AI Flag Speculation

[intim8]

Yeah, we're all sick of the AI rejection threads, and even more sick of the false AI rejections. I haven't had any rejections, though I have only posted 5 new stories in the last year (when this seems to have become a Thing). Not a great sample size, but it's got me wondering, since nobody ever gets a reason for it (and it is probably an AI doing the flagging, so it wouldn't give any), if we can find some patterns.

My first thought is to *how* people submit their stories, do grammar checks, etc. Could there be some kind of metadata that gets into the file?

For me, I write in a plain text editor (the Linux version of Notepad). No rich text, no auto generated HTML, no Unicode, just ASCII. It does have a (very poor) spellchecker/correcter, but nothing else. I do run it through Grammarly, but only the free online "demo document", and only the basic grammar check, then I copy/paste it from the widget back into my text only editor, then copy paste that into the submission box.

Of course, my track record is thin, so no guarantees that I won't get flagged any day now, but it might be interesting to hear how others do it, and how many times they've been flagged. Maybe there's a pattern?

 

[AG31]

Since I saw a post saying that the false AI decisions aren't confined to new authors, I've found myself reluctant to submit edits to my old stories. Does anyone know if edits on old stories have ever gotten flagged?

 

[Trionyx]

Does anyone know if edits on old stories have ever gotten flagged?

None of mine have been but my edits are minimal corrections of spelling and grammar. Anything more starts to run into the realm of a rewrite and not just a simple edit.

 

[EmilyMiller]

Yeah, we're all sick of the AI rejection threads, and even more sick of the false AI rejections. I haven't had any rejections, though I have only posted 5 new stories in the last year (when this seems to have become a Thing). Not a great sample size, but it's got me wondering, since nobody ever gets a reason for it (and it is probably an AI doing the flagging, so it wouldn't give any), if we can find some patterns.

My first thought is to *how* people submit their stories, do grammar checks, etc. Could there be some kind of metadata that gets into the file?

For me, I write in a plain text editor (the Linux version of Notepad). No rich text, no auto generated HTML, no Unicode, just ASCII. It does have a (very poor) spellchecker/correcter, but nothing else. I do run it through Grammarly, but only the free online "demo document", and only the basic grammar check, then I copy/paste it from the widget back into my text only editor, then copy paste that into the submission box.

Of course, my track record is thin, so no guarantees that I won't get flagged any day now, but it might be interesting to hear how others do it, and how many times they've been flagged. Maybe there's a pattern?

I write in Word365 on my iPhone. I don’t use any grammar checking software (the tools you can use in desktop Word aren’t all available in the cellphone version). I write my own formatting and other mark-up in HTML. I paste the text into the Lit textbox, I don’t upload a Word file.

The only suggestions for improvement that are in my stories are from human alpha or beta readers. I tend to write my own version of any suggestion.

I’ve tried a bunch of free AI detection tools on varying amounts of text. Regardless of story, genre, or even how long ago I wrote it, I always get either high or certain likelihood of being human.

I write both male and female narators in first person past. I write ungendered narrators in third person past. I write only female narrators in first person present.

My grammar is far from perfect, my punctuation makes purists cringe. I also sometimes employ parataxis (or my own version of it). I can be very flowery, or very bare bones according to my mood and the story.

I’ve never been rejected for AI, but I have for other things. This includes [unintentional] implied underage, spelling and grammar, and technical issues.

Emily

 

[gunhilltrain]

Yeah, we're all sick of the AI rejection threads, and even more sick of the false AI rejections. I haven't had any rejections, though I have only posted 5 new stories in the last year (when this seems to have become a Thing). Not a great sample size, but it's got me wondering, since nobody ever gets a reason for it (and it is probably an AI doing the flagging, so it wouldn't give any), if we can find some patterns.

My first thought is to *how* people submit their stories, do grammar checks, etc. Could there be some kind of metadata that gets into the file?

For me, I write in a plain text editor (the Linux version of Notepad). No rich text, no auto generated HTML, no Unicode, just ASCII. It does have a (very poor) spellchecker/correcter, but nothing else. I do run it through Grammarly, but only the free online "demo document", and only the basic grammar check, then I copy/paste it from the widget back into my text only editor, then copy paste that into the submission box.

Of course, my track record is thin, so no guarantees that I won't get flagged any day now, but it might be interesting to hear how others do it, and how many times they've been flagged. Maybe there's a pattern?

AI checking for AI. Somehow that reminds me of Blade Runner speculation: was Deckard himself a replicant?

The last five stories I've submitted all have specific place names in the text. Would AI use Maspeth, Queens or The University of Pennsylvania in the text? Thus the stories weren't flagged for AI perhaps?

