of a system may thus be characterised by tangled hierarchies or Strange Loops and emergent/complex behaviour when a network is created between the individual components of the system within the confines of a selective environment. "Life" could be considered to be either emergence, or the property of an emergent system when it can self replicate (i.e., preserve its complex organisation by a transfer of information). All these definitions are simply a matter of semantics.
As von Neumann realised, environment is one of the key ingredients that will make for a self replicating and self evolving system. Without a sufficiently complex (selective) environment, the individual units within a system will never form a complex dynamic network of interactions and will remain stagnant.
Self replication is tied in intricately to evolution. In any evolutionary system (i.e., where the environment imparts a "survival of the fittest" selective pressure because it is constantly changing), the ability to evolve is selectively advantageous. Self replication is a necessary condition for evolution. While it is possible evolution of a species or an organism can occur without replication, our biological universe works in this manner. Alternately, any modification (evolution) in a nonreplicating system can be thought of as process by which replication and change are merged into one.
In any evolutionary system, the ability to self evolve is selectively advantageous, particularly in the context of preserving the overall metasystem. While there may be exceptions to this, I think it is true given the neodarwinian paradigm we currently hold. Interestingly, however, if a creature or species developed the ability to self evolve (i.e., perform directed evolution on itself in response to environmental changes), then it would violate the neodarwinian paradigm (i.e., that changes (mutations) occur independently of the biological needs, i.e., reproductive fitness, of the organism).
The idea of self evolution isn't as far fetched as it would seem: mini evolutionary systems (i.e., the immune resistance and clonal selection) exist within our own bodies that do this to a certain degree---except that the genes aren't incorporated into the germ line.
Also, humans are good example of a species capable of self evolution (à la Gattaca and many other science fiction movies). It remains to be seen whether any undiscovered molecular mechanisms exist to accomplish this goal, one of the focuses of my research.
Within the confines of an evolutionary environment, the ability to self evolve is selectively advantageous (particularly when there are other species/populations also evolving). Species that possess traits that allow them to choose mates with beneficial traits and discard those without engage in a primitive form of self evolution. To self evolve, there must be some sort of sentience or self awareness (in other words, there must be a "direction" that is present that tells an organism its status in relation to its environment). Sentience is a byproduct of (the complexity required for) the trait of self evolution. In other words, I believe the reason why we have the complex brains we have today with all its abilities is so we can make a statement about performing genetic engineering on ourselves. This claim however requires that there be extreme feedback from the future which is hard to conceive, but the precursor trait to the genetic engineering self evolution trait likely manifested itself in less complex and clearly manipulatable ways, for example when early humans started to domesticate plants and animals and observed that selective breeding produced specific phenotypes.
Note that I am not saying that the complexity in our brains responsible for the awareness (knowledge) that leads to genetic engineering is directly inheritable, for at least two reasons. First, complex traits such as intelligence require the interactions of many many atoms, molecules, systems, and cells (neurons). Due to the chaotic/complex nature of these interactions, they are unlikely to be prone to any predisposition. Second, our genomes are fairly small and reencoding a specific arrangement responsible for a particular piece of knowledge becomes intractable. These are the same reasons why complex memories aren't inherited (though some may dispute me on this). This still doesn't explain why simple memories (such as antibodies against a disease) aren't inherited. Perhaps there is an evolutionary advantage (or even test) to be able reacquire information.