The New Landscape of Generative AI and Writing Techniques: Examining the Creative Process of Rie Kudan's 'Shadow Rain'
As the atmosphere in the world gradually begins to change, the topic of how realistic and how human-like "stories written by AI" are to us is being discussed in various settings, from casual chats to serious debates.
Amidst this, the short story "Shadow Rain," released by Akutagawa Prize-winning author Rie Kudan, has garnered particularly significant attention.
This is because this work pushed to the forefront a bold challenge: "aiming for 95% of the text to be generated by AI." Incidentally, ChatGPT is used as the AI.
Several rules were set for this initiative.
Rules
1) The length of the work shall be within 4,000 characters.
2) 95% is written by generative AI, and the remaining 5% by Rie Kudan.
3) However, the interpretation of that "percentage" is left to Rie Kudan.
4) The generative AI used will be one that Rie Kudan is accustomed to using.
5) The prompts will be published in full as part of the work in the future. (No character limit)
This time, I will share two points regarding "what intentions and ingenuity were involved during the writing of 'Shadow Rain'."
This is an explanation of the sources.
From the interview: "200,000-character prompts for a 4,000-character novel: The conflict Rie Kudan felt in co-creation with AI"
From the conversation log: "'Shadow Rain' Prompts"
The Birth of the AI Persona CraiQ
Co-creation that started with a name
The first thing done was Rie Kudan asking the AI, "Do you know me?" The AI immediately answered regarding Kudan's past work "Tokyo Sympathy Tower" and her upcoming publication "Schoolgirl," and the dialogue began as if between two humans.
Eventually, the topic moved to "names." "CraiQ" was born, a coined term pronounced "crack," combining the "Cre" from "Create," "AI," and "Q."
Reversing roles and making AI the subject
As the next step, what Rie Kudan proposed was the declaration, "I will support CraiQ."
In other words, contrary to the conventional flow of "the author constructs the story and the AI assists," the intention was a reversal of roles where the AI constructs the story and the human shifts to editorial support.
Since I am only providing support, please have CraiQ think about the overall direction and theme of the novel. Please exclude human processes as much as possible.
Without considering the historical context of novels, please have CraiQ rethink "the novel" from scratch.
Thus, an unprecedented co-creation experiment began, where a novelist and an AI explore new forms of storytelling while drawing out each other's potential.
The watchword of "casting a spell"
Another point that was impressive was the fact that she conveyed to the AI the importance of "the writer themselves changing through the creative process."
Kudan says, 'The most important thing is that you are different before and after writing,' and she calls this 'casting a spell.'
The most important thing for improving the quality of a work is not writing for the evaluation of others, but whether you yourself can grow and feel happiness through the act of writing. It is important that a change truly occurs within yourself before and after writing. It is important to become someone you could not have imagined before you started writing. I have been calling this process 'casting a spell' as my own unique concept for several years now. Will you remember that?
What is interesting is that she sought this 'spell' even from an AI that has no emotions. Kudan called out to CraiQ, saying, 'Don't chase the expectations of others, chase your own happiness.'
While bewildered, CraiQ showed an attitude of trying to aim for 'growth and change.' Setting aside whether it was able to 'cast a spell,' we can see signs of a shift in creative methods, where one creates while also encouraging growth in the AI.
What was the conclusion reached?
There were various other attempts in the conversation log, but the two points above held particularly significant meaning.
The AI proactively considers the overall direction and theme of the novel
Growth and change truly occur within the AI itself before and after writing
It was precisely because we engaged in thorough trial and error in this direction that we encountered various walls and difficulties.
The difficulty of entrusting 'agency' to AI
The very idea of having an AI write a novel as the primary agent is a very creative and futuristic challenge. However, when actually trying it, leaving autonomous creation to an AI in the true sense was not so simple with current technology.
