Guide:Generative AI in your Writing

From Bravo Fleet
Revision as of 10:15, 16 February 2024 by Beckett (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
You must always be the author of the stories you publish on BFMS

Bravo Fleet has clear policy surrounding the use of generative AI in your writing. By ‘generative AI,’ we mean large language model chatbots such as ChatGPT and Gemini. It is well understood in Bravo Fleet that you cannot use this generative AI to ‘do your writing for you.’ This does not mean, however, that you cannot use AI in your writing at all. This article guides in interpreting our Creative Integrity Policy regarding using AI in writing and how one can appropriately harness AI.

AI to Write Your Stories

Believing one cannot use AI at all in one's writing is a misunderstanding of the Creative Integrity Policy. It is easier to view the policy as being about authorship and plagiarism. Simply put, when you write for BF, earning activity ribbons and being rewarded for your effort, it is assumed that you are the author of that writing. Posting a story you had ChatGPT generate for you is, as such, no different to posting a story someone else wrote for you. Both instances would see you rewarded for writing you didn’t do, which is against policy and against the spirit of BF.

Generative AI vs Writing Assist Tools

Generative AI creates content from a user prompt, using its access to source texts to remix them into new arrangements while writing assistance tools such as Grammarly and ProWriteAid make adjustments to a writer’s existing text. (Grammarly now also has a Generative AI feature.)

Where it gets complicated is that you could ask ChatGPT to edit an entire story for you all at once, by pasting a story in and then asking it to proofread for you. You shouldn’t do this, for a lot of reasons:

  1. With an actual writing assistance tool, you have control over what changes to accept and what to reject; as grammar rules are not binary or absolute, each instance needs to be looked at carefully to make sure that your meaning isn’t being altered.
  2. Some grammatical fixes will involve inserting words, so you should be doing this yourself by accepting changes or making fixes, rather than letting ChatGPT mess with your wordcount.
  3. Once you put your text into ChatGPT, it trains the language model, which means that your phrases could appear in other peoples’ responses from ChatGPT, so you lose control of your intellectual property.
  4. You will never learn how to avoid making mistakes in your writing if you have ChatGPT edit for you, as you can’t go through and see specifically what went wrong without a lot more extra work.

The bottom line is that Generative AI makes new content while writing assistance tools help you edit your own writing.

AI as a Writing Tool

None of this is to say that you cannot use generative AI such as ChatGPT in your writing. On the contrary, it can be a powerful tool to help you generate ideas, develop storylines, or add detail. This section gives several examples of good practice in using AI in your writing and explains why this differs from breaches of the Creative Integrity Policy.

AI for Treknobabble

Sometimes we all need a little help making our Starfleet characters sound clever

Writing in Star Trek sometimes means writing characters coming up with technological solutions or scientific explanations surrounding the challenges they face. Many of us do not have the same scientific backgrounds as our characters and when developing scenarios or character choices, we may have to do a little research to add some authenticity to our writing. This might constitute Googling issues, hitting up Wikipedia articles, or consulting Trek canon to find similar examples from past episodes. This last can be particularly relevant as Star Trek is not a ‘hard scifi’ setting, and sometimes its scientific explanations are closer to magic.

Very often, it can take a lot of work to find only a few key phrases to make a character’s explanation of a situation sound convincing. This is an occasion where AI can be incredibly powerful in enhancing your writing. With the right prompts, it is capable of describing a scientific phenomenon in relatively simple terms or of explaining a feat of technological wizardry using language that would be at home in Star Trek.

Example

In a mission for the USS Endeavour, I wrote about a Borg Cube that had been destroyed in transwarp, which scattered its wreckage across a sector. This was an ‘original’ concept, meaning it was my idea, not derived from AI. It’s easy to understand and provides a story hook where ships must travel across an area to secure technology. But this was also a problem that my characters would need to study, comprehend and explain, and Memory Alpha articles about transwarp were a little light on jargon or explanation. I turned to ChatGPT for aid with the following prompt:

In this Star Trek fanfiction, a Borg transwarp conduit has collapsed. The Borg Cube inside it was ripped apart, its wreckage scattered across a sector. Give me some technological explanation of what happened when the transwarp conduit collapsed.

