Am I a writer if I use AI?

So, a member PM'd me a few days ago asking about the fact that I use AI. I won't say who it was, but it's been bugging me for a few days.

The first question was "Does it steal ideas from you?"

My answer:
"No, it doesn't steal ideas. That would be impossible anyway, since every idea has already been done in one form or another. It does not, contrary to popular belief, copy and paste lines or fragments of lines and mash them into suggestions from writers. It cannot, for example, reproduce the text of most novels unless the novel is *especially* famous, even if it knows the plot. It can, for example, tell you about specific passages from, for example, Shakespeare, or the beginning of Moby Dick, but it has to be at that universally classic level for it to do so.

If you write a story about vampire giraffes, it learns that the word "giraffe" is more likely to appear close to the word "vampire" in certain contexts. It enters that information into its database, making the probability that "giraffe" will appear near "vampire" increase by a vanishingly small amount - we're talking millionths or billionths of a percent here."

Question: How do you use it? Brainstorm ideas?
"Yes, I use it to brainstorm. I do not use it to suggest ideas. I feed it ideas I already have, which it then posts a response to. I then ignore that response, but something it says almost always triggers a tangential idea, which I then expand on myself. Then I feed that new idea back to it, and so on."

A further addendum - It's like the "What's the first thing that comes to your head" thread. It's an association game - the AI says something, and that triggers a thought which can be expanded further.

And the last question, the one referenced in the thread title and that really did... I dunno, upset? annpy? trigger? me was "Do you see yourself as a writer even if you use it?"

My answer:
"Of course I see myself as a writer. All the writing is mine. All the prose, and the ideas, the plot and the execution is mine. I might have used one word in five chapters that the AI used, because I like it, but I never take whole sentences. Nor do I simply substitute that word into an existing sentence. I write an entirely new sentence incorporating that word. Why wouldn't I see myself as a writer? I can write far better than an AI can, because it's my voice, not one that is an amalgam of a mass of data. I can come up with better ideas, because I can be original. The AI can only come up with very generic ideas, which are so generic as to be unusable, not to mention those ideas don't take into account the story, tone and focus that I'm trying to achieve. It would take a LOT of effort to make it do that, and if I'm going to expend that effort, I might as well just write it myself.

It's no different than using a thesaurus, or reading a Wikipedia article and clicking on a link that catches your attention, except that the thought sparked by the AI can be much more tangential than a Wikipedia article, and is often completely unrelated to whatever I was talking to the AI about in the first place."

So this is for the record. There are lots of good and legitimate ways to use AI. Being a critic of using it is fine, but at least try to really understand how it works and what it actually does rather than having a knee-jerk reaction based on misleading headlines.

Sure, you CAN use it to generate a whole novel for you, just like you can use a fork to dig a pit. It'll work, but the result won't be anywhere near the same as if you used the right tool, or used a tool for the purpose it was designed for.

 

It's a trigger. Sometimes it will suggest that a bit here needs to flow better or be quicker, or need some more humour. I might agree and do it, but I won't use any of its suggestions on how to do it.

Or if I'm plotting, almost invariably, it will say something about or make a suggestion about a scene that triggers an indirectly related thought, that then balloons into an idea. It's a sounding board, same as receiving a critique, but live. Being live, it allows you to develop a train of thought much faster than posting it and seeing if someone responds.

You can't use its responses directly, because they're just too generic. And also, if related to a character, they're usually not keeping in with that character's personality. So for example, I had a character making some cutting remarks to her partner in the opening scene. The AI started suggesting she do it in every scene I gave it, which would have turned her into a totally snarky bitch, instead of maintaining the playful tension between them.

 

It would be interesting if you could give a specific example of how it said something that triggered a thought that ballooned into an idea.

 

I'll try and find one, since I have all the chats archived. Quite a lot of long conversations for me to trawl through, so give me some time.

 

But as an example of how meandering conversations can go, last night, I was thinking about dinner, and got a recipe out of ChatGPT. That led me to ask it about tomatoes, which led to scientific names and Carl Linnaeus, which led to the pronunciation of IVLIVS CAESAR in Classical Latin, which in turn led to Vulgar Latin vs Ecclesiastical Latin.

My brainstorming sessions would have similarly meandering conversations.

