In this post, to demonstrate how creativity heuristics and generative AI can work together, we will describe a three-step approach to using creativity heuristics along with Large Language Model (LLM) tools such as ChatGPT to generate initial concepts during the idea generation process. The ideas served up by AI tools are mostly generic ideas and can provide a stepping stone or baseline for further ideation by creative teams.
The approach we describe in this post includes four steps: Firstly, we "teach" the LLM model to ideate using multiple creativity heuristics to generate raw ideas for multiple product categories. Secondly, we refine the specificity of ideas through the use of LLM (Language Model). Third, we generate prompts for text-to-image models using the specified ideas from the second step. Finally, we employ text-to-image models to help visualize the concept.
Step 1: Generating Raw Ideas using Creativity Heuristics
In the first step, we will guide ChatGPT into using five creativity heuristics for the idea-generation process. For simplicity, we provide a fairly high-level description of the five heuristics which are among the 11 thinking tools of Innomantra's Function Innovation Methodology. Our prompts are in blue and italicized, and the response of ChatGPT is in green.
You will play the role of an ideator. For every product that I tell you, you will give me an idea that involves (1) eliminating a component of the product (2) combining two components of the product (3) segmenting, dividing, or modularizing the product or its components (4) copying, replicating or adding redundancy to the product or its components (5) varying or modifying a feature of the product or its component dynamically based on some external input.
Sure, I'd be happy to help with that! Please go ahead and provide a product for me to generate ideas for.
- Eliminating a component: A toothbrush without bristles, relying instead on ultrasonic vibrations to clean teeth.
- Combining two components: A toothbrush with a built-in toothpaste dispenser, so that the user only needs one device for brushing their teeth.
- Segmenting, dividing, or modularizing: A modular toothbrush with interchangeable brush heads, allowing users to easily switch between different types of bristles and shapes.
- Copying, replicating or adding redundancy: A double-headed toothbrush, with bristles on both ends, so that the user can clean all their teeth more quickly.
- Varying or modifying a feature dynamically: A toothbrush that adjusts the vibration frequency and intensity based on the user's brushing habits, providing a more customized cleaning experience.
- Eliminating a component: An electric car without a traditional dashboard, relying instead on a heads-up display projected on the windshield.
- Combining two components: An electric car with integrated solar panels that can both power the car and recharge the battery while driving.
- Segmenting, dividing, or modularizing: A modular electric car where different components such as the battery, motor, and suspension can be easily swapped out and upgraded.
- Copying, replicating or adding redundancy: An electric car with dual motors for increased power and reliability.
- Varying or modifying a feature dynamically: An electric car with a dynamic battery management system that adjusts energy usage based on driving conditions and the driver's behavior, optimizing for efficiency and range.
- Eliminating a component: An MRI scanner without the need for a bulky, enclosed tube, using instead a "open MRI" design that eliminates the claustrophobic feeling for patients.
- Combining two components: An MRI scanner integrated with a real-time patient monitoring system, allowing medical professionals to monitor vital signs during the scan.
- Segmenting, dividing, or modularizing: A modular MRI scanner where different components such as the magnets and gradients can be easily replaced or upgraded.
- Copying, replicating or adding redundancy: A dual MRI scanner system for increased precision and accuracy, or for the ability to perform scans on two patients simultaneously.
- Varying or modifying a feature dynamically: An MRI scanner with a real-time adaptive imaging technique that adjusts the imaging parameters based on the patient's anatomy and tissue properties, improving image quality and reducing scan time.
Step 2: Increasing the Specificity of the Idea
In Step 2, we ask the LLM to increase the specificity of one of the ideas. Here our objective is to obtain additional details about one of the raw ideas so that we can better understand it.
Can you expand on this idea with rich details: "A toothbrush without bristles, relying instead on ultrasonic vibrations to clean teeth."
Certainly! The idea is for a toothbrush that uses ultrasonic vibrations to clean teeth, instead of traditional bristles. Here's how it could work:
The toothbrush would have a handle and a head, similar in size and shape to a traditional toothbrush. The head would contain a small ultrasonic transducer, which would generate high-frequency, low-amplitude vibrations. When activated, these vibrations would cause tiny bubbles to form and collapse in the user's mouth, creating a scrubbing effect that removes plaque and food particles from teeth and gums.
The toothbrush would be powered by a rechargeable battery and could be activated by pressing a button on the handle. The user would simply place the head of the toothbrush in their mouth, and press the button to turn on the ultrasonic vibrations. They would then brush their teeth as they normally would, using ultrasonic vibrations to clean and massage their gums.
One advantage of this design is that it eliminates the need for bristles, which can wear down over time and harbor bacteria. The ultrasonic vibrations would provide a more effective and hygienic cleaning experience, without the need for replacement brush heads. Additionally, the compact and lightweight design of the toothbrush would make it easy to use and travel with.
Of course, this is just one possible implementation of the idea. There may be other variations or refinements that would make the toothbrush even more effective and user-friendly.
Step 3: Prompt Generation
In this step, we will ask the LLM to convert the above rich description of the product into a visual description that can be used as input for a text-to-image model.
Please create a prompt to use with DALL-E for the above idea that gives a detailed description of the visual characteristics of the product.
Here's a prompt for DALL-E to generate an image of the ultrasonic toothbrush:
"Visualize an ultrasonic toothbrush for cleaning teeth and massaging gums. It has a sleek, modern design with a handle that is ergonomically shaped for a comfortable grip. The head of the toothbrush contains a small ultrasonic transducer that generates high-frequency, low-amplitude vibrations. The toothbrush is powered by a rechargeable battery, and a button on the handle allows the user to activate the ultrasonic cleaning. The toothbrush is white and silver in color, with a clean and minimalistic look that would appeal to those seeking a high-tech, hygienic alternative to traditional toothbrushes.
Step 4: Concept Visualization using Text-to-Image Models
Finally, we use the prompts we generated in the previous step, and use them as inputs for text-to-image models. The first set of images was generated using DALL-E, OpenAI's text-to-image model. Unfortunately, the images generated showed little novelty or variation compared to common electric toothbrushes.
Next, we tried a different text-to-image model. The following images were generated using Midjourney, a generative AI model recognized for its highly realistic and evocative image generation capabilities. Here the images were far more realistic and polished compared to those generated with DALL-E. However, again, the concepts looked very similar to regular electric toothbrushes.
Finally, we used Stable Diffusion to generate design concepts based on the prompts that were generated. A large fraction of the concepts were 'junk' ideas. However, some of the design concepts generated were highly novel and interesting. These concepts are shown in the below collage.
We felt that these design concepts could be used as triggers to refine the AI-generated idea. The top ideas can then be moved to later phases of the product development process.
Role of Human Creativity in AI-Enabled Idea Generation
It's important to note that the use of generative AI does not replace the role of human creativity. Rather, these tools serve as supportive aids in establishing baselines, improving detail, and facilitating visualization. Blindly relying on AI tools alone is unlikely to result in innovative (i.e., both novel and valuable) ideas.
In the above example, our goal is to demonstrate what can be achieved solely through the use of generative AI tools, without human intervention. However, it is important to remember that the most impactful and innovative ideas will often result from a combination of human creativity and AI-assisted tools.
As part of Innomantra's Functional Innovation Methodology, we leverage creativity heuristics, generative AI tools, and the power of IdeaScale's Idea Management platform to enhance our clients' idea-generation capabilities. To learn how we have supported leading companies such as ABB, Bosch, Google, Samsung, and NetApp to turbocharge their idea management process, schedule a meeting today at calendly.com/innomantra.