Here, I provided the specific genre: Star Trek fanfiction. I explained the scenario, making sure to mention the words ‘transwarp’ and ‘Borg,’ and asked for a specific output. Note that I did not ask for any prose where a science character might explain this to their captain. I received the following:

The catastrophic collapse of the Borg transwarp conduit resulted from a destabilization in the subspace infrastructure that facilitates transwarp travel. At its core, a transwarp conduit relies on a controlled warp field within a subspace tunnel to achieve faster-than-light travel. In this instance, a disruptive resonance cascade occurred within the conduit, triggering a rapid decay of the coherent warp field.

As the conduit collapsed, intense subspace shear forces tore through the Borg Cube that was traversing it. The localized subspace implosions caused by the collapse shattered the Cube's structural integrity matrix, disintegrating its superstructure into countless fragments. The violent release of destabilized transwarp energy further fragmented the ship's assimilation and propulsion systems, scattering Borg technology across the sector.

The collapse was akin to a cascading failure in the intricate web of spatial and temporal manipulation that underlies transwarp technology. The catastrophic event left behind residual subspace fractures and temporal differentials, creating hazardous pockets of distorted space-time in the affected sector. The remnants of the Borg Cube, now strewn across this unstable region, serve as eerie reminders of the destructive consequences when the highly regulated balance of transwarp subspace is disrupted.

A few key things stand out from this response:

  • ChatGPT explains its answer with terminology and a level of internal consistency that would be at home in a Star Trek episode. What’s a ‘disruptive resonance cascade?’ I don’t know, but I can see Spock saying it on the bridge of the Enterprise.
  • It provides some technological language I could use for my writing, but it also extrapolates in the final paragraph with the discussion of subspace fractures and temporal differentials.
  • This is three paragraphs of dense language and terminology and cannot be used in its current state for creative writing.

I used only a little of the exact text in my eventual writing, and even then, this was limited to basic principles and some key terms.

When I began to write my story, I did not use the text generated by ChatGPT directly, just the concepts it presented. An important principle of the creative integrity policy is that any text we include in a story should fundamentally be our own words. It also did not fundamentally change the story that my science officer could explain how localised subspace implosions shattered the Cube’s structural integrity matrix, but it made the scene feel more ‘authentically Trek’ that he could do so. It would have been unreadable tripe to have him regurgitate those three paragraphs verbatim, but reading them still helped shape the Treknological explanation of what had happened.

Here’s a cut-down extract from a Story on Endeavour when the captain of a science ship is explaining his crew’s findings to his superiors:

‘My science officer’s having a field day with the study, sir, I can tell you that,’ Xhakaza said brightly. ‘We’re learning considerably more about the nature of transwarp by studying the subspace disruptions from the conduit’s collapse.’

‘Is there any indication,’ Faust pushed, ‘whether the Cube was destroyed first and that collapsed the conduit, or vice versa?’

Xhakaza shook his head. ‘We know for sure that the conduit’s collapse triggered localised subspace implosions. Based on the reports from the Redemption, my officers reckon that shattered the Cube’s structural integrity matrix. It’s just not clear if the Cube was already in trouble when that happened.’

‘So we don’t know much,’ said Faust.

‘We know a little more. It’s not good news.’ Xhakaza sighed. ‘We had the reports from Endeavour about the Mesea Storm. My science officer agrees: the collapse of the transwarp conduit has left subspace fractures potentially across the sector. That probably caused the plasma field’s unpredictable agitation.’

Rourke worked his jaw. ‘Any other consequences of that?’

‘We should watch for pockets where space-time has been disrupted. That could be extremely hazardous.’

You’ll see only a few of the key terms from ChatGPT’s response have made it in. It’s been necessary to have characters ask questions and others reply, and the scientific terminology is largely window-dressing. But it evidences that Starfleet officers are studying and comprehending a phenomenon, and raises the tension by suggesting future dangers.

Generative AI is a tool - you need to use it properly to get the best results

A more impactful result of using ChatGPT was its extrapolation of the situation. I could have probably written some sort of Treknobabble about transwarp collapsing myself with the use of Memory Alpha and perhaps some VOY episode transcripts. That would have both likely contained egregious uses of the word ‘tachyons’ and taken a lot longer for, likely, worse results. What I hadn’t done for myself was extrapolate the idea of the ‘residual subspace fractures’ and the possible ‘pockets of distorted space-time’ as a consequence of the transwarp conduit’s collapse. That ‘idea’ was all ChatGPT’s.

Will this be an ongoing issue in the Midgard Sector? Will the subspace fractures impact travel across the region? Will Starfleet operations be disrupted by these distortions in space-time? I don’t know, but it’s an option that wasn’t there without ChatGPT.