 

While I don't know the first thing about using AI in any sense past "google do a search for me," I am of the opinion that using AI as a springboard is no different than having some other kind of source of inspiration (for example, prompts in book form, articles in a magazine, simply living life outside the house). I would say that if AI writes the entire book/piece for you and then you call it your own, that feels dishonest and slovenly, but that isn't what I'm getting from you here. I think the fear (in my own mind) comes from AI being used to generate work with malicious intent behind it. Someone "stealing" or something. But a lot of that comes from my own ignorance of the actual functions of AI.

You are still a writer because you add nuance only a human brain can. There are some writers today who are big names and I don't have an emotional connection to their work, even though it's brilliantly written in terms of construction and characterization. The writing still manages to feel superficial in some way. That's what I feel AI writing is like. If a writer is able to connect the words to something deeper than an algorithm, I see no reason why you'd be less of one for getting inspiration however you choose.

 

So here's an example of how one interaction went - I had the final scene in mind for a story, where my MC (a space pirate) and his on-and-off love interest (a space cop) say goodbye at the end of another caper. I gave the AI the outline of the scene. It suggested a parting line for the cop where she playfully suggests that next time, she'll arrest him as they make their way off the station where they've just pulled off a rescue mission and heist. It gave me some suggested dialogue, which I ignored. It's not that the dialogue was bad, per se, it was just too bland and generic, something that you might see on any work, said by any female character.

That made me think that the ending was too pat and abrupt. It needed spicing up with something more than just some dialogue between the two characters. What the AI had suggested wasn't in keeping with the relationship between them - the MC wasn't just going to walk off with a "Sure Princess, come and catch me, kthxbye."

So instead, I had the MC plant the incredibly valuable statuette he's just stolen on her ship, leaving the cops to explain how said incredibly valuable stolen statuette had come into their possession, and leaving him completely off the hook for the entire thing, and in control of the outcome. The AI's review of the scene had also made me realise that I hadn't actually given the MC a good reason to steal the statuette in the first place, other than that it was valuable and it happened to be there - and in fact, had played no important role in the story whatsoever.

 

I disagree. I asked GPT to write about walruses in the style of Ayn Rand. In the first paragraph (or second, I can't remember) it named one walrus 'John Galt.' It doesn't know what it is and isn't copying. It can't know. Yes, every idea has been 'done in one way or another' but that's only if you reduce it to its base elements, which isn't what people are talking about when they compare one story to another, or praise one's creativity. It's the specificity that grants value, makes us read one book and not another, and makes us seek out other stories.

This might be two questions.

1. Am I a writer?

2. Was it entirely I that did the writing?

If I was habitually uncharitable (okay, I am, but not for this topic), I would say you're getting the AI to write the rough/first draft of the scene. It's spitting out a generic framework for you to build off of. In the same fashion, if you paid/asked someone to produce a very rough first draft of your scene, then you used that as an inspirational launch pad, would you still take full credit for the final work? The difference is one collaborator is a person, while the other collaborator is a predictive machine trained on human work, but the outcome is a rough draft either way.

To the original question, I say you're a writer. However, that term doesn't hold as much power over me as it does others. When people try to make that term exclusive by obsessing over the fine details of set membership, like is happening now, well they're mainly just procrastinating and should all go suck on a cob of corn since it will be a better use of their time. I'll just use a different word: "No, I'm not a writer, I'm a booksmith." If you create a story even entirely using an AI, the general public won't bop each other over the head if someone says "writer" instead of "curator" or "AI operator." They'll still call you a writer and get on with their lives.

I'd say it's more debatable, or at least not as pointless, to question whether it's entirely you that wrote the scene. For the record, I still don't care either way. If you do care, that's where you should focus your philosophy IMO.

My refusal to use AI in any writing capacity is I'm worried it will make me think less. If it gives me a rough draft, the bones will be too formulaic for me to make anything compelling with it, no matter how much hot sauce I splash on top. If other people claim it makes them think more, I'm not going to disagree with them, but I also can't relate.

 

And you will find, that if it is a very well known author, it will sometimes recycle names. So it called a walrus John Galt. So what? Did the rest of the story take any specific details of plot or actual written work, phrases or otherwise from Rand?

It has almost certainly noticed a strong relationship between the words John+Galt+Ayn+Rand. And that's literally almost all it's doing - when asked to write something in the style of something else, it will just use the relationships it knows of with the kind of words that the other person uses. Ask it to rewrite the Bible in the style of a pirate, and all it does is stick a few pirate related phrases in there like "Arr, me hearties.".

