(or, How to Stop Worrying and Let the Language Models Crash the Party)
I’ve lost count of the contests that open with a cheery “No AI-generated work will be accepted.” The disclaimer is always followed by hand-wringing about “authentic voice,” “the sanctity of craft,” and “protecting literature from soulless algorithms.” Lovely. Also unenforceable. Unless the committee plans to DNA-swab every paragraph, they’re working on the honor system and a whiff test.
So here’s our counter-proposal:
Let the machines in. Pay them no fee, hand them no trophy, but let them lurk behind any author who wishes to employ them. Then blindfold the judges, open the polls to readers who don’t know—or care—who (or what) pressed the keys, and see what floats.
Below is the formal call, the rationale, and the inevitable objections answered in the least diplomatic tone I can manage without disqualifying myself from my own contest.
1 | Why we’re doing this
Because the anti-AI rule has become the fig leaf that covers a deeper panic. A 2024 survey out of Stanford HAI found that 68 % of casual readers preferred passages lightly edited by GPT-4 over the author’s raw draft—even when the readers knew a model had intervened . Meanwhile, PEN America’s report on “witch-hunt dynamics” shows authors losing contracts for admitting they used a chatbot to unblock dialogue .
We suspect three things can simultaneously be true:
- An experienced storyteller who rides the AI wave will outrun an equally experienced purist.
- A beginner armed with the same model will still trip over plot holes.
- Most readers cannot reliably tell who did what, and—heresy of heresies—might not care if the result sings.
Until someone constructs a controlled test, the debate is theology.
2 | Contest parameters
- Theme: Ghost in the Machine. Interpret freely: literal hauntings, existential dread, joyful transhumanism—just keep it under 2 000 words.
- Entry fee: $0.
- Prize: $500 (PayPal or crypto; yes, crypto is tacky, but bank wires charge fees that would make a Medici blush).
- Rights: You own everything. We only ask for a 90-day digital display license. After that, torch it or sell it.
- Deadline: We’ll set a firm date as soon as Google quits treating our brand-new domain like unclassified papyrus. (Late summer is the working rumor.)
3 | The lab coat details
Blind reading
Stories arrive stripped of bylines and metadata. Our voting interface shuffles order per refresh and shows randomized excerpts—think Tinder for fiction. No author name, no Twitter handle, no secret handshake. Popularity contests die in the dark.
Any tech allowed
Claude brainstorming, ChatGPT line edits, Midjourney mood boards that inspire your climax, or a complete refusal of silicon assistance—choose your poison. We do not need to know your recipe to taste the stew.
Post-mortem stats, anonymised
Entrants will complete a confidential questionnaire after results lock: What tools? How heavily? Did AI spearhead or merely sand the corners? We’ll publish aggregate numbers—“73 % of top-quartile stories involved AI at the revision stage”—without pinning labels on individual manuscripts. Nobody gets doxxed for their workflow.
4 | Objections, briskly incinerated
“But that’s not fair—AI can’t submit by itself!”
Correct. A model lacks motive. Somewhere, a human chose the theme, fed the prompt, tweaked the output, exported the file, and risked rejection. Authorship hasn’t evaporated; it’s grown composite. The experiment measures curation, not raw keystrokes.
“Readers will game the vote.”
Not easily. Hidden excerpt rotation plus IP-throttled ballots equals statistical head-wind against brigading. And every raw vote, plus timestamp, lands in a public CSV the day we announce the winner. Audit away.
“If the winner used AI and we never find out, the prize is tainted.”
Only if you believe soul is a checksum. We’re wagering $500 that craft and reader delight outrank tool purity. Contests that ban spell-checkers can throw the first stone.
5 | Why anonymity still matters
Several entrants will be mid-career novelists under contract. Asking them to confess AI assistance is asking them to risk real-world livelihood in a climate where some publishers still equate “I used GPT for brainstorming” with plagiarism. The Electronic Frontier Foundation calls anonymity “a shield from the tyranny of the majority” ; we wield that shield so writers can experiment without career suicide.
6 | Practical inspiration for the undecided
- Dictate a scene on your phone, dump the transcript into a model, and challenge it to supply sensory detail you ignored.
- Or reverse it: generate a sterile outline in GPT, print it, scribble marginalia until only the bone structure remains AI-visible.
- Or abstain entirely, polish each verb with a whetstone, and prove the doomsayers wrong.
Art often blooms under constraint; consider this contest an elective constraint, like writing in terza rima or hammering a sonnet into fourteen equal coffins.
7 | FAQ in embryo
Will you run AI detection?
Only as an academic control. Tools like GPTZero have false-positive rates north of 15 %. We’ll publish those error bars in the final report to show how unreliable “AI sniffing” still is.
Can a model win on its own?
No. Terms require a legal human recipient for prize money. If you slap your ChatGPT beach-read into the entry form untouched, you’re still that human—lazy, perhaps, but human.
What happens if a crowd favourite is later proven entirely machine-generated?
Nothing. The point is to measure preference, not moral purity. If pure machine text captivates, that’s a data point—one we’ll annotate, not annul.
8 | What happens next?
We finish indexing. You spread or doom-post this call. Submissions open. In-box fills. Voting goes live. Aggregate stats drop. Think-pieces follow. Maybe the results confirm our hunch that hybrid skill beats raw algorithm; maybe the ghost outwrites us all. Either way, we’ll have something weightier than vibes to argue about.
Light the lamp. Feed the model. Sharpen the quill.
Let the best ghost win.
Writing contest details and submissions page.
Leave a Reply