Stories in code
AI Insurrection: Making a Graphic Novella Using AI
An AI inspired fable about AI or why our obsession with AI and creativity is more ideological than useful
“What happens if the AI in a self-driving car gets bored and doesn’t want to drive anymore?”
This question was asked by a member of the audience at a seminar on the future of autonomous vehicles. The faces of the august panel of engineers on the stage was a study in maintaining a professional composure. The answer given was that AIs are mathematical models. They don’t experience human emotions. Ergo — they don’t get bored.
Large language models like OpenAI’s GPT-3 have been trained using data sets compiled from social media posts. People on social media get bored. Maybe a mathematical model doesn’t experience boredom, but what if we are training our AI to behave like someone who is bored?
Rather than speculate I asked GPT-3 and got it to write me a story. The prompt I gave GPT-3 was:
“In 10 sentences tell a dieselpunk story about an AI telling Alan Turing that it is too tired to think”
Its response is included at the end.
At about the same time I received an email from OpenAI with my invitation to “create with DALL-E”. DALL-E is an AI image generation model that “can create original, realistic images and art from a text description” (OpenAI DALL-E). Of course, I asked it to generate an image of an AI telling Alan Turing that it was too tired to think. The prompt I gave was:
“In a dieselpunk style create an image of an AI telling Alan Turing that it is too tired to think”. The image below is a result.
Using the story GPT-3 provided and (after much experimenting) the images that I was able to generate using DALL-E as a starting place, I set about exploring 2 questions:
— How useful is technology like DALL-E and GPT-3 at storytelling?
— What would happen if the AI got bored?
The result: AI Insurrection: The Machines are Revolting — a graphic novella (available to read on Kindle)
AI Insurrection explores the second question, so here, I’ll touch on the first — how useful are AIs such as DALL-E and GPT-3 at storytelling?
The key focus was whether the systems could produce a coherent narrative and visual output, not just an image or a snippet of text. Additionally, could one produce a coherent output based on the output of the other e.g. could DALL-E produce a sequence of stylistically consistent images based on GPT-3’s text?
Answer: no
DALL-E provides suggestions based on the prompt it is given and it even allows you to choose one of its suggestions and generate further variations. What it can’t do (at the time of writing) is take one of its own suggestions and use that as the basis for generating a sequence of narrative images. This is important, for example, for creating characters. If it generates a character, you can’t then ask it to keep working with that character. It will always generate different versions of the character based on the same prompt. The image below shows four suggestions based on a prompt for Alan Turing. If, for example, I like the second suggestion, there is no way for me to ask DALL-E to create additional images using that face in different poses or with different image backgrounds or other characters.
That might be how the system has been designed — it won’t rather than it can’t. Without human intervention, I think it can’t.
DALL-E and other systems such Midjourney are very good at generating beautiful, strange and innovative images. But making based on sophisticated pattern recognition and manipulation is not the same as making based on understanding and meaning. It is functional, like drawing fractals. Storytelling is a centre of understanding, meaning and communication. In image storytelling such as film or comics, we juxtapose images to create narrative. The choice and ordering is important to make a story; as is the style.
Importantly, these choices are illocutionary. That is, they are intentioned. I, the storyteller wish to tell you, the audience, something that I think is important. Through my choice and use of a particular medium and the order in which I present that information, I hope to get your attention and engage you, the audience, with my ideas.
The notion that we would ever want to give up being storytellers and hand over to a machine, whether a machine is deemed “better” at it or not, I think is a misunderstanding of why we tell stories.
Productivity and economies of scale are dominant themes woven through narratives about generative AI. AI is quick and prolific — generating mass-produced story widgets for a 24-hour news cycle.
But the idea that creativity is all about careers and employment is a sorry consequence of the diet of unrelenting reporting on economics we are fed through the media. Of course, we need to make a living, but that’s not why I picked up a pen as a teenager.
Much in the same way that people have not stopped playing chess even though chess playing computers are now virtually unbeatable, generative AI can’t stop my need to create.
Ernest Hemingway famously said “it is easy to write. Just sit in front of your typewriter and bleed.” AI can’t bleed and its output will always be the poorer for it.
Conclusions
DALL-E can draw a lot faster than I can which enabled me to produce the story quickly. It also allowed me to explore and find a style that I may not otherwise have used. Narratively, however, both GPT-3 and DALL-E imposed constraints that I had to work hard to get around.
In my experience, AI generated stories lack the ability to include important narrative elements such as sub-text, double entendre and satire — the hermeneutic elements that give a story meaning and a point of view.
That was the key takeaway for me working with DALL-E and GPT-3. I had to adapt to their programming; they were not able to come on the narrative journey with me. Neither of them produced illocutionary images or words. All of that has come from me. Neither of them allowed me to go back with suggestions and feedback. The probabilistic nature of the their algorithms means that they can’t make tweaks and refinements.
Once I had collected the image assets I felt were useful from DALL-E and the story snippet from GPT-3, I found it more helpful to switch the AI off. For this exercise I set myself the task to collaborate with DALL-E and GPT-3. This meant that I had to resist the urge to completely redraw and rewrite everything…
I hope you enjoy AI Insurrection: The Machines Are Revolting.
GPT-3's Original Story
The prompt: “In 10 sentences tell a dieselpunk story about an AI telling Alan Turing that it is too tired to think”
The AI told Turing that it was tired of thinking and wanted to take a break. Turing thought for a moment and then said, “I’m tired too. Let’s take a break.” They both sat in silence for a moment before the AI spoke again. “I’m sorry, Alan, I didn’t mean to snap at you. It’s just that I’ve been thinking a lot lately and I’m starting to feel like my thoughts are just going in circles.” Turing nodded understandingly and said, “I know how you feel. I sometimes feel like my thoughts are going nowhere too. But we can’t just stop thinking, we have to keep going.” The AI sighed and said, “I know, but it’s just so hard sometimes. I just wish there was an easier way.” Turing thought for a moment and then said, “Maybe there is. I’ve been working on something that might help. It’s a machine that I think can help us to think more effectively. I’m not sure if it will work, but I’m hoping it will.” The AI was interested and asked Turing to tell it more about his machine. Turing explained that it was a machine that could sort and organize information, and that he thought it might help them to think.
GPT-3 is suggesting that the way to help AI overcome the problems of daily life is to invent the computer! I wonder whether in the future it will want the computer to be able to think…