Using ChatGPT to help me write some Treknobabble, I’m left with several key lessons:

  • State your parameters clearly. The more you give ChatGPT, the more it gives you. Say that you’re writing Star Trek fanfiction, and it will give responses more appropriate to the genre. You can ask direct questions or just for a technological explanation.
  • Quote the response sparingly. Not only is it necessary for adherence to the Creative Integrity Policy, but you’re looking to sprinkle your work with some key scientific terms - not regurgitate an AI-generated wiki page at your readers.
  • Generative AI provides only suggestions. Use what’s useful for you; ignore what isn’t.

This is different from getting generative AI to write your Story for several reasons. In this case, the core idea was my own. More pertinently, I used ChatGPT more like a research tool, albeit one which could generate hyper-specific answers relevant to the genre of Star Trek fanfiction. It was still necessary for me to read and interpret the response and then make authorial decisions on what to use and how to write and present the content.

AI for Mission Planning

Another way that generative AI can be useful while still within the bounds of policy is in plotline development. ChatGPT is capable of generating, from scratch, outlines for your missions with as much detail as you ask for. I prefer to use it when I have a broad concept but could do with additional ideas or for figuring out how to move the plot from A to B.

For example, in the mission above about collapsing Borg transwarp, the USS Endeavour was stranded on the far side of Romulan space with a long journey home ahead of them, venturing through the territory of the old Star Empire. While I had a few ideas of what I wanted to write, I punched that scenario into ChatGPT to see if it could conjure up anything else. Here’s the prompt I gave:

I'm writing a Star Trek fanfiction, set after the events of Season 3 of Star Trek Picard in the year 2401. The USS Endeavour has been catapulted by Borg transwarp outside of Federation space, into the distant Beta Quadrant. They're maybe a month away from Federation territory, and their most direct route back is through the territory of the old Romulan Star Empire. But the Star Empire collapsed after the supernova in the 2380s, and Starfleet has never been to this region of space. It's likely there'll be the remains of the fallen empire, Romulan warlords who've tried to seize regions for themselves, and untold mysteries on the USS Endeavour's journey home. Can you give me some quick story prompts for adventures the Endeavour might have?

And the reply:

Certainly! Here are some story prompts for the USS Endeavour's adventures in the Beta Quadrant:

Warlord Negotiations: The Endeavour encounters a powerful Romulan warlord who controls a significant portion of the former Romulan Star Empire. The crew must negotiate safe passage through their territory, but the warlord has ulterior motives that put the ship and its crew at risk.

Lost Research Facility: The crew discovers the remnants of a secret Romulan research facility hidden deep in an asteroid field. The facility holds advanced technology and data that could be valuable to the Federation, but it's guarded by automated defense systems and mysterious security protocols.

Temporal Anomaly: The ship encounters a temporal anomaly that sends them back in time to a period when the Romulan Star Empire was at its height. The crew must navigate the delicate politics of the time while finding a way to return to their own era without altering the course of history.

Cultural Exchange: The Endeavour encounters a newly formed Romulan settlement led by a charismatic leader who is open to establishing peaceful relations with the Federation. The crew is tasked with building trust and understanding between the two civilizations, but old prejudices and political tensions threaten to derail the process.

Mysterious Signal: The ship picks up a distress signal from a remote planet in Romulan space. Upon investigation, they discover a group of Romulan refugees struggling to survive. The refugees claim to have crucial information about a hidden threat in the region that could have devastating consequences for both Romulans and the Federation.

Rogue A.I. Threat: The crew encounters a highly advanced Romulan artificial intelligence left over from the Star Empire's heyday. This A.I. has its own agenda and views both the Federation and the Romulan warlords as threats. The crew must find a way to outsmart or deactivate the A.I. before it poses a danger to everyone in the quadrant.

Betrayal Within: A member of the Endeavour's crew is revealed to have ties to a Romulan faction and has been secretly communicating with them. The crew must navigate internal tensions and uncover the reasons behind the betrayal while dealing with the external threats in the Beta Quadrant.

Temporal Distortion Field: The ship encounters a region of space with intense temporal distortions that affect both technology and the passage of time. As they try to navigate through, the crew experiences strange phenomena, encounters alternate versions of themselves, and must find a way to escape the distortions before it's too late.

