top of page

The great American AI mystery?




As the screenwriters' strike goes on, with AI-produced scripts a bone of contention in the negotiations, I've been thinking a lot about the use of AI software in the creation of crime fiction. On some level, turning to algorithms to write a cozy whodunit or a gritty police procedural feels natural enough. Predictability is implicit in genre writing. Consciously or not, readers expect mystery plots to follow certain narrative rules and be comfortably formulaic. And if there's one thing AI thrives on it's formula.


According to the British journalist and author Christopher Booker (also a climate change denier who ironically looks a lot like avid environmentalist Ed Begly, Jr.) there are in the entirety of drama and literature only seven basic plots. And according to Georges Polti, the original genre decoder, there are no more than 36 plot variations in all of storytelling. Whether these scholars' assessments are on the mark or not, I think it's fair to say. there's nothing new under the crime fiction sun, plotwise.


In most debates about AI in the creative space, AI is presented as the alternative to an IRL author. But what about the many companies that actually cater to IRL writers? If IRL writers are limited to seven basic plots and 36 story variations, isn't it reasonable for them to use an AI program to quickly outline the plot possibilities?


The popular AI-light website Plottr.com does just that. It allows writers to brainstorm story ideas with help from a well-informed computer program. To begin, the writer supplies the program with three elements: the main conflict, the principal characters and what they want, and the story's setting. From there, the program will generate outlines for the entire plot. I haven't tried it, but from what I've read, it seems like a kind of thesaurus of plot twists – potentially handy and relatively benign.


But can AI programming capture the nuances of style and characterization that transform standard whodunits and police procedurals into best-sellers or classics? Some AI developers seem to think the answer is "yes."


Out of curiosity, I did a brief search of software designed to improve or complete writing tasks. Most of the AI writing marketplace focuses on essays, blogs, copy and report writing. A few, like Sudowrite, specialize in creative writing. As with most of these companies, Sudowrite offers to help customers with a variety of creative writing tasks. With your input, it can brainstorm story ideas, write a first draft, or come up with sensory-rich descriptions. You can even "expand" your story "when you have a rough draft done but you just want to make it longer."


[Make it longer? Like a class assignment? That right there made me cry a little, as the current mania for 400-page-plus mysteries, filled with fluff, is my biggest peeve as a reader. Are bloated manuscripts the result of Amazon's Kindle Unlimited, which pays writers by the number of pages read, I wonder? Regardless of the reason, paperbacks heavy enough to use as weapons are a plague. But I digress.]


The Rewrite function, according to Sudowrite, is "for when you know a passage needs to be improved, and you want Sudowrite to do the word-smithing." Okay. Nothing terrifying about that – as long as the program's wordsmithing is better than yours, and not just grammatically correct and with more words.


Below is an example of suggested rewrites, or rephrasings, that Sudowrite's program came up with for a passage from my first draft of The Birthday of Eternity, Book Two in the Comfort & Company series (coming soon!). I'll let readers decide on the merits of each.



To be fair, the Sudowrite guide cautions that the software works best when there is more original material to learn and pull from; I only submitted a page or so. I'm skeptical, however, that I would have gotten radically different phrasing even if the program had seen the whole draft. Clearly, the suggested phrasing is syntactically nothing like my own.


The guide also encourages users to be as precise as possible about their rewriting goals. Do you want the selected text be more dramatic, more descriptive, etc.? Under the "Describe" option, the user can filter the process to focus on any or all of the five senses and on "metaphors." Below is what happened when I asked the program to create a metaphor for me:


Crazy, huh? Kinda like those loopy Chinese-English translations you used to see on restaurant menus or in owner's manuals, but with many, many more words. (I say "furniture," you say "plague of moths and termites"; I say "the Reverend," you say, "last year's insect – deformed, feeble and old." Whatever!)


Sudowrite cautions that their software output is only as good as its input. Garbage in, garbage out. In other words, if the original text is schlock, the Sudowrite text will also be schlock, but with different (and no doubt longer) turns of phrase. The company also stipulates that if "the writing you’re getting is not in your voice, look at your document and remove anything that isn’t a match to what you’d like Sudowrite to write for you. Or, use Rewrite to improve it."


At this point, I'm thinking, if my original is halfway decent (no garbage in) and I'm capable of studying my text for voice and removing the inconsistent parts – as is required for the algorithm to work – and I've seen what the Rewrite program is capable of, why would I need a computer program to identify and fix my writing? Will seeing bad versions of my prose help me improve as a writer? For a reasonably confident writer, it might. For novices, however, it might just confuse and demoralize.


I don't mean to pick on Sudowrite specifically. I think it represents the current industry as well as any. But for me, asking an AI program to write and edit dramatic stories is like asking for directions using the most winding route possible and with no expectation of ending up anywhere near where you want to go. Or to put that even more metaphorically: asking a swarm of poisonous vipers to encircle you, then lead you into the void of the dead unknown.



Comments


bottom of page