There's been recent talk about the GPT-3 algorithm being able to simulate human writing enough to pass basic evaluation. This prompted a number of people to inquire if the days of writers are numbered. We can all rest easy in my opinion. Before you can make an artificial writer, you really need an artificial human. Even the task of writing LIKE a particular writer, making a model using the body of a writer's work as a training regimen. To create NEW unique narrative that isn't just a random patchwork of the writer's previous work you would have some pretty huge hurdles to overcome.
The difficulty is being able to quantify an author's style. Any code would have to make some basic assumptions about story. In simplistic terms, a story consists of characters (protagonists) trying to solve or overcome a problem. The narrative consists of the introduction of the problem, the toil (or attempts to solve the problem), and the resolution where they ultimately fail or succeed in the endeavor. Where this challenge becomes incredibly complex is how you would distill the writer down to "features" or "characteristics". Some writers are known for characters and characterization, others for plot or setting.
There is such a vast range and so many minutiae that this part alone is problematic from a scope standpoint. In terms of an algorithm, you could attempt to tag an author statistically by common phrases or perhaps topics and setting details but this is several magnitudes away from the essence of a person's writing that is unique to their personality, perspective, and history. You could analyze a person's stories by associating them with universal myth paradigms and if you had enough examples could say the author tended to write stories of that particular type. How does the algorithm come up with a character? How does it do dialogue? How would the software determine the narrative conflict and how that aspect is resolved.
Essentially, you'd have to code a story-teller's brain. You'd have to teach it how to visualize characters in terms of mannerisms, speech, need and desire lines, with decisioning algorithms that could walk them through the maze of the concocted narrative. There are ways you could boil this down, but none of the algorithms I read about would get you close. You can't boil a literary style down to statistics (not in any meaningful way) where you could mimic it. Ultimately, you couldn't emulate a style from a sampling of the writer's work. The work grows out the person's psychology.
A story is like a pachinko ball that begins with inspiration and cascades down through the narrative mechanism bouncing off of personal experience, bias, and all manner of additional factors. To throw even bigger wrenches into this implementation is that some classic writers like James Joyce didn't even write stories, they wrote streams of consciousness that don't follow the story model. Other writers specialize in vignettes and other literary artifacts that likewise defy categorization. How do you teach something to recognize these outliers? So, while AI driven fiction writing is interesting in concept we're still a long way from distilling a person down to some formula, and then use that formula to influence the variation of a generated narrative artifact.