Not all of these were suitable; I have discovered that ChatGPT will suggest ‘time travel’ or ‘temporal disruption’ at any possible moment when developing Star Trek plotlines. You’ll notice none of these are astonishingly original, nor do they provide more than a premise - in the ‘Mysterious Signal’ story idea, it hints at a ‘hidden threat’ that could have ‘devastating consequences’ but doesn’t explain what that is. When asked, ChatGPT came up with a very melodramatic idea of a ‘dormant subspace anomaly’ linked to experimental technologies gone awry, which... could allow people to travel through time.

Because of the brevity of ChatGPT’s prompts, and the fact they are not always appropriate to the story you want to tell, you cannot simply ask it for an idea and then write it. It will give you a premise that you have to develop, and while you can press ChatGPT to elaborate on points - as I did with the ‘Mysterious Signal’ prompt above - you are still making key decisions as the author: what prompts you find interesting, what points you ask ChatGPT to flesh out, what you choose to use, what you choose to not use, and what you choose to develop for yourself.

The ‘Warlord Negotiations’ concept was close to an idea I’d already had when I gave the prompt, but I ultimately fleshed out a mission outline combining it with the ‘Lost Research Facility’ concept. In this plotline, Endeavour needed to cross the territory of a Romulan warlord, who offered safe passage in exchange for the Starfleet crew venturing to an abandoned military research facility and securing lost Tal Shiar R&D records on his behalf.

ChatGPT's plot suggestions: If in doubt, add time travel

There were more threads I developed both myself and with ChatGPT’s help: a subspace mine that damaged Endeavour so they had little choice but to accept the warlord’s help, power struggles within the warlord’s faction, the details of the research facility’s security (yet again it suggested technology to disrupt the passage of time). Sometimes ChatGPT came up with useful specifics (the technology of the subspace mine); a lot of the time, I used it as a glorified ‘rubber duck’ - I vented my ideas at it, and, rather than use ChatGPT’s responses themselves, those responses helped me know whether I liked my idea at all.

ChatGPT tends towards both the grandiose and generic. It routinely wanted to suggest stakes that threatened the Federation, and I had to routinely put in prompts to remind it that this was one lone Starfleet ship, far from home, so the wider political ramifications of the story were limited. It would often come up with mysterious or evocative-sounding ideas that were nevertheless very thin on detail, and if you pressed it for answers to the exciting questions it posed, the answers were usually very generic - or simply raised more questions. It was very useful in helping design a subspace mine that could incapacitate Endeavour to a specific degree, because I knew what condition the story needed the ship to be in. It was less use coming up with what technological secrets might be in the records of the research facility, with ideas which were almost universally far too powerful (or were, you guessed it, time travel).

I don’t feel like the plotline was something I couldn’t have come up with on my own. But using ChatGPT didn’t only kick-start the development process, it prompted me to write down my ideas and then read and make decisions about suggestions. Every once in a while, it came up with a really good concept - outside of technological details - that I enjoyed, but this mostly happened when I had already done a lot of the groundwork and it had a relatively original base from which to extrapolate. On its own, its ideas are quite limited and generic. It did, however, make an excellent tool for brainstorming, for throwing ideas against the wall, seeing what stuck, what had legs, and getting a little pushback.

Often, we want to do that with a person if we’re stuck on a story, where what usually happens is we vent out our idea, they haplessly try to make some suggestions, we say, ‘That’s nice, but no,’ and the process of explaining our idea reveals a solution to us anyway. With ChatGPT, you don’t have to worry about wasting anyone’s time!

I didn’t use the storyline I developed with ChatGPT in the end for reasons unrelated to the quality of the concept. I might use the premise in the future at some point; a ship doesn’t have to be weeks away from Starfleet support to try to enter negotiations with a local warlord, and I found I was developing an antagonist and a region I felt it would be a waste for Endeavour to leave behind once the story was over. But it made the development process a lot easier, helped me work through my ideas, pitched me some new ones, and let me know what my story wasn’t about as much as what it was about.

It was not, for all of ChatGPT’s trying, about time travel.

Conclusion

You must always be the author of the Stories you post on BFMS. That means that you can brainstorm with friends, ask them to proofread your work, and incorporate their suggestions, but you can’t publish something they wrote and claim it as your own. The use of generative AI is much the same. It is a powerful tool for research, especially for ‘research’ about science fiction concepts. It can also be useful for brainstorming and developing plot outlines. Using it in this way is not against policy; what matters is that the end product is your own writing, in your own voice, and a product of your own labour.

Happy writing!