How AI Affects Human Creativity
Happy New Year, and welcome to the January issue of Hacker Chronicles!
Over the holiday break, I read a scientific paper titled Human Creativity in the Age of LLMs: Randomized Experiments on Divergent and Convergent Thinking. LLMs means large language models, and they are the basis of generative AI.
In this issue I share what I found most interesting in that paper and how it informs my strategy for the use of AI.
A heads up that I'll interview John F. X. Sundman about his hacker novel Acts of the Apostles in my next newsletter issue. I won't spoil the book's plot or resolution, but the interview will cover interesting aspects. It was a great read with lots of old 90s stuff I have not thought of in a long time.
/John
Writing Update
Publishing Submerged
Between Christmas and New Year, I got all the changes and comments from my copyeditor and have started responding.
You should expect out-of-band emails about the book release. Most likely one on pre-sales with cover reveal, one on sales start, and one when the audiobook is out. Those of you who have pre-ordered the Binary Release will get separate emails about that.
Writing the Third Novel in the Series
Before I jump into writing my third novel, currently titled Taiwan Yield, I need to fill the well. That is, I need to replenish energy and inspiration.
I'll do it in four ways:
- Study neural networks. I want to get a good understanding of the foundation of modern AI. This involves programming and takes a lot of time, but I'm already a bit into it and it's delightful.
- Read. I wasn't able to read much in 2024 because of my writing. It’s time to enjoy others' work for a while! I'll try for ten books before I start writing again.
- Decide how to use AI in writing. I need to make up my mind on generative AI and not just let things develop organically. This very newsletter issue will tell you how I'm approaching it.
January Feature: How AI Affects Creativity
My first novel, Identified, was written during ten years and completely without AI tools.
My second novel, Submerged, took me three years to write and is being edited by humans right now. While writing it, I used ChatGPT as a thesaurus and for researching things. And I used ProWritingAid to improve grammar and word choices, as explored in my November issue. I'd say Submerged has had very little creative input from AI tools.
The juncture I'm now at is how to write the third novel considering the raging development in AI? The questions I'm asking myself may apply to you too.
My Questions On Writing Going Forward
- Will my writing improve if I use AI tools more? Here I'm thinking of at least three dimensions — vocabulary, description/expression, and storytelling.
- Will my creativity improve over time if I use generative AI as a source of inspiration? Or will it deteriorate?
- Will my books regress to the mean if I use generative AI tools? This is both about the data the AI was trained on and the risk of using the same AI tool as other authors. If it goes badly, all fiction will be less diverse down the road.
- Will I be able to write much faster with AI tools and thus reach my goal of a new novel every eighteen months? I'm sure readers would love a higher cadence.
- Will I be more or less happy writing with AI tools? Writer's block can be frustrating, but coming up with something clever myself is very rewarding.
- Will I end up a laggard and curmudgeon if I don't embrace AI in writing?
After those questions have made a few rounds in my head, I sometimes end up asking myself the ultimate question:
Why am I writing? And will that 'why' be compromised if I create with AI?
Let's head into the research paper and then come back to how I'm answering the above questions.
Research Paper: Human Creativity in the Age of LLMs
Human Creativity in the Age of LLMs: Randomized Experiments on Divergent and Convergent Thinking (pdf) is still in preprint but has apparently gone semi-viral on TikTok. TechXplore wrote about it too.
It's authored by four researchers in Toronto and measures generative AI's effect on creativity, both when AI is used to directly solve problems and when it's used to guide or train humans in problem solving.
Divergent and Convergent Thinking
The authors use an established dual definition of creativity:
Divergent thinking. Generating new ideas, exploring multiple possibilities, and going in unconventional directions. The study asked people to come up with new, creative, unconventional uses of everyday items such as a pair of pants.
Convergent thinking. Refining and reducing a diverse set of ideas to the best one. Converging on a solution. The study used an existing setup for this which asks the person to find a common theme between three words, such as shelf, log, and worm having the common work book, for bookshelf, logbook, and bookworm.
