Creativity Collider: How to Generate Novel Ideas with ChatGPT

#chatgpt #chatbot #creativity #creativeindustries #highereducation

Ready to release the full creative potential of Large Language Models (LLMs) like ChatGPT? While AI continues to ace common measures of intelligence in diverse disciplines, many have grappled with generating truly novel ideas by using the technology. However, the key to prompting ChatGPT’s hidden abilities lies in understanding metacreativity—the ability to think critically about creativity and apply those insights to generate wonderfully new ideas.

In this article, I’ll share a fun and effective strategy that will enable you to overcome AI’s perceived limitations, turning ChatGPT into a creativity collider capable of generating novel, original ideas by smashing seemingly unrelated concepts. So, put on your lab coat and safety goggles, and get ready to experiment with unexpected collisions that might just change the way you think about AI and creativity.

“Who’s the more foolish, the fool or the fool who follows him?” – Obi-Wan Kenobi

ChatGPT is most impressive when it comes to intelligence and problem solving, however many are lamenting the technology’s shortcomings when it comes to creativity.

• The Atlantic proclaimed “ChatGPT is Dumber Than You Think.” The article, authored by Ian Bogost, dismissed the AI’s ability to write compelling poesy, even fairly simple imagist poetry.
• ?Forbes dismissed the chatbot’s ability to write creative marketing copy. The article forecasted that only low-talent creatives, who already produced contrived material, would face significant disruption.
• Even though a reviewer in the Washington Post was impressed by the chatbot’s intelligence, Geoffrey A. Fowler concluded it couldn’t produce any creative writing “that didn’t make me cringe.”
• Wired tested ChatGPT, Bing Chat, and Bard by asking the chatbots to author Saturday Night Live sketches lampooning Donald Trump. The results were disappointing and unfunny.
• An unpublished article, “On the Creativity of Large Language Models,” concludes that LLMs cannot be considered creative when it comes to creative writing, and are only capable of producing banal prose. Though, they do have the potential to do so in the future.

According to these reports, ChatGPT is a great collaborator for those who enjoy uninspired, cringe-worthy creative outputs.

I would suggest, as politely as possible, that AI isn’t the problem. Indeed, a study by Jennifer Haasse and Paul H. P. Hanel tested 6 chatbots, and found the most advanced platforms generated ideas that were just as or more creative as 90.6% of the human participants.

The chatbots could effectively ideate solutions to everyday problems (what we call mini-c creativity) as tested by the Alternative Uses Task. It is a common creativity test that asks participants to name all the alternative uses for an everyday object such as a paperclip or brick—check out this really cool BBC documentary to see it and other creativity measures in action. The authors conclude: “whether GAI [generative artificial intelligence] is creative can be answered pragmatically with ‘yes, as much or as little as humans.’”

In the case of chatbots, the answer to Obi-Wan's aphorism lies in the former: the human fool leading the AI is clearly more foolish.

The Secret to Unleashing the Creative Potential of AI (and Humans!)

ChatGPT is trained to generate the most appropriate response to a prompt, not the most creative. When generating text, ChatGPT uses a process called autoregression, where it predicts one word at a time based on its training, the prompt, and previously generated words. That’s why it often functions like an Internet search engine that reproduces what’s already been written instead of actually crafting new and novel ideas. If you ask it to generate creative or original ideas, it’ll simply output existing options that are tagged or labelled as such. That’s why all the above reviewers dismissed ChatGPT’s creative ability.

To get truly creative results, one needs to teach ChatGPT how to make associations between distantly related concepts. In fact, our brains use a similar process to generate novel ideas.

• A study by C. S. Q. Siew and colleagues on metaphoric creativity in humour found that novelty emerged from the conceptual distance between compared elements. The greater the distance, the greater the novelty. Entitled “Nymph piss and gravy orgies: Local and global contrast effects in relation humor,” the article is a must read.
• Another study by Dan R. Johnson and colleagues found that narrative creativity (i.e. storytelling) is characterized by divergent ideas and the distance between divergent associations.
• A third study by Yoed N. Kenett and colleagues found that the ability to interpret novel metaphors is a reliable predictor of creative ability. Similarly, several creativity tests measure an individual’s ability to make associations such as the Remote Associates Test (which is used to study insight) and Divergent Association Task (which is used to predict creative ability).

Consequently, effective creative processes activate metaphoric thinking and often incorporate randomness to encourage distant associations. If you want ChatGPT to produce creative ideas, you need to teach it how to collide distant concepts.

Use ChatGPT as a Creativity Collider

In two previous LinkedIn articles, I teased how I used ChatGPT to generate novel ideas for pizza dishes and supervillains who crush and kill creativity. For both, I applied relatively simple processes that circumvented ChatGPT’s tendency to generate banal outputs by instructing the AI to make associations between the challenge and completely unrelated concepts.

After introducing and clarifying the product innovation challenge, here's how I quickly prompted ChatGPT to generate novel and original ideas for pizza.

