literary AI redux

RECENTLY, I TRIED OUT the AI features on Storywise, a service designed to connect aspiring authors with independent presses. In addition to providing a platform for writers to pitch and submit their books to publishers, Storywise, for a modest fee, offers authors a chance to have their full manuscripts critiqued by an AI program.

The whole process took only a minute or two, and the volume of analysis the bot came up in with in that time was impressive. The critique was organized mostly by the book's narrative structure, roughly translating into three acts, and included editorial suggestions, a list of similar titles, marketing insights, a choice of covers, and an overall assessment of the storytelling.

Compared to the AI bot that two years ago tried to write metaphors for me (a laughable disaster), the Storywise manuscript bot is a pro. It's startling how quickly the technology and the programming is progressing. In addition to its overall organization and the breadth of its analysis, the Storywise bot was good at identifying unnecessary repetitions and plot inconsistencies. Some of its editorial suggestions for moving scenes up or down proved helpful.

Overall, I'd say the bot's sense of story structure is its biggest strength. What it missed in its analysis of my book, however, was significant. My mystery includes a short story-within-the-story, starring a character who is the alter ego of the primary antagonist. Although it's clear from the text that the antagonist is writing this story, the AI software did not recognize the inner story as separate from the outer. AI listed the main character in the short story as an independent supporting player in the outer plot.

Also missing from the AI critique was any mention of the sub-plot, which accounts for about 20 percent of the book. Whether to expand or delete the sub-plot was one of the questions I expected feedback on, so its absence altogether from the analysis was both disappointing and surprising.

SOME OF WHAT the bot did choose to focus on was equally surprising. It contended that a scene that takes place between lunch and dinner at a Chinese restaurant needed more sensory details, specifically smells. My reaction? Unless the odors are unusual—something the reader might not associate with a Chinese restaurant—why make note of them?

At the same time, the bot described the scene's dialogue exchange as long to the point of "dreariness." (Can an AI program be drearied?) Could the cat-and-mouse tenor of that introductory scene and its subtext have been lost on the bot?

That and other comments made me think that the program was working from some sort of Creative Writing 101 checklist. The bot didn't really understand, as a human reader would, why a smell might or might not be worthy of mention, or how subtext might work, or why some exchanges might be protracted for dramatic purposes.

And, of course, the program made no mention of the potential emotional impact of the writing. It couldn't assess whether a reader would be moved by a cathartic moment or thrilled by a tense chase. It made no mention of the quality of the prose. And humor? Forget it!! Bots still don't get jokes. To put it in Star Trek terms, the Storywise bot performed like "early" Data, an android striving to be human-like but not quite getting there.

One final Storywise oddity: After making two tiny edits based on the critique, I removed my original manuscript and uploaded a revised version. A good chunk of the analysis remained the same (it still wanted those Chinese restaurant odors), but some of it was arbitrarily different, including the marketing strategy. The tenuousness of the critique didn't exactly inspire confidence!