Experiment Setup
The researchers divided participants into three groups: One with no AI help at all (the control group), one with direct help from AI, and one with AI guidance on how to approach the task.
All three groups then played a Game of Snake to reset their brains and then did the task once more with no AI help to see if prior use of AI affected their abilities.
The researchers measured the outcome in several ways: Originality of ideas, how many ideas, diversity within a set of ideas from one person, and diversity of ideas within the entire group of people.
As with all such studies, the results are limited to the setup and the test subjects. But there were some striking results that point toward where generative AI may take us.
Experimental Results on Divergent Thinking
These were the results on divergent thinking, i.e. being creative and coming up with new ideas:
- Control group
- Participants performed better in terms of originality when they had no prior exposure to AI.
- Participants who received direct help from AI
- Participants who received direct help from AI submitted fewer ideas than what the AI suggested to them. The hypothesis here is that individuals reject proposals from the AI and prioritize generating their own ideas, even though those may be worse.
- Participants who received direct help from AI did not produce fewer ideas later when they did not get AI help.
- Participants who received AI guidance
- Participants who received AI guidance on how to approach the task produced significantly fewer ideas compared to those without AI exposure. The researchers point out that this may be because reading and applying strategies require more time, potentially reducing the number of ideas generated.
- Participants who received AI guidance produced less diverse ideas when subsequently working on their own.
Experimental Results on Convergent Thinking
These were the results on convergent thinking, i.e. on refining ideas to the best one:
- Participants who received AI guidance on how to approach the task performed significantly worse than those who didn't get guidance or help.
- Neither direct AI help nor AI guidance enhanced the subsequent unaided performance. The authors mention how convergent thinking often relies on having an ’aha’ moment and AI guidance or help might have disrupted getting there.
- The participants in the two groups who got AI help reported a reduction in creativity after the experiment (self-assessment). The authors suggest that reliance on AI, particularly AI guidance, may undermine the confidence in our own creative abilities.
My Thoughts on the Results
If what we want out of a task is just the results, there's no need to have humans work to come up with them. But AI may be counterproductive if what we want is to train a human in solving a certain kind of problem or if we want personal satisfaction of having created or solved something.
Our instinct to reject AI suggestions is particularly striking to me. It also happens among humans. Who hasn't heard of the manager who only accepts a new idea when they believe it's their own?
But once this field matures, we will probably be able to design AI that helps humans in just the right way to actually train or boost creativity.
Current Answers To My Questions
So how do these experimental results affect my thinking on using today's AI in my creative writing?
I have already read a handful of books on how to write fiction, have listened to hundreds of podcast episodes on creative writing, and been inspired by all the books I've read. I'm no carte blanche. Others' work has trained and inspired me.
I have also gotten tons of direct feedback on my writing from developmental editors, alpha readers, and beta readers.
The important thing in comparing is where the lines are drawn.
My general training has been detached from my writing, both in time and in the fact that those podcasters have not created material on my writing.
The direct feedback on my writing has always come after I've created. I could have sought feedback on my plot before writing anything, but I haven't done that so far.
With that in mind, here are my questions and current answers.
Improved Writing
Will my writing, such as vocabulary, improve if I use AI tools more?
I think so. I'm open to the idea of asking an AI tool about various ways to describe an emotion or a scene. Turning what I have in my head into English words is something I accept training on.
Improved Creativity
Will my creativity improve over time if I use generative AI as a source of inspiration?
I think it may deteriorate unless I'm deliberate. I want my creativity to improve and expand. An AI prompt like "What would be a good plot twist after XYZ?" feels risky. Tempting too of course, because who knows what exceptional plot twist may be at my fingertips?
