I’ve been following the news about ChatGPT generated short story submissions with some dismay. If you’re not aware, some markets, for example the excellent Clarkesworld, have been so swamped they have been forced to temporarily close to submissions.
The problems with dealing with this kind of spamming seem to be those of scale, granularity, and cost. Small magazines don’t have the staff resources to effectively filter out all the bogus submissions, automated checks cost money they don’t have, and are not 100% effective – they generate both false positives and negatives.
It’s a problem.
I decided to ask my two sons, Ashley and Tom, if they had any ideas. As co-founders of Scirra, and authors of the Construct3 game creation software, they have a lifetime of experience dealing with cutting edge internet technology, coding, and transaction validation.
Ashley had several interesting ideas.
Ask for additional information that is harder to fake, or easier to tell if faked – a cover letter, a questionnaire, a video where you talk about your writing process…
Just make the process longer. Someone who clicked a button to generate a story might not be willing to spend a while on submission, but someone who really invested weeks/months/years on it probably would. The reviewers don’t even need to look at all the provided information – again it’s just there as a filtering system.
I think this is a good one, and I’ll come back to the idea later. There’s going to be a bit more work for the slush readers, but you are also pushing effort back onto the submitter. And not too much to deter the genuine writers who you want to let through. As one of those myself, I can easily understand why this is necessary.
Start a short story magazine solely dedicated to AI-generated stories. Perhaps all the submissions will then go there instead. If the aim is merely to take submissions from other publications, then quality control doesn’t really matter and readers can avoid it, but the AI goons get their stuff printed.
I like this, it’s a witty, clever idea and I’m sure someone with good coding skills could put this together. It might even become a thing in itself.
I see the problem being that the ChatGPT ‘authors’ are submitting to professional markets. Their overweening arrogance and entitlement means they believe the rubbish they generate is good enough for a paying market. They won’t submit to for-the love markets.
Fight fire with fire and try to find an AI that can identify other AI-written stories. Sounds like that already hasn’t worked though and there will definitely be an arms race as people try to invent AI that passes AI checks.
These do exist, and Tom suggested one called Zerogpt, but as he also mentioned, there is that problem with false positives and negatives. No doubt this will improve.
Close submissions to all except trusted authors, e.g. previously published. However it’s still difficult to establish a new trusted author and I guess that defeats the point of magazines for new/aspiring writers.
This can only ever be a short-term solution. It might happen, but ultimately you are harming your own market – the spammers have won.
Tom came up with a similar idea to Ashley’s first suggestion:
My immediate thought is to try to make it all work by design. For example, if inclusion in the magazines means the author gets paid a small amount of money, you can insist that all stories include payment details in the event of a successful submission or they will be ignored. Then it’d be easy to filter out stories submitted by the same people that are low quality.
I really like the principal of these two suggestions. Maybe asking for payment details up front isn’t the best solution, but the idea of asking for a unique piece of information that can effectively identify a repeat submitter feels like the way to go. New submissions could be run past a database of previous entries and most bogus submissions would fall at the first fence.
Ashley also suggested that perhaps, this too shall pass. No doubt it will. Just as scientific journals are learning to deal with paper mills and plagiarism, so will these markets – those that survive. The thing is, these magazines, from the professional markets to those who can only afford token payment, are the life-blood of the short story world. To see them being pressured like this is shameful.
Of course, I’ve had a play with ChatGPT. It wrote me a story, and it was rubbish – simply an idea without a single word of dialogue, and a deeply unoriginal idea at that. When I criticised the story ChatGPT apologised. What a complete sham. This software neither knows nor understands concepts such as criticism, offence, or apology. And yet it has been designed to masquerade these and other behaviours with apparent sincerity. It’s a CreepFake.
As has been said elsewhere, the real threat from AI is our own gullibility.
We need to think very carefully about this before it is too late. And too late it may be quite soon.
David Gullen is a two-times winner of the British Fantasy Society Short Story competition and his work has appeared in The Best of British SF 2020, and 2021. Other work short-listed for the James White Award and placed in the Aeon Award.
Born in Africa and baptised by King Neptune, David has lived in England most of his life. He currently lives behind several tree ferns in South London with his wife, fantasy writer Gaie Sebold, and the nicest cat you ever did see. Find out more at www.davidgullen.com.
Reblogged this on Loving Life in the Rain and commented:
A worrying turn of events!