I would like to talk about AI. Again.
In our Winter 2021 issue, the cover art and all the illustrations accompanying the stories were at least in part created by an AI-powered image generator. However, it was a somewhat complicated, laborious process: interesting to undertake, but repetitive and time consuming, fraught with unusable misfires, and (due to my modest computer resources) prone to system crashes. If all went well, I could create a workable image in fifteen minutes… but it might take a dozen failures to acquire one winner.
The recent attention which has fallen on another AI image generator, Midjourney, may underline two inevitable necessities to enter the mainstream: quality of output and ease of use, both of which Midjourney has in spades. Text prompts with any degree of detail produce four image options in approximately 60 seconds, and they are almost invariably good; sometimes weird, sometimes wonderful. It will create variations on those initial images in maybe half that time, and upscales the ones you like into large, highly detailed final versions in perhaps a minute or two at most, depending on how busy the site is.
Unsurprisingly, it gets quite busy.
All the art in the present issue of Mythaxis was created using Midjourney, which in addition to being very accessible is also startlingly inexpensive to use—which connects to a subject I mean to return to in a future editorial. In some cases these images are Midjourney’s unaltered output; in others, I’ve composited its output to a greater or lesser extent to create the final product; and I’ve also had to perform a few instances of cosmetic (in some cases, emergency) surgery to deal with problems thrown up by the AI. It isn’t a perfect science yet, though I think you’ll agree that the results are impressive—but while they make for a very attractive-looking issue, this isn’t the aspect of Artificial Intelligence I really want to discuss.
A lot is made of how Artificial Intelligence is going to take over all our jobs; how it can already compose music easily mistaken for the work of history’s greats; and how its often amusing attempts at creative writing seem less laughable with each new viral report. Ever since our first visual experiment with using AI, we behind the scenes at Mythaxis have discussed how to further explore the potential of AI in the field of magazine publishing. And, after much debate, we’ve come to the conclusion that the one expendable participant, the most toothless cog in our machine, the weak link in our chain, the fifth wheel to replace, is…
I won’t deny, this came as a bit of a blow. Nevertheless, to this end I enthusiastically joined in plotting my own downfall, and I can confirm that the first step on the path to my obsolescence has now been taken, because we are indeed training up an AI to replace me. But it’s going to have to prove itself before I hand over the reins. And I suspect I’ve got a little while in the hot seat ahead of me yet.
Our fledgling experiment went something like this: first, we exposed a learning algorithm to the seventy-plus stories which have previously been published in Mythaxis during my editorship, so that it could analyse the material which I considered best amongst the hundreds of submissions we have received. Then, we challenged it to survey the one-hundred and fifty-eight stories which we received in our most recent submission window. If its understanding of my taste in fiction was accurate, surely it would rank most highly the stories which I actually selected, no?
The answer was… No.
In fact, of my eight acceptances that window, my so-called replacement rejected seven. And the one story it agreed with me about was also the one it was least certain of out of the eight. It really didn’t like the others! So, from my perspective at least, we didn’t so much create an Artificial Intelligence as an Actual Ignorance.
However, maybe this isn’t terribly surprising. It should be noted that seventy-four stories is not a very large sample size. Also, that the characteristics of the stories we publish vary in many different ways: from flash fiction to long shorts, science fiction to fantasy to horror, written in flavours (or flavors) of English from American and British to those influenced by fluency in other languages. There’s a lot of complexity, in short—not least, I would hope, in the thing we’re actually attempting to simulate here: me.
Therefore, we shall persist. We plan to refine our strategies, and educate our little monster. We’re calling it the Slushbot. Starting now, every three months our currently oh-so-limited AI will have a chance to test itself against my judgement. As new submissions come into the slush pile, I will make my choices, and Slushbot will make its. My taste is impeccable; we will see if Slushbot’s taste improves.
We do all this with a goal in mind: we hope to be the first magazine edited by an AI. So, should it ever match my picks from a window’s submissions to a sufficiently frightening extent, Slushbot will get to edit an issue of Mythaxis. For one window, we will give it the final say over which stories to include, and I will go on holiday.
Guess I’d better start packing my bags… you know, sooner or later. Next year maybe. Or whenever hell freezes over. Under a blue moon. As a pig flies over it.