*
Biology is the science of living things, therefore the definition of life is a matter outside biology.
Indeed, the definition of life is, properly speaking, outside of science; despite that the mainstream current definition is derived from chemistry - based on the replication of molecules with the possibility for variation and selection. But that this really constitutes 'life', does not come from science but is a non-scientific assertion or assumption.
The other mainstream definition of life refers 'metabolism' - but the 'real' nature of life is just a matter of choosing a definition - there is no right answer; and furthermore replication and metabolism may have evolved separately and using different molecular types.
*
(For example, Freeman Dyson has argued that metabolic life may have evolved firstly among proteins, and this metabolism was parasitised by replicating RNA molecules - and later protein and nucleic acids co-evolved to join in that symbiosis we observe as 'the cell'.)
*
Natural selection must have something to work on - and that something must be sufficiently stable to allow for some reasonably large number of generations to do the work of natural selection.
*
(In a deeper, metaphysical sense; natural selection presupposes an understanding and identification of 'form'; because form dictates what it is that evolves, and when that form stays the same or when it changes to another and different form. Unless form, its constancy and change are already known, in a definition originating outwith biology, then the workings of natural selection could never be observed.)
*
Therefore, life must have been initiated by chance; then this spontaneous life must have 'fallen-into' some natural form, pre-existent order, stereotypical pattern, auto-catalytic system, or 'strange attractor' which kept it going for a while - because only then could natural selection do its work.
Even with natural forms - when it comes to life: what chance has given, chance also can take away.
And any form of life will have a tendency towards extinction from what has been termed error catastrophe (unless it has evolved methods for preventing this).
*
Error catastrophe is what happens in a metabolising or replicating system due to the spontaneous occurrence of errors to processes and copying. Such errors will naturally accumulate over time, unless there is some means to prevent them accumulating.
(Mutation accumulation is a special type of error catastrophe:
http://iqpersonalitygenius.blogspot.co.uk/search?q=mutation+accumulation )
And the vast majority of errors will damage functionality (because only a tiny proportion of undirected changes will improve functionality), and each new accumulating error will tend further to damage functioning - tending towards a catastrophic loss of function with death of the individual and/or extinction of the lineage.
*
But however life is defined, the principles apply that life must form and have some degree of stability by some combination of chance and natural forms; and then the first job of natural selection is to stabilise life.
Put it another way - life may form spontaneously - but it will not last without the help of natural selection.
*
Natural selection must primarily be about sustaining life, because only when life is being sustained, is there a possibility of the of life being improved.
Because, it is statistically improbable for an error to be an improvement; and therefore it typically requires considerable time (in terms of generations), and or a considerable population (because numbers amplify the number of generations) before the undirected ('random') occurrence of a beneficial error.
*
So, life happens by chance but life is also is doomed by chance to be short-lasting; therefore the first 'job' of natural selection is to keep life going just as it is - just as it has arisen by chance.
And only after life has evolved such as to have been kept going just as it is by chance, is there any possibility of natural selection to produce adaptation of organisms.
First sustaining, then later (perhaps) improvement.
*