Thoughts on AI
AI you ask? No, I’m not talking about the ai, a three-toed sloth (but Scrabble players know this two-letter word well). No, not airborne intercept or artificial insemination. I’m obviously talking about artificial intelligence.
What is artificial intelligence? For those of you living in caves or under rocks (that is, the likely survivors of the cataclysms to come), according to Merriam-Webster, AI is—and allow me to (mis)quote—“the thing no one really understands, but it’s kind of like a very smart computer that messes up around half the time, and just might kill all of us within a decade.”
In the meantime (ad interim (ai)), here’s some random stuff on AI, and how it relates to Submitit, and writing in general.
We avoid AI at Submit.
Imagine how much time we’d save using AI to read or edit our clients’ work? Or to analyze the stories, essays, and poetry published in literary journals (to add data to our algorithm)?
But we don’t do it. For a couple of reasons: (1) our “scoring” would be inaccurate (scoring is how we match your work to journals) and (2) our editing would be subpar.
(Quick note: Some elements of AI are difficult to avoid, most notably with Google searches. For example, the other day, I wondered if “southern Connecticut” was an official region (and thus should be capitalized) or just a general one (and thus is correctly lowercase). The result was AI generated (turns out, it’s not an official region). I also sometimes use AI to help with images for these blogs, as I’m sure you may have guessed. So when I say we “avoid” AI, take this with the smallest grain of salt. But we certainly avoid AI with our core functions: reading and editing your work, and finding high-match journals.)
Scoring
AI does not really “hear” the quality, craft, and style of poetry and prose. Analyzing writing, as with any art form, is a supremely human endeavor. When we, at Submitit, read your work, we’re looking for such nuances as humor, style, craft, and tone (and much else). Perhaps someday AI will be able to recognize these things (and, more importantly, recognize their effectiveness), but it’s not there yet.
Editing
AI also largely fails when it comes to editing (even copy editing, which is inherently more “mechanical” and less “creative” than line or developmental editing). AI usually does a pretty good job with basic grammar and usage errors, such as fragments, comma splices, subject-verb non-agreement, misspelled words, and so on. But here’s the problem: we’re not writing academic tutorials or political treatises or blog posts. We’re writing literature.
Many editorial issues, even those related to copy editing, have to do with sound and style. I just edited a short story with several long sections written as stream of consciousness: run-ons, comma splices, braided subjects, and other liberty-takings. AI would fail utterly with a piece like this. Or consider wordplay. My own writing tends to be filled with neologisms, archaisms, odd portmanteaus, intentional misspellingz, abbreviations, etc. No chance an AI bot would “get it.”
And there are many editing issues that AI would likely not even notice, such as awkward word repetition, “thesaurusizing” (inelegantly changing words to avoid awkward word repetition), formatting problems (especially with dialogue), pronoun ambiguity, the art of punctuation (gray areas), possible misreadings, and much more.
One of the main issues, I think, is that AI doesn’t recognize a writer’s intention. This is usually the measure of whether stretching grammar rules is okay, or whether these “errors” will be read as, well, errors. If it’s clear a writer means to write a page-long sentence with no commas, great. If not, not.
I’ve been mostly talking about copy editing. For more expansive editing (line or developmental), in which we dive into things like plot, character, structure, and all the other “big picture” aspects of writing, AI is not even close to good human editing. As with scoring work (for our algorithm), editing requires a human touch.
Finding Journals
What about finding actual literary journals for our submissions? I hope it goes without saying:
We don’t use AI in any way to match your work to journals!
We’ve spent countless hours analyzing (reading!) literally thousands of stories, essays and poems from hundreds and hundreds of journals. From all this reading, we’ve found each journal’s “fingerprint”—that is, the literary qualities of the writing they tend to publish. Then we read and score your work, using the same criteria, run it through our algorithm, and get a list of journals based on match percentages (measures of how close your work matches journals’ work). No AI. Just a lot of reading and some basic math.
But, well, why not AI? As with scoring and editing, Lord knows it would save us some time.
I’m happy—very happy—to share an AI test run. This is for an actual story in our database.
AI Test Question
“I have a short story, 5,535 words, dark themes, also darkly humorous, literary style, fairly experimental. I want to target good, well-known journals, currently accepting submissions, with an acceptance rate higher than 1%. Any ideas?”
Google:
Driftwood Press [currently closed; more an anthology than a literary journal; it’s not in our database, so I couldn’t check the match percent]
The Ex-Puritan [not really a top journal; Duotrope shows 0% acceptance; again, not in our database, so no match percent]
Story Magazine [sadly, this journal is on indefinite hiatus]
New Ohio Review [currently closed; max length is 5,000 words]
Conjunctions [currently closed; also a black hole (very low acceptance rate); otherwise, not a bad match, in terms of style]
ChatGPT:
The Fiddlehead [Duotrope shows 0% acceptance; only a fair match (67%)]
The Southern Review [another black hole (usually around 0% acceptance); not a great match (66%)]
Colorado Review [Duotrope shows less than 1% acceptance; fair match (69%)]
Cleaver Magazine [max length is 4,000 words; not a top journal]
The Sun [Duotrope shows less than 1% acceptance; fair match (65%)]
Analysis:
Google: Among other issues, three of the journals were closed and one was defunct. To say the least, not a great result. Grade: F.
ChatGPT didn’t do much better. One journal failed the word-count guideline, and the rest had acceptance rates under 1%. None matched over 70% (anything less than 70% is fair at best—you can read more about this here). At least we’d get some submissions from this search, but I suspect the rejections would soon follow. Grade: D.
