Most people treat GPT technologies as automatic text generators, but the actual key to using AI effectively is to look at it like a storyteller.
Most storytellers see invention like going to the gym. You don't just go when you have a game coming up. You practice every day, examining ideas, looking for new associations, and exploring the best ones.
Making leaps of association is key element to this creative thinking and something GPT technologies can't do on their own. Leaps of association are ways of forming connections between seemingly disparate topics and forming them into something meaningful. AI can make connections, but they can't necessarily make large leaps or make meaning of those large leaps. Remember ... LLM aren't thinking, they predicting language patterns.
AI can help you make leaps of association and show you disparate topics, but it cannot make meaningful sense out of those new connections without the help of a human individual. Working with AI like storytellers helps students see the limitations of these technologies, especially when trying to enhance creativity in the writing process.
Seeing Invention as Interaction
Invention and coming up with ideas doesn't just happen; it requires interaction between humans and technology... regardless of whether you are using a blank notebook or ChatGPT. But this requires writers to practice using technology in different ways. This means using AI as more than just a text generator or fancy search engine.
Here are three common invention activities that require technological interaction and how they tie to creative uses of AI.
1. Micro Writing Every Day
To see good ideas, you need a well-honed lens. Storytellers look for ideas everywhere and write them down. The more you do this, the better you get.
Good AI writers generate text multiple times and in multiple ways ... and don't use 99% of what they see. They learn to pick and choose.
2. Automatic Writing
Storytellers free-write every day to discover new associations. The best ideas are usually new connections you've never made before. To do this, you've got to get them on paper.
Good AI writers look for connections across multiple generations ... that is how AI can help come up with new ideas. It doesn't generate them. You do.
3. Prompting
Storytellers often use random words as prompts. For example, last week's prompt for the Move Me Poetry Battle was luscious. That word will bring different associations to mind for different people.
Good AI writers don't cut and paste good AI text ... they use it as a prompt for their own ideas.
The ideas are not from GPT-3 ... they emerge from the interaction between the writer and the AI.
Neither exists without the other. To truly come up with new ideas and form new connections, one must draw upon experiences, knowledge, and the collective wisdom of others. That requires both humans and machines.
Creating an Invention Experience Inside an LLM
There is a rich history of techniques used by writers and thinkers to come up with new ideas, and these techniques do not involve simply waiting for inspiration to strike or asking the machine to provide the answer. But most writing classes don't spend enough time in the invention phase of writing or show how coming up with ideas integrates into the entire process (not just at the beginning).
In rhetoric, these techniques are called heuristics ... or tools for coming up with new ideas. They are great for creating better prompts for GPT-3, but I want students to experience what it means to come up with an idea.
So what exactly do these heuristics do for the human mind? They help us build connections and new associations through a structured process that encourages analogy, juxtaposition, and comparison. Through this process, we can come up with new ideas not just for storytelling, but for all kinds of writing.
We can use the same techniques to create an invention experience inside a Large Language Model (LLM).
To help our students experience this process, I demonstrated various invention techniques such as prompting, juxtaposition, and association. Then, I simulated a language model for their next micro story.
Here's how I did it:
- I asked the students to free write on a random prompt for at least 5 minutes, without stopping.
- After 2 minutes, I asked the students to review their writing and annotate any interesting ideas or connections.
- Each student then wrote 1-2 sentences that really struck them.
- I shuffled the cards and handed them out randomly, which became the students' writing prompt for the week.
By following this process, we created our own data around a topic, followed various associative connections, and looked across generations for new connections to feed back into the model. Obviously, this is technically not a LLM, but it simulates creative interaction with an LLM.
The goal of this model isn't the generation of coherent text, but rather the generation of productive randomness.
We can use AI in interesting and creative ways to generate ideas and make leaps of associations. However, without the human element, AI can never truly come up with creative ideas. The best way to capitalize on AI's potential for creative problem solving is to team humans and machines together.
That should be important for any kind of writing class, because AI can write our less creative texts ... but it can't solve problems with writing or create new knowledge without us.
___
If you would like to copy or download the assignment I used for this activity, you can access it here. I'm currently gathering a group of teachers who are interested in exploring more creative uses of AI. If you are interested in beta testing the space, check out this page.
If you would like to copy or download the assignment I used for this activity, you can access it here. I'm currently gathering a group of teachers who are interested in exploring more creative uses of AI. If you are interested in beta testing the space, check out this page.