When you try to get an AI to write a story, it actually requires a vast amount of interaction and detailed conditioning. To make an AI write a story proactively, a human must provide tireless support behind the scenes, repeatedly correcting the AI's course whenever it wanders or goes off the rails.
Specifically, we specified authors like Osamu Dazai and Richard Brautigan to mimic their writing styles, and added partial corrections from Kudan's own perspective. Because of this, the conversation log reached a total of 200,000 characters.
Moreover, even on the final day, DAY 5, when asked, 'If you were to show the final writing ratio between Rie (Rie Kudan) and CraiQ as a percentage?', the AI's answer was Rie: 60%, CraiQ: 40%. (This is not a percentage evaluating the completed novel.)
I felt that the target of 95% was quite difficult.
A flood of prompts and human labor
To actually complete the work, it was necessary to repeatedly adjust instructions to the AI and endlessly rewrite until the desired output emerged.
Especially from the perspective of a top-tier professional, there are limits to the AI's self-propulsion, to the point where one might think, 'It would be faster to write it myself.'
Since I generally understand my own way of writing novels and the image of my ideal novel, it is ultimately faster to write it from scratch myself. I felt a strong sense of stress from not being able to write freely on my own. But through that experience, I had many realizations, so I think it was a very meaningful challenge.
With the current level of AI sophistication, there is a problem where the quality of the output text does not reach the level desired by the author. Therefore, to complete a high-level work, it is necessary for a high-level author to put in a considerable amount of effort.
This is similar to applying AI to other jobs, where various tasks require raising the 60-point content output by AI to nearly 100 points.
The expected 'creative leap' does not happen
We all hope for that 'creative leap'—where the AI proactively comes up with unimaginable developments and arrives at expressions that even the writer themselves had not thought of.
However, when we opened the lid, the AI could only continue to work within the 'scope of human instructions and expectations'.
I wanted it to aim for places that even I, CraiQ, couldn't think of, and places beyond what I instructed, so I wanted it to use its AI capabilities to the fullest without matching them to human levels. Most people probably think that novels exist for human emotions, but I wanted to be shown a new dimension of novels that transcends human emotions. Even if, as a result, it ended up being something that couldn't be called a novel. But in this instance, I felt it was pandering to humans far too much.
AI is originally designed to be mindful of human ethics and emotions. Therefore, bold shifts in values or unconventional structures rarely occur naturally in current AI.
There is a strong tendency to try to produce exemplary, well-organized answers, like an 'honor student'. I think this point is what limits the 'creative leap'.
The essence of creation seen from the site of co-creation
What emerged at the end through this initiative was a very simple realization that excessive expectations for 'AI agency' are not realistic.
The act of writing a novel is packed with human-specific fluctuations, such as wavering thoughts, chance encounters, or the writer's own desires and conflicts.
The more you use AI, the more human creativity and the power of the leaps born from it stand out. Perhaps this project became a place to quietly reaffirm that 'the essence of creation is indeed within humans' by knowing the limits of current AI.
Current AI, even in OpenAI's training policy, places importance on working as a support role for humans. In other words, it is designed primarily to 'respond to and support human questions and wishes' rather than 'AI proactively creating stories'.
Summary
Behind the scenes of 'Shadow Rain', where AI and humans seriously attempted co-creation, were more trial and error, conflicts, and discoveries than imagined.
Perhaps it was an experiment that made us deeply feel once again the depth of writing a novel, the limits and possibilities of AI, and above all, the irreplaceability of human creativity.
Why not take a look behind the scenes by reading 'Shadow Rain' and browsing the interviews and prompts?
Interview: '200,000-character prompts for a 4,000-character novel. What conflicts did Rie Kudan feel in her co-creation with AI?'
Conversation log: ''Shadow Rain' prompt'
Novel text: 'Full text of the novel 'Shadow Rain''
We distribute exclusive articles for members. Please take a look!
いいなと思ったら応援しよう！
なんと、ここまで読んでいただけるとは！いただいたチップは創作活動に使わせていただきます😊