1. I selected an inspirational source that I thought might be a good fit. I chose the Baroque aesthetic because I wanted ideas that would be visually appealing, intricate, expansive, decadent, and highly unorthodox. Similarly, melted cheese often turns a golden hue when cooked at a high temperature, so I thought there might be some interesting opportunities to reproduce Baroque colour palettes and textures.
2. I asked ChatGPT to identify characteristics associated with the Baroque. It accurately identified key characteristics including ornamentation, gilt, emotion & intense contrast, illusionism, and awe-inspiring grandeur.
3. I then asked ChatGPT to apply those characteristics to generate creative ideas for pizza dishes. You can find the visually stunning and delicious list of ideas in my article “Welcome to the Metacognitive Economy!”

I used a similar process to generate ideas for supervillains by introducing works of critical theory as inspirational sources. It worked pretty well.

I also applied a slightly different approach to generate ideas for supervillains with creativity-crushing superpowers.

1. I introduced the design challenge: I told the AI my intention was to design supervillains who frustrate and kill creativity. I clarified key insights from Teresa Amabile’s “How to Kill Creativity” will be applied to design the supervillains. But, asked ChatGPT to acknowledge receipt and await further instruction.
2. I clarified the first inspirational source by asking ChatGPT to identify and summarize all the creativity killers introduced in the article. I analyzed the output and offered a few important corrections.
3. I generated a list of distant associations. I selected one creativity killer, and asked ChatGPT to make as many associations with the concept as possible. I clarified that each new association should increase the conceptual distance. I continued this process until the associations were a little unusual but still understandable.
4. I collided concepts. I selected a 3 to 5 of distant associations, and asked ChatGPT to apply those concepts to ideate supervillains who crush and kill creativity.
5. I iterated and expanded the conceptual distance. After each batch of ideas, I provided feedback telling ChatGPT which ideas were creative and which were banal. I asked ChatGPT to generate more ideas like the creative ones, and asked it to increase the conceptual distance with each new idea. I continued this cycle until there were a number of creative and effective options.
6. I then repeated the process with each creativity killer.

I've used similar prompt strategies to generate and develop novel ideas for Saturday Night Live sketches, new assignments, and more. Approaching ChatGPT as a creativity collider seem to be a reliable strategy for prompting novel and original ideas.

Metacreativity: A Competitive Advantage

You can be certain that something akin to ‘AI Thinking Templates’ will shortly flood the marketplace. By this, I mean people will use the ChatGPT API to make apps that will automate creative processes such as brainstorming, forced connections, empathy mapping, SWOT, etc. The user will describe the challenge in the prompt, select the thinking tool, and then AI will generate outputs correspondingly.

These apps will facilitate illusionary creativity: they’ll generate interesting ideas, but they also generate similar ideas for anyone else using the technology, thereby mitigating any meaningful value or strategic advantage. Creativity is fundamentally social; an idea is 'creative' only when it is novel compared to all previously generated ideas.

Originality will increasingly emerge from unique and authentic creative processes that facilitate AI ideation. So, dear readers, start thinking about the unique ways you generate ideas and transform those insights into broadly applicable creative processes. Embrace your idiosyncrasy, pursue diverse experiences, and bring what makes you unique to the creative process to make unusual associations.

Practice Using ChatGPT as a Creativity Collider

Here’s a quick exercise to practice generating novel ideas with ChatGPT.

1. Introduce the Challenge. Tell ChatGPT that you’d like some help generating novel and original ideas for a title to this very LinkedIn article that incorporates humorous metaphors. Introduce your target audience. Tell the chatbot that you will provide a copy of the article. Provide a copy, and ask it to acknowledge receipt and wait for further direction. It may ignore this direction and start generating banal ideas. That’s fine.
2. Teach ChatGPT How to Collide Concepts. ?Share the key insights from “Nymph piss and gravy orgies: Local and global contrast effects in relation humor.” Explain how humorous metaphors are created by comparing two very unlike concepts. Emphasize that the greater the conceptual distance, the more novel or original the ideas will be. Provide some examples, such as “nymph piss” and “gravy orgies.” Ask the chatbot to apply a similar strategy to generate 30 highly original ideas for titles that feature humorous metaphors. Depending on how ChatGPT is behaving that day, you may need to re-paste the LinkedIn article.
3. Provide Feedback & Iterate. Tell the chatbot what ideas you like and why. Ask for more ideas along those lines and keep advising it to increase the conceptual distance for each new metaphor. Once exhausted, ask the chatbot to focus on different insights or aspects of the article (such as pizza, metacreativity, unusual connections, etc.).
4. Store Ideas. You’ll generate way more novel ideas than you can possibly use. So, save those novel options for future projects. For example, I placed “Cognitive Carousel: Spinning Unlikely Ideas into New Creations with ChatGPT” and “Neural Network Novelties: Unearthing ChatGPT’s Unexpected Connections” into my ideas archive. I’m clearly sucker for consonance.

Give the exercise a try. In a couple of minutes, you’ll have dozens of potential options, many of which will be novel, catchy, and effective. If you especially dig any of the generated?ideas, please post them into the comments. And if you had fun, share the post with your network and ask them to try colliding ideas with their chatbot of choice.

Once you understand the process, apply it to a new creative challenge. Try out different inspirational sources drawn from your personal experiences and everyday life. In upcoming articles, I’ll share a few more strategies for generating creative, novel, and original ideas with Chat GPT; so, consider clicking that follow button.