So in the most recent Readers and Book Lovers diary here yesterday there was a link to another piece on a blog called “Inside Hook”, titled “ChatGPT Is Now a Published Author of Over 200 Books on Amazon”. That piece included the line: “And apparently, other people have the same ideas, as several titles by ChatGPT have launched on the site’s self-publishing program, with several of them actually utilizing the AI to write about the AI (e.g. How to Write and Create Content Using ChatGPT). The rest appear to be other children’s books, guides and even a collection of poetry.” I’m pretty sure the poetry collection referred to is mine.
Lately I’ve done a couple diaries here at DKos about ChatGPT and my adventures poking around with it. In “Playing Around With ChatGPT”, I describe how I experimented with the AI’s capabilities by prompting it to write some poetry, and in “Some Poems Done with ChatGPT”, I presented some of the poems produced by the machine that I liked.
I found the whole idea of it interesting, so I decided to write an Introduction giving a history of artificial intelligence and machine learning, then present a number of poems that had been written by the AI, and submit it to Kindle Direct Publishing, Amazon’s self-publishing program. I titled it The Soul of a Machine: Poetry From an Electronic Artificial Intelligence, Written by a Machine, and Edited by a Human. Since I wanted to be transparent, and since for me the most interesting part about it in the first place was that the machine had written the poems and I wanted people to see and judge how well (or otherwise) they thought the machine had done, I tried to list “Artificial Intelligence” as the Author, and myself as the “Editor”.
And that baffled Amazon. They rejected the Author listing I submitted, arguing that it would “confuse the readers”, even though I had made it excruciatingly clear in both the Author listing and in the book’s description that the AI had done the poems, not me. After several rounds of back and forth, I reluctantly agreed to their request to list my name as the “Author”, and to drop the “Artificial Intelligence” from the credits, but I did not change the title or the description. I did rewrite the Introduction a bit to make it more a book about AI than a book by AI, though it still remained clear that the machine had written the poems. Amazon then withdrew its objections and published the book. (Though for some unfathomable reason known only to the Amazon Gods, the paperback listing still has “Artificial Intelligence” listed as the Author but the Ebook listing does not, even after our big long drawn-out debate over it.)
Ironically, given Amazon’s original objections, I think that is, in the end, far more confusing to the reader than my original arrangement.
So I then decided to try another experiment. Back around ten years ago I did a lot of DIY scifi fan films for YouTube, in which I wrote the script, made all the props and costumes, played all the roles, and did all the FX. The vids were not monetized and I did them just for the fun of it. One project I worked on was a script for a long film, around an hour, about humans being put into suspended animation and tended by clones to avoid environmental destruction. I titled it “Stasis”. Alas, although I finished the script and even started making some of the props, I never had the chance to shoot the movie. So the script sat on my computer. After a while I decided to turn it into a scifi novella, and I worked on it now and then for the next ten years. A couple of weeks ago it was about 2/3rds finished.
So as an experiment, I decided to take another look at the AI to see how useful it might be as a writing partner, rather than as writing generator. I began prompting it with scenes from the screenplay and asking it to generate a number of possible ideas, descriptions and dialogue for each scene. Most of what the AI suggested was, alas, crap. But some of it was, I thought, pretty interesting. So I took that, rewrote it, expanded it or cut it as needed, and worked it into the narrative that I had already done. When the novella was finished, I published that to Kindle Direct as well, using the title “Stasis”.
What I found was that while the AI’s own compositions were clunky, repetitive and usually crappy, it did come up with some clever ideas and some useful pieces, and I incorporated those that I liked into my story. So although I was legitimately the author of that novella, there was some input in there from the AI. I found it to be a useful tool for generating a variety of ideas, which I could then accept, reject, or adapt. It was like having a room full of writers who were each coming up with new ideas and bouncing them off each other.
That, I realized, was really the strength of the AI—it was good at generating new and interesting ideas. Most of them were of course crappy, but some of them were pretty good (just as with most human brainstorming).
So I then decided to do another experiment fully utilizing a human-AI partnership in that same way, and I asked the AI to generate a new story arc for a short fantasy swords-and-dragons novella. Some parts of it I liked, many parts I didn’t. What I ended up doing was writing a very long and very detailed outline which incorporated parts of the AI’s storyline and added a lot of my own to make a narrative that was broken into dozens of tiny little pieces. Then, for each individual scene, I asked the AI to generate some ideas about description and dialogue, within the overall storyline. At various places I would ask it questions like “I need a plot twist here—what do you suggest?” or “I want to add a new character here—what would be a good idea?” And I’d often ask it to “give me a description of this person, thing or place”. At one point I needed a good Norse name for a sword, and it replied with a nice name and a good explanation of what it meant, which fit well into the storyline.
Yes, most of what it came up with for these queries and questions was crappy, too. But some of it was pretty good, and I ended up rewriting and adapting some of the machine’s suggestions and fitting it into the outlined story. And so it went, scene by scene, almost paragraph by paragraph. Although every paragraph ended up being reworked and rewritten by me, usually several times, a nontrivial portion of the storyline and scenery was inspired by and taken from the AI’s suggestions, altered as needed. I just recently finished that work, titled it “Dragon’s Bane”, and just published it to Kindle Direct.
And that, I think, is where the AI’s strength really lies. It is not ready yet to be writing novella-length works on its own—it tends to be too clunky, too inconsistent and too repetitious. But it is REALLY useful for generating a menu of ideas and suggestions, and several times I took the story’s plot in a different direction than what I had originally planned because I liked one of the alternative ideas suggested by the AI.
So I view it as a tool, almost a partner, for the writing process. I think that’s where its true strength lies (at least for now)—not in generating text, but in generating ideas and proposals, which the human writer can then adapt and modify to make it work within their storyline.
As for the AI-generated books on Amazon, I have not read any of them. Based on my own experience with ChatGPT and the experiments I did with it, I expect that all of those books will be crap. But then, it’s sadly true that nearly all self-published human-produced books are crap, anyway.
AI is not yet an author. But it already can be a good tool for an author. The human, however, still remains central. The human writes the music and conducts the orchestra: the machine just suggests instruments and maybe a few bars of harmony. But in the end it all comes together in one complete piece.
At least that is how I see it.