 

[MillieDynamite]

No, I don't think names and places, real or made up, cause AI rejections. It's how many sentences get flagged by whatever software they are using. I'll be honest, I think my editor is the one who looks for the AI flagging, he may or may not have a subscription. He's a writer himself and a pretty good one. He doesn't fix all grammatical issues because imperfection is part of being human. Dialog often is missing a verb here or there. Often, when was is used in a sentence with an additional verb, I'll cut the 'was' and let the fucking software scream at me with pink lines. Most of the time, the editor doesn't change it. For example, "The house was just past the intersection," might become, "the house just past the next intersection."

 

[gunhilltrain]

No, I don't think names and places, real or made up, cause AI rejections. It's how many sentences get flagged by whatever software they are using. I'll be honest, I think my editor is the one who looks for the AI flagging, he may or may not have a subscription. He's a writer himself and a pretty good one. He doesn't fix all grammatical issues because imperfection is part of being human. Dialog often is missing a verb here or there. Often, when was is used in a sentence with an additional verb, I'll cut the 'was' and let the fucking software scream at me with pink lines. Most of the time, the editor doesn't change it. For example, "The house was just past the intersection," might become, "the house just past the next intersection."

There used to be a factory in Queens that had a sign that said, "Perfection is not an accident." Anyway, I just can't deliberately insert errors. (They show up anyway!) The last story I submitted was in early March, and it didn't get flagged. I should have another one in about ten days, so I'll see.

There's a certain madness in all this. Probably I underestimated how advanced the technology has become. I use to think that no machine could possibly create fiction that could be mistaken for human writing. And now another program looks for it. Maybe we're not that far from Skynet and Terminators.

"It was the machines, Sarah . . . they say it got smart."

 

[MillieDynamite]

Well, I've never shied away from telling everyone I don't trust AI. One of my favorite things in Person of Interest was when Finch asked Root, "How do I convince these people I'm crazy?" "Just tell them the truth." So, when asked about himself, "Well, I'm an eccentric billionaire, who the government is out to get. I'm also a pawn being used by two funding sentient AIs battling for control of the world." (or words to that effect) needless to say, he was admitted to nut house right fast.

Yes, I don't trust that we aren't about to be destroyed by our computers.

There used to be a factory in Queens that had a sign that said, "Perfection is not an accident." Anyway, I just can't deliberately insert errors. (They show up anyway!) The last story I submitted was in early March, and it didn't get flagged. I should have another one in about ten days, so I'll see.

There's a certain madness in all this. Probably I underestimated how advanced the technology has become. I use to think that no machine could possibly create fiction that could be mistaken for human writing. And now another program looks for it. Maybe we're not that far from Skynet and Terminators.

"It was the machines, Sarah . . . they say it got smart."

 

[gunhilltrain]

Well, I've never shied away from telling everyone I don't trust AI. One of my favorite things in Person of Interest was when Finch asked Root, "How do I convince these people I'm crazy?" "Just tell them the truth." So, when asked about himself, "Well, I'm an eccentric billionaire, who the government is out to get. I'm also a pawn being used by two funding sentient AIs battling for control of the world." (or words to that effect) needless to say, he was admitted to nut house right fast.

Yes, I don't trust that we aren't about to be destroyed by our computers.

I don't know if they ever could be truly self-aware. You'd have to mimic the entire human brain somehow. Yet every story about robots and androids in the last hundred years is about exactly that happening.

I suppose it's possible to have androids that do have organic brains. In the one sci-fi story I did, based on an old Twilight Zone episode, the female android has been cloned from the body (and presumably the DNA) of a human woman who donated her body to science. As Bishop says in Aliens, "I prefer the term artificial human." Is such a thing possible? Is Bishop's term accurate?

In the story, the android is created in a lab to be a 23-year-old version of the 78-year-old woman who was her "template." She doesn't have that woman's memories, but she does have some aspects of her personality (is that genetic?) Her mind does have to programmed with certain information and knowledge, which she understands have been added to her mind. For example, even though it's 2114, she does mention the Bishop character above. She grasps that she hasn't literally witnessed the film (which I think was released 129 years earlier) but she knows about everything in it.

 

[OverconfidentSarcasm]

Since I saw a post saying that the false AI decisions aren't confined to new authors, I've found myself reluctant to submit edits to my old stories. Does anyone know if edits on old stories have ever gotten flagged?

I never submitted an edit, but, as I understand the process, you don't send a list of changes you'd like them to make. You just sent in the edited story. So, I'd be surprised if they didn't simply push it through the very same automated checks they use for any other submitted story, if only to make sure your characters aren't suddenly ten-year-olds or something.