My approach will be to come up with my own best idea first, and flesh it out well, and only then consult AI. That way, I don't risk rejecting an idea I could have come up with myself just because I don't want the AI to create my story, and I can use AI to challenge me rather than replace me in creativity.
The best possible outcome of this approach will be if I push my own creativity harder, knowing I will ask the AI later and wanting to come out ahead. I should want to beat the AI and that could become a driver!
I will opt for my own ideas if I only get slightly better ideas from AI, and opt for synthesis if the AI comes up with something vastly better.
Regression to the Mean
Will my books regress to the mean if I use generative AI tools?
There's quite a lot of this going on already with how we are taught to write. Hero's journey for instance. Or the formulaic nature of the romance genre. Or archetypical characters such as the sidekick or the mentor.
For AI in specific, it comes down to what is now called prompt engineering. Can I ask generative AI unique enough questions to get reasonably unique results? I'm going to have to challenge myself there.
Improved Writing Speed
Will I be able to write much faster with AI tools and thus reach my goal of a new novel every eighteen months?
I think so, yes. But I don't want to become faster that way. This plays into the question of why I write.
Happier With or Without AI
Will I be more or less happy writing with AI tools?
There are pieces of writing that don't make me happy today, such as advertising text.
But there is a potential opening here. If I refine an AI model with my own corpus of writing, I could see myself happy if it clearly produced my kind of writing.
Becoming a Laggard
Will I end up a laggard and curmudgeon if I don't embrace AI in writing?
I think creative writing will to a large extent become AI-driven in the coming years. The incentives are there — money, fame, satisfying readers who want more, plus just getting your writing done and spending time on other things.
But quite a few authors will use AI sparingly or not at all. What such authors produce could become an authenticity niche. We just need to figure out how to validate the humanness of a book or a script.
I think I've pushed myself enough to not be a curmudgeon. Whether I become a laggard will hopefully be a decision.
Why I'm Writing
Finally the ultimate question — why am I writing? And will AI affect that?
Creativity and Speculation
Foremost, I love creating clever stories and speculating about the future. I got started on story craft as a role playing Game Master back in high school. I don't want AI to replace that joy.
A Way of Life
I've been doing this for thirteen years now. It's become part of my life, which means I'm constantly aware of situations, characters, and ideas that could make it into one of my books. I take notes on language things (idioms, words, cool phrases), conflicts in the news, larger-than-life characters, and hacks I come up with myself when I spot a vulnerability in a system.
Incorporating AI would not challenge this, just alter it.
Legacy
For many years I was shy about why I compose music and later why I write fiction. I didn't even want to dwell on it much myself because it felt, and feels, dark. But what it comes down to is death anxiety.
It's not that I have a constant fear of death. And I don't worry much about my health or being mortal. But I do worry about having no legacy. I don't want to die with nothing left behind worth keeping.
As I've grown older and become a dad, it's become easier to admit that I want to have a legacy. My kids will have direct memories of me and if I have grandkids, I'll hopefully make a lasting impression on them too. But that's it. Any subsequent generations will only have stories and what I've created. I want to carry on living in that sense.
My writing allows me to share thinking, creativity, and personality in a broad way. Sure, I write near future which will be dated by the time I die. But well-written books are enjoyable even though they are from a different time. Isaac Asimov's and Agatha Christie's sure are!
So, where does that take me with AI tools? Will it be my legacy if my plot points were generated by AI? Will anyone care in the future? Will they even know?
Had my goal been to become a prolific author and earn lots of money on my writing, I would likely be much more open to letting AI tools develop and create for me.
But when the day comes and my life is over, I want to think that what I leave behind is truly by me.
What Are Your Thoughts?
Have you thought about the same questions I have? What are your reactions to the research results? What do you think of my approach?
Currently Reading
Oh, the delight of browsing the bookcase and picking my next read. I revel in it and landed on Den skrattande polisen / The Laughing Policeman by Maj Sjöwall and Per Wahlöö. There's a 1973 American movie based on the novel.