Hacker Fiction Net

Generative AI in Writing, and the NaNoWriMo Conflict

You might not follow what writers are discussing online. But let me say that it's been a week of rage, accusations, and deleted accounts.

I'll begin with the recent conflict and then move on to deeper thinking on generative AI in writing and the arts.

The NaNoWriMo Conflict of September 2024

Longtime readers of my newsletter know that I love and participate in the National Novel Writing Month, abbreviated NaNoWriMo. It's a non-profit community focused on helping people write. The official goal is to write 50.000 words during the month of November.

NaNoWriMo is no small community and it has commercial sponsors.

The Q&A That Started It

A week ago, someone asked "What is NaNoWriMo's position on Artificial Intelligence (Al)?" on their official forum and got a response that blew the writer community up (archive link because they've changed the statement since):

NaNoWriMo does not explicitly support any specific approach to writing, nor does it explicitly condemn any approach, including the use of Al. (…)

We also want to be clear in our belief that the categorical condemnation of Artificial Intelligence has classist and ableist undertones, and that questions around the use of Al tie to questions around privilege.

In a section explaining "Classism," they said:

Not all writers have the financial ability to hire humans to help at certain phases of their writing. For some writers, the decision to use AI is a practical, not an ideological, one.

Similarly on "Ableism," they said:

Not all brains have same abilities and not all writers function at the same level of education or proficiency in the language in which they are writing. Some brains and ability levels require outside help or accommodations to achieve certain goals.

The Spicy Side Dish of a Sponsor

One of NaNoWriMo's current main sponsors is ProWritingAid – an AI-based writing tool. Here's from their FAQ:

(…) the AI used for most reports and features (excluding Rephrase, Sparks, Critique Report, and Inspiration Report) simply processes and analyzes your documents to provide suggestions on how to improve your writing. ProWritingAid only uses generative AI on our Sparks, Rephrase, Critique Report and Inspiration Report features, all of which can be turned off.

Animated gif showing ProWritingAid's feature called Sparks expanding a paragraph of fictional writing.

The Reactions Pour In

I picked up on this conflict when writers I follow started commenting on it and sharing what others were saying. Here are a few representative comments:

randomwolfeguy wrote on Mastodon:
As a person with Asperger, it pains me to see NaNiWriMo using ableism to appear inclusive. AI LLMs have ripped work wholesale and now regurgitate it. We never asked for this, and yet OUR work is being used without permission and even implicitly recommended by a "creative" organization.

FediThing wrote on Mastodon:
The claims by #NaNoWriMo about "poor people" needing to use LLMs are a bit like the old idea in English-speaking countries that people must learn to speak with a posh accent. This is what was parodied in Pygmalion (which became the musical My Fair Lady).

And NY Times bestselling author and lead on Star Wars: The High Republic Daniel José Older wrote on Xitter:
Hello @NaNoWriMo this is me DJO officially stepping down from your Writers Board and urging every writer I know to do the same. Never use my name in your promo again in fact never say my name at all and never email me again. Thanks!

Eventually both The Verge and Wired wrote articles about the backlash.

My Thoughts on the Controversy

The creative writing community has not made up its mind on generative AI. There are those who embrace it, those who are skeptical, and those who hate it or think it should be illegal. I'm personally in the skeptical category and we'll get to that in a bit. But what do I think of this NaNoWriMo debacle?

I think what NaNoWriMo was trying to say was …

• It's OK to use AI tools for creative writing.
• NaNoWriMo is primarily about humans who write.
• Dismissing or criticizing writing simply because it's generated by AI is dismissive of people who need AI to write.

Who Needs AI Tools to Write?

Taking a step back, I find it interesting to think of people who may need AI tools to write. I'm certainly not of the opinion that anyone should be blocked from creating stories. And it's believable that there are groups of people who have not been able to get stories they have in their minds into text before generative AI.

However, there is also something about the struggle to get good at writing and if that is no longer necessary, writing as an art has changed. So the question kind of becomes whether we want, or even accept, such a change? Then agagin, that question may be moot because we can't prevent or reliably detect the use of AI in writing.

The Problem With the Models

The bigger problem for me is how some large language models were created. I recently heard that old masterpieces by Jane Austen get flagged as AI-generated when run through tools that try to detect exactly that. Why? Because all old writing that's no longer under copyright has gone straight into training the AI models. So Jane Austen's writing looks like AI writing because she posthumously trained the AI. The same apparently goes for the Bible.

If we limit our analysis to AI used by humans, what we've effectively done is ingest millions of books into a statistical model that can produce new text by mimicking the writing skills of humans throughout history. That capability goes beyond what any human could ever do. It's the joint effort of a million writers, albeit a bit like patchwork.

My Take on Generative AI in Writing

I've written bits and pieces about generative AI in earlier issues of this newsletter.

In the May issue 2023 "The Turing Test and Our Fear of Humanoids" I mentioned artists get inspired by prior work and how reading good books is my favorite way of getting out of writer's block. AI models are not inspired, rather trained on prior work. But it's similar.

In my September 2023 Review of the movie Source Code I mention that AI models regress toward the mean and produce kind of averaged-out writing, not writing with a single human's voice. This means that a human like myself would need to train or refine a model with my writing to be able to use it while maintaining my voice. That's doable once you have enough polished writing of your own.

January this year, I wrote about the first time I listened to AI-composed classical music. I was not able to enjoy it. The knowledge that it was generated by a computer affected me. Up until that moment, I had not realized that knowing that a talented, hard working human had composed or performed a piece of recorded music contributed so much to my appreciation.