It cannot *really* write a story in the style of a particular author, like the previous short story contest, because it doesn't *know* what a plot, or a scene or character is. It just knows about relationships between *words*.

 

Am I? And how does that work, given that I didn't ask it to plot out a scene for me in the first place? I gave it the scene that *I* plotted and designed. It then commented on the scene, that prompted some ideas, as I explained above. How is that it coming up with a framework?

And how is it different from posting a piece for critique in the workshop, then someone making a comment on it, and their comments leading me to consider a revision of some parts of the piece, for whatever reason? Is that the critiquer coming up with a generic framework for me?

It could entirely be my fault for not explaining how I work with it very well.

 

What I personally don't understand is why you have fixated on this idea of a "rough draft", when I made it clear that all the plot ideas - including the rough draft of the scene - came from me, not from the AI.

And I'm certainly not the only one who uses AI in this way.

Whether or not an AI will make you think more or think less entirely depends on how you use it. It's an aide, not a crutch.

 

Hey now, if we're going to blame anyone for anything, we can start with all the hazardous things I inhale during work and hobbies.

This is where I got the notion that it's creating a rough draft for you. You gave it an outline:

So I compared to asking, say, a ghost writer to come up with a scene from some plot beats and an outline. Why would it suggest dialogue if not filling in the blanks? Did you specifically ask it to make suggestions about the dialogue? What did you actually prompt it for, and what did you actually provide?

By framework I'm thinking of the meat. For example, the prompt might be: write a scene where James loses his cutlass duel to Ghost Pirate Orangebeard, but he never actually planned to win in the first place. It was all a ploy to give Hook Hand Alice time to line up The Great Bean's starboard side cannons with the crystal statue that is Orangebeard's phylactery. That's an outline with beats, but there's nothing really there yet. If the AI can map out what happens in between those beats, IMO that's what's creating the first draft, however rough, in the same way you might provide those initial details to a collaborator for example.

For science, I tried it again. No name-dropping this time, unless Odak is a Randian name I'm unaware of, but I think a blind test would leave the reader thinking of Ayn Rand (even though this excerpt is high in summary).

Prompt: write a story about walruses in the style of ayn rand

 

I asked it to provide feedback on how well it fit in with the rest of the story so far, some of which had already been written. I didn't ask it to come up with any dialogue.

It may well be an outline with beats, but at no point did I ask it to, and nor did it "write a scene". If you want to be specific, I asked it something along the lines of:

"Here is what is going to happen in the final scene, how does this sound as a way to end the story?" or something along those lines. It will *always* be positive about whatever you give it. It's up to you, the user, to use whatever result is to refine or develop the scene.

Usually, the AI will misunderstand the scene at first, because I haven't given it enough detail of what's happening and why it's happening. That will be obvious in its response. I will then explain it in more detail, which in turn makes me put those details down on paper with more refinement and discuss them further.

In this particular case, I was misremembering what happened. I reviewed the chat again, and this is how it went. I started prior to the final scene with the MC's partner disabling the security system and being chased by an army of droids. In summarising that scene, the AI called it a "distraction", and that they could use it to escape.

I realised that the AI had misunderstood the purpose of what the partner was doing, but I liked the idea of a distraction. So I came up with a different idea - a high stakes poker game on the casino floor, at the climax of which, the partner would burst in, followed by the army of droids in from the previous scene. It suggested that the partner could "scramble the bots' commands to throw up another obstacle". OK, so I liked that idea, and came up with the idea that the partner could infect the security droids with serving droid protocols, causing them to spread into the crowd offering drinks with all guns blazing.

That, in turn, leads the AI to comment on them leaving the casino unnoticed while the chaos unfolds. I didn't like the idea of them casually walking out, so I had the FMC use her royal status to bluff their way out past reception to their getaway, and the MC suggesting to reception that they both "have plans" tonight.

This leads the AI to suggest that, when they part, the FMC tells the MC that those plans will include pursuing charges the next time they meet. I liked the idea of building some kind of (playful) threat into the scene, but actually having her do so puts the focus on her, rather than the MC. So instead, I had her on the verge of doing so, but being forestalled by the MC telling her he's left her a present, and then her finding the valuable stolen statue secreted on her ship, and her acknowledging that, once again, he's come out on top.