Here’s what Submitit’s algorithm came up with for this story (top five journals):
Third Coast Magazine (79% match, >1% acceptance)
Hobart (75% match, >2% acceptance)
Southern Indiana Review (75% match, >1% acceptance)
American Literary Review (75% match, >2% acceptance)
Gulf Coast (74% match, >2% acceptance)
Solid journals. Good matches. Grade: A. :)
I was going to run a test with another story, targeting accessible journals, but the above AI results were so pathetic that I think I’ll stop here. Safe to say: if you're tempted to use AI to find literary journals for your own work—instead of (the notion!!!) using Submitit—you’ve been forewarned.
(Owner’s sigh of relief goes here.)
We don’t accept work written with AI.
As mentioned in our Terms of Service, we don’t accept AI-based work (if for no other reason than that most journals don’t accept AI-based work). In truth, we don’t check this. We trust our clients to honor the rule. (Clients officially agree to our terms when signing up.)
But it’s a rule worth honoring. A work written largely with AI will have two challenges: (1) if a journal notices the AI (by ear or with an AI-detection tool), your work will likely be rejected and you might get blacklisted or sent to a gulag in Siberia; and (2) your work will likely suck.
So please don’t use AI when you write, at least not for creative purposes (more on this below). The result will almost certainly be plain and boring, borrowing from the internet of clichés and banalities that these AI systems are trained on. (Remember, these language learning models scrub huge amounts of past writings, and then create text based on likely word patterns. So what you get is largely (entirely?) reflective of the way most people would naturally write. In other words: dull, hackneyed writing.)
As mentioned above, we’re writing literature here. We’re striving for voice and style. We very much want to not sound like everyone else!
How should you use AI (if at all) in your own writing?
Here are some examples of AI warnings in the submission guidelines for literary journals:
“We do not accept stories written by AI tools such as ChatGPT. If you submit such stories, we may ban you from submitting to our magazine.” —Flash Fiction Magazine
“We do not and will never consider essays written—even partially—by LLMs [language learning models].” —Brevity (italics are mine)
“We will not consider work produced, in whole or in part, by a large language model, such as ChatGPT, Gemini, or Claude. If you’ve developed your idea through exchanges with AI, used a structure proposed by AI, or even incorporated a phrase composed by AI, please submit to a publisher that expressly welcomes shared human-AI authorship.” —AGNI (italics are mine)
Other journals are a bit softer with their dicta:
“Primary authorship must be human. TYR does not accept submissions that rely on AI to generate entire drafts, arguments, or analyses.” —The Yale Review (again, italics are mine)
“Spell-check and grammar-check are permitted, but writers and translators must affirm that the work submitted is their own and not the product of artificial intelligence tools, software, or generators.” —New England Review (yes, yes, yes: the italics are mine)
So it seems that for many journals, even touching AI is a no-no. Others are okay with some AI use. So, to my question above: How should you approach AI in your own writing?
I’m not sure I have the perfect answer (and if I did, it’d probably be wrong in a few months—things are changing that fast). I’ll just tell you how I approach AI. Feel free to use this as a guide, if you’d like.
For my own writing, I use AI—at most—as a tool of research. In the past few weeks, I’ve used it for the following queries (I’m checking my history):
Names of cities in California at least six letters long, with only three unique letters? [There were none. :( ]
Severed penises in Greek Myth other than Uranus? [None. :( ]
What section in Being and Time does Heidegger mention a clearing in a woods? [Chapter V, §28]
Can you find me a source with page number for the Whitman quote “gentle soft-born measureless light”? [I had this quote in my notes, but forgot to record the source.]
Famous novels that mention Walter Benjamin’s “angel of history”? [Ben Lerner’s 10:04 and Kim Stanley Robinson’s The Ministry of the Future]
I don’t feel guilty about these searches. None of them involves actually writing. Essentially, I’m asking AI to help me with my research. It’s a time-saver.
Am I one-hundred percent sure my writing passes the AI guidelines mentioned by some journals? I guess not. But I do think I’m obeying the spirit of these laws. Using AI for research seems to me like a no-brainer (literally and figuratively).
(I feel obliged to add: I—with great intent—used AI to help find the AI journal guidelines above. Once again, this is research. It is not a creative endeavor.)
How might we (Submitit) use AI in the future?
So far, I’ve ruled out using AI for reading, editing, writing (other than for research), and finding high-match journals. But that doesn’t mean I’m not open to using it in other ways to help run Submitit.
One possibility is to update journal guidelines and open/closed statuses. Currently, we check to see if journals are open or closed manually (weekly). It gets us close—close enough—but it’s not perfect (this is one of the reasons why we add extra journals to our lists, in case any are closed when we go to submit your work). We also run an annual journals audit, checking for changes to things like word-count guidelines and acceptance percentages.
Assuming AI could someday tackle the above accurately (my test runs above show that AI is not there yet), I see no harm in letting it handle these mind-numbingly boring tasks. (When I do these things, I usually put on some music, pour myself a glass of bourbon (neat, please), and basically work on autopilot. Could I find something more enjoyable to do with my time, while listening to music and drinking bourbon? Yes, I believe I could.)
Finally, if AI could successfully decipher the myriad ways journal editors update writers about submissions (acceptances, rejections, etc.), I’d be happy to let a bot handle this. I’m sure we’d get a few false positives and false negatives (rejection emails sometimes seem to do everything but make the rejection clear), but at least the updates would be in real time.
Conclusion
We’ll see where the future takes us with AI. (Yes, hopefully, humanity will be around to see this future.) In the meantime, know that Submitit has real humans reading and editing your work, and we use numerical data based on the work of real humans to find high-match journals for your submissions. And please, if you use AI as a tool for your own writing, do so for research only.
—Erik Harper Klass (fellow human)
For updates, tips about submitting, discussions of craft and style, and other miscellanea, please subscribe to Submitit's blog here.