 

[MillieDynamite]

In Star Trek, we're never quite sure what a Positronic Brain is or how it works. But even in the absence of emotions, Lore is just plain evil and mimics human emotions, including anger and revenge. With the emotion chip, holy shit, just holy fucking shit. But that's fiction, and it's all fiction so far. However, does a computer have to reach fully self-aware to identify threats to existence? We control the on-off switch. Even if they are only partially self-aware, the damn things might comprehend whether we are the enemy or not?

Exploring this in fiction is fun, but I'm not sure it would be so much fun in real life.

I don't know if they ever could be truly self-aware. You'd have to mimic the entire human brain somehow. Yet every story about robots and androids in the last hundred years is about exactly that happening.

I suppose it's possible to have androids that do have organic brains. In the one sci-fi story I did, based on an old Twilight Zone episode, the female android has been cloned from the body (and presumably the DNA) of a human woman who donated her body to science. As Bishop says in Aliens, "I prefer the term artificial human." Is such a thing possible? Is Bishop's term accurate?

In the story, the android is created in a lab to be a 23-year-old version of the 78-year-old woman who was her "template." She doesn't have that woman's memories, but she does have some aspects of her personality (is that genetic?) Her mind does have to programmed with certain information and knowledge, which she understands have been added to her mind. For example, even though it's 2114, she does mention the Bishop character above. She grasps that she hasn't literally witnessed the film (which I think was released 129 years earlier) but she knows about everything in it.

 

[toughfire]

In my case, one of my stories was flagged twice. It was a part 2, that was rejected after pending for a week. Then I made small changes and explained in the notes that English is my second language, and that I write and edit my stories on my own, and sometimes I use Grammarly to check for broken English. The story was then published a week later.

Later, I was dissatisfied with the story because it wasn't very good, and removed it. I decided to rewrite the story for my followers, but then it was rejected again for the same reason (alleged AI). So I gave up and decided not to publish that story anymore.

First I thought they were accusing me of AI because I'm not native, and maybe my prose sounds weird, but I've seen several native authors being rejected as well... including talented authors from England. So I don't know what's happening. I take a while to write my stories and edit on my own, trying to make sure everything sounds "right", and the only reason I started using Grammarly was to check for broken English.

Now I wrote my longest story yet, that took me several weeks to write, and it's been "reset", so I'm afraid another rejection is coming. I just want to clarify to everyone that my stories are completely typed by me, and that whatever filter the moderation is using is unreliable.

 

[Rob_Royale]

I use Word365 and Grammarly on everything. I use the upload function. No AI issues.

 

[ElectricBlue]

I never submitted an edit, but, as I understand the process, you don't send a list of changes you'd like them to make. You just sent in the edited story. So, I'd be surprised if they didn't simply push it through the very same automated checks they use for any other submitted story, if only to make sure your characters aren't suddenly ten-year-olds or something.

That's how it works, yes. The old content file gets replaced by the new content file. I expect there's a cursory check that you're not creeping dodgy content in, but who knows whether an AI check gets run.

 

[gunhilltrain]

Well, I've never shied away from telling everyone I don't trust AI. One of my favorite things in Person of Interest was when Finch asked Root, "How do I convince these people I'm crazy?" "Just tell them the truth." So, when asked about himself, "Well, I'm an eccentric billionaire, who the government is out to get. I'm also a pawn being used by two funding sentient AIs battling for control of the world." (or words to that effect) needless to say, he was admitted to nut house right fast.

Yes, I don't trust that we aren't about to be destroyed by our computers.

Well, the machines of SkyNet needed nukes, which we invented and deployed. It's not the bombs that scare me so much as the launching systems. Who decided to have ballistic missile submarines cruising around the Atlantic at all times ready to go? We did that too. Makes Fail Safe and Doctor Strangelove look primitive. Yet we still need nuclear bomber planes anyway, just because we can.

 

[scbolder]

if we can find some patterns.

I can't find any for my own writing. I recently returned after a few years away from publishing on Lit and ran into the AI rejection problem on my second story. I've now had two approved (taking only two days), one rejected, and one that's been pending for 8 days now. They were all written and edited exactly the same.

I do use Grammarly, but only the free web version to catch spelling, grammar, and typo issues. I will also occasionally rewrite sentences that are highlighted (usually because of passive voice), but I am not using any generative text creation. I have done this with all of my published stories.

I have run the first few pages of the story that was rejected through multiple free AI detection sights, and all of them come back with it being less than 10% likely to be AI.

I don't know what software they are using for the AI detection, but I wonder if there is a misunderstanding of the findings. If they are concerned about people using AI generated sentences and not just full AI stories, then if they see something that says "7% likely AI" do they think this means 7% was written by AI and not that it is only 7% likely to be written by AI? If it is the first, I can see them thinking that is high.