So what did I get out of it? The only thing I got directly was the idea that the partner scrambles the bots' commands, and even then, the details of what actually happens came from me. Nonetheless, everything else developed from ideas sparked by what the AI had said, even if I incorporated those ideas by doing the exact opposite of what it was suggesting. So in the end, I ended up with a detailed scene that was built with the help of AI feedback, but with only one general element that was actually suggested by it.

But maybe that's just how my mind works, and the approach won't work for other people.

 

I'm not familiar with Ayn Rand's writing, so I can't really comment.

 

I have no skin in this game and no stake in this argument. I also have never read Ayn Rand. I'm simply a neutral observer.

I simply googled "John Galt", and the first response (from Wikipedia) was, in part:

My reaction? OK, fine. I'll probably forget all this by dinner time. By the GPT probably won't. If it was that easy for me to connect "John Galt" to "Ayn Rand", I can only imagine that -- for GPT -- it would've been much, much easier. He's not an obscure character.

Since GPT calling a walrus "John Galt" doesn't impress me, I agree with Naomasa: it called a walrus "John Galt". So what? *shrugs* Sorry, Territory.

 

Hey, I'm the person in question.

I'm so sorry I made you feel that way. It wasn't my intention to upset you.

Personally, I don't mind it if you use AI to bounce ideas off of, or brainstorm. AI should be used as a tool -- and as with any tool, to be used wisely.

My own worry is that if I use AI, I risk letting it decide everything for me, like:

Let's say I go to Chat GPT, give it a basic summary of my plot and character, then ask it:

"OK, what should Amos (my main from my Colonial WiP) do here?"

AI: "He could do this, this, that, here are pros and cons."

Even if I altered the ideas it suggests, I would still feel icky because it was an AI that gave me that idea to begin with. As for stealing ideas, perhaps I should've clarified it further.

Let's say I feed it this scene:

Spoiler: An example scene from my WiP

“Informants work on behalf of their client — be it a law enforcer, a consultant like myself, or even a government body. As such they must follow proper protocols and guidelines to ensure the overall investigation runs as smoothly as possible. The objective is the key priority, Amos. Nothing more, nothing less. We can look out for each other, we can even have a bit of fun, but what we do must never, ever, put the mission in jeopardy. You understand me?”
Amos slowly nodded.
“But don’t worry, Amos,” Nathaniel added with a smile, “you won’t be alone. I trust you’re familiar with Darryl?”
A look of recognition flashed over his face. Surprise. “D-Darryl?” Amos’ voice was rising. “He’s your informant?”
Nathaniel shushed him. “Quiet now! When the need arises, yes. He’ll be your guide and instructor. Do what he says, exactly what he says, and don’t argue with him.”
Amos nodded. “Understood.”
“And Amos,” Nathaniel said, “if it ever comes down to it — if he tells you to run, and don’t look back—”
“I’m blind, sir,” Amos said.
Nathaniel sighed, half-wondering if the magistrate actually had a point. The lad is too impulsive… “You do what he says, Amos. If he tells you to wash your hands of all this and return to your life as a tavern boy, you do exactly that.”
Amos frowned. “Yes, sir.”
“Good. Good.” Nathaniel collected his journal, stood.
He was halfway out of the threshold to the main common area when he heard Amos.
“Sir?”
He turned around. The boy shuffled closer, a hand partially raised. Remarkable, how he knows his way around… the thought came unbidden. Amos held out his other hand, ostensibly to shake Nathaniel’s. “Thank you, sir.”
He grabbed the boy’s hand, shook tightly. “A pleasure. A contact will be with you soon.”
Amos’ eyebrow raised. “Not Darryl?” He kept his voice low this time, fortunate.
“A proper consultant has more than one assistant, Amos. Farewell.” His boots squeaked on the floorboards, leaving the newly minted informant to his thoughts.
He stepped outside, inhaled, looked at the sky. Jayson once again resurfaced. The bright, youthful man with dreams of becoming a proper consultant in his own right.
Jayson…
Help me protect Amos…
Nathaniel’s fingers tightened into the leather of his journal. He remembered Dunbar’s warning.
I will not allow Amos to suffer your fate. Or may God damn my soul.
He shuddered, walked down the cobblestones, lost in his thoughts.

Would the AI steal that scene and characters or...

That's what I was confused on.

Again, very sorry for making you feel that way. Wasn't my intent.

 

Then don't ask it that. Instead, ask it for its impression of what your character is like. See what it says, and see if that is what you were trying to write your character as. If it is, the AI's character summary will inspire an idea to you as to how your character would react in that situation.