Authenticity and Trust

We're already in a place where there are whole books written by generative AI. Editors and publishers are dealing with what they call a tsunami of AI content.

Finnish writer Jukka Aalho has published AI-generated poetry in collaboration with ChatGPT called Aum Golly.

The cover of Aum Golly 2 with poetry generated by AI.

Today, we have no way of knowing if something was written by AI, a human, or both. Jukka Aalho decided to be open about it but that's not mandatory. I believe readers are interested in knowing though. It has to do with authenticity – what origin does this book have?

If I'm right, we will have to establish how to state origin. New terms of art will appear, such as:

• "Based on human writing" for a human draft refined by AI
• "Developed, written, and edited by humans" for pure human stories, and an equivalent term for pure AI stories
• "AI-generated based on a human idea" for prompt-based writing

It'll be hard to validate the claims of origin. That's where we might have to establish chains of trust. Having a close relationship with an author, like you reading my newsletter, can build direct trust between us. That trust could be transferred in testimonies where you could say "I vouch for John Wilander's books being written by him."

Ultimately, we will need laws that impose consequences on people who violate trust and lie about origins.

Why Care? If It's Good, It's Good

There are people who say "Why do you care who or what wrote the book? If it's good, it's good." The same goes for music, paintings, photos, and movies.

I think our experience of art includes an emphatic connection with the artist(s). This includes being impressed, a sense of kinship, and inspiration for what oneself might want to achieve. By extension, I'm saying that part of "being good" is being created by humans.

Are we only impressed by human creations? Absolutely not. Nature is an incredible source of inspiration and awe. Some view it as God's creation and thus a higher form than any human can aspire to. I recently decided that me and the family have to visit the Grand Teton National Park and see its grandeur.

Grand Teton National Park, created by nature, or God.

Authenticity and trust is not about policing what gets published or shaming someone. It's just about letting people choose and judge based on origin if they want. Telling people that they must not consider a book's origin is limiting. It's like saying we must not include author names on books.

Using AI Tools vs Letting AI Write For You

I felt like I needed to listen to a writer who's pro AI, someone who is embracing this development.

I found it in The Creative Penn by Joanna Penn, especially her podcast with the same name. She uses AI tools in her writing and thinks we all should consider it.

Joanna Penn.

Penn doesn't let tools write books for her, but she habitually asks them for inspiration and writing that's not her books.

A great example was when she had a plot with several characters set up with distinct backgrounds and professions. She then told a generative AI tool that setup and asked it for good interpersonal conflicts to incorporate in her story. Apparently the tool was really able to deliver. She wrote the conflicts herself.

She also promotes using AI for marketing ideas, writing blurbs for the backside of your books, writing subject lines for your newsletter issues, and writing updates on social media. Basically all the writing that's not the core of being an author.

Penn thinks it's better to embrace AI than to try to fight it and lose.

My Use of AI in Creative Writing

I pay for ChatGPT and use it for two things related to my writing.

My Primary Use: Answering Questions

I may be researching a topic and need to find keywords or a basis to dig deeper. ChatGPT is great at that.

Say I want a character to operate a tank in one of my stories. I asked ChatGPT "How do you operate a tank?" and I got this (significantly shortened):

Operating a military tank is a complex task that requires specialized training, coordination, and understanding of the vehicle's systems. Here's a high-level overview of the process:

1. Crew Roles and Responsibilities

   • Commander: Oversees the operation of the tank, gives orders, and coordinates with other units.
   • Gunner: Operates the main gun and secondary weapons, aiming and firing at targets.
   • Driver: Controls the movement of the tank, including steering, acceleration, and braking.
   • Loader (in some tanks): Manually loads the main gun with ammunition.

2. Starting the Tank

   • Ignition: The driver starts the tank using the ignition system, similar to starting a car but more robust. Some tanks use key ignitions, while others use switches.
   • Systems Check: All systems, including communication, weaponry, and engine, must be checked for functionality.

3. Driving the Tank

   • Steering: Most modern tanks use a steering system similar to a car, with a steering wheel or levers. Older models might use tillers or sticks.
   • Throttle and Brakes: The driver controls the speed using the throttle and stops the tank with the brakes. Some tanks have automatic transmissions, while others require manual shifting.
   • Visibility: The driver has limited visibility, often using periscopes or cameras. The commander usually assists with navigation.

My Secondary Use: Draft Covers

Here's the beta cover of my upcoming novel Submerged, generated by Dall-E:

Summary

Generative AI is only going to get better, but I think there are significant hurdles ahead such as the need for more data (where is it going to come from?), the need for quality data (who will vet it?), and the need for power both to model and infer (what are the trade offs?).

Humans will find ever new ways of using AI tools for the arts. Some authors will go all-in AI, some will covertly use AI to boost their writing, and some will reject AI all together either on moral grounds or because the joy of writing yourself is what makes them want to be writers.

New people will be able to write because of new AI tools. If that gives them and their readers joy, I am all for it. But questioning AI doesn't have to be abelist. There are questions about copyright, creativity, art, and authenticity that go far beyond enabling more people to write.

I will personally explore AI writing tools with caution. The challenge of becoming a great writer entices me and solving hard problems in writing gives me a sense of purpose. I hope my readers want that kind of authenticity.

This text was my most recent newsletter issue. I don't normally re-publish immediately here on the blog but wanted to partake in the ongoing NaNoWriMo conversation. My newsletter is focused on hackers in fiction and next issue I will go deep on the first episode of the TV series Mr. Robot! Subscribe below.