Linguistic rules are more like DNA than algorithms, their transmission is subject to random mutation which is what accounts for linguistic change.
Well, actually, linguistic rules are nothing like DNA. They are just slightly more like DNA than algorithms when it comes to their transmission. And only in the sense that there are errors in their transmission that look kind of like the random mutations on which the natural selection operates in the process of evolution.
Imagine sentences as a kind of organisms and rules as the DNA that produces them. Each utterance is generated based on the rule (DNA code). It leaves traces of its DNA in the minds of the speaker and the hearers which is then used the next time the utterance is used. But the transmission is not perfect, there are errors (random mutations) in the utterances most of which just fall aside but some of which are adopted and in the future the new shape of the utterance is used based on new DNA. (Not unlike memes, then, but for the memes, they are their own carriers).
This is an extremely limited metaphor that only occurred to me when discussing the regular nature of linguistic variation (of the sort studied by variationist sociolinguists). It seemed to me that they are still operating with a fairly algorithmic notion of the ‘rule’ that does not seem compatible with the notion of variation. In many contexts, they talk about patterns, trends and tendencies but when it comes down to thinking about what the rule looks like, things get very traditional.
In fact, language rules are not very much like language rules, at least not those contained in most grammar books and the works of Stephen Pinker. I would say that they are more like whatever it is that lets us walk around the city. Flexible, reacting to what is around and inside us, but still remarkably regular and predictable. The sort of thing that lets us recognise landmarks, step over dog shit, nod at acquaintances and hug friends. Most of it completely automatic, some of it self-aware and some of it following generative rules. In short, something not unlike what construction and cognitive grammar talks about.