Even if the AI has given you an idea, unless you've asked it to give you a detailed scene with a step-by-step description of what happens, it's still upon you to develop the scene. You would only use the AI to give you a general direction, not write the whole outline of the scene for you.

In my discussion described above, the next scene came from one word that showed the AI was misunderstanding the purpose of the preceding scene - "distraction". It wasn't even suggesting that as the next scene, it was misunderstanding the *previous* scene as being a distraction (or it just used the wrong word to describe it). That's all that I needed to spark the idea for the next scene.

Have you ever read a random WIkipedia or news article and gotten an idea from it? It's no different to that.

 

No, it wouldn't. All it would do is build word associations. It would note that the word "informant" is very slightly more likely to appear in close proximity to "law enforcement". It wouldn't take your scene and actively suggest it to someone who wanted an idea for a scene, and it isn't going to lift whole sections of text from it. It won't spew out any text that you wrote word for word. It might, by sheer coincidence, produce a short sentence that as the same as yours.

The reason it can produce, say, "What light in yonder window breaks?" word-for-word is because the association between those words is so strong where the words "Shakespeare" and "Romeo and Juliet" are involved is because of the millions of times that particular line appears on the internet in proximity to those particular phrases.

 

Makes sense. I mean, just today I read a wiki article about how Claudius became Emperor by accident, because everyone thought he was a joke due to percieved weaknesses related to his disabilities (that he got in early childhood.) I started forming an idea for a character I could use based off that in my story.

Aaah, so in other words, if someone used Chat GPT to help develop the storyline and the characters, they don't need to worry about Chat GPT actively sharing it with someone else? I've asked it if the chats were private, and it said yes, so long as the function to allow it to be monitored for AI training was turned off.

 

That's exactly right. Now, with Wikipedia, if you see a link that interests you, you can immediately go to that article. With ChatGPT, if you were discussing a character or a historical figure that you're basing your character on, you can just ask it about that, allowing your discussion to roam. That's one aspect of doing research with AI, but if you need the answers to be accurate, always double-check.

ChatGPT can't actively share it. I can't ask it "What did Link the Writer discuss with you?", or anything like that. Even if you keep the training option on, all it learns are word associations, and any individual chat can only make a minute difference between the billions and billions of associations it already has. For every association you increase the weighting on, there are hundrds of thousands of others that are making associations that pull in different directions. It would be almost impossible for someone to reproduce your character ideas just through a chat with it.

I'll give you an example - if you ask ChatGPT to write you an original Sherlock Holmes story, chances are very high that it will write a story which includes supernatural elements, including vampires and/or giant rats. Those things only appear in a single Conan Doyle story, The Adventure of the Sussex Vampire, and the rat is only mentioned once. But so much fan and derivative fiction has associated Holmes with the occult and those two things, that ChatGPT thinks those elements are more important in association with Holmes than they actually are. It hasn't stolen any ideas, it's just learned that the words "vampire", "Sumatran Rate" and "Holmes" very often appear close together. It doesn't "know" that Conan Doyle only mentioned them in a single story.

 

So in other words, Chat GPT is basically the bottom low-risk AI to use if you wanted to use it to bounce off ideas/play with your characters and not have to worry about your ideas being taken.

 

Correct, because it's not a database of text. It doesn't store all the stuff it's been trained on. It stores the connections between words that it's been exposed to.

 

Okay, that example illustrates it well. I see what you mean in that you're not having it write the first draft for you. You're using it incite brainstorming, and its ability to regard your specific plot must be working well for you.

You're a writer, and it's you doing the writing. I wouldn't consider a friend you explain the plot to ultimately be a collaborator either. Many a writer's wife is just that, but she's usually in the acknowledgements rather than the cover, and even that's not a requirement.

Some writers thrive by having someone to talk to about their story, others just need a lot solitude. I'm in the latter camp, but it's all just process.

In spite of what I've just said, I'm reluctant to use any human comparisons to LLMs, so I'll have to add that this is still different than a human sounding board in various functional ways. It always will be. Does it matter here? No, but it's also important to not forget.

 

I'm not suggesting that anyone must, or should use it. That's a personal choice.

But what I am saying is that just the fact that the writer uses AI at all, is not a reason for that writer to be stigmatised - depending on how it is used, and AI being involved, does not mean that the writer is using AI to actually do the writing.