TLDR: A new computational model and tool, INSIGHTS, has been developed to generate more novel and useful innovation opportunities. Unlike general generative AI, INSIGHTS integrates creativity theories and techniques, using five functions to systematically explore opportunity spaces. Evaluations showed INSIGHTS outperformed Notebook LM and ChatGPT4o in generating more creative outcomes, particularly in usefulness, by leveraging structured knowledge and specific functions like ‘atypical-opportunities’ and ‘pivot-opportunities’. The model offers a human-centered approach to innovation, providing greater control over the creative process.
In the rapidly evolving landscape of artificial intelligence, generative AI technologies like ChatGPT and Notebook LM are increasingly impacting creative professional work. However, their reliance on pre-existing data and probabilistic language modeling often limits them to generating more incremental creative outcomes. The challenge lies in their capability to explore truly new spaces of ideas that are not only useful but also genuinely novel.
A new research paper introduces a computational model and an associated tool, named INSIGHTS, designed to address this limitation. The model aims to generate more novel opportunities in professional innovation processes, informed by established creativity theories and techniques. It was developed by Neil Maiden, Konstan/nos Zachos, James Lockerbie, Kostas Petrianakis, and Amanda Brown.
The INSIGHTS Model: A New Approach to Creativity
The core of INSIGHTS is a descriptive model that integrates characteristics of creative outcomes from various creativity theories. This model defines how to discover more novel outcomes using a set of five key functions:
- Generative-rules: Systematically explores opportunity spaces using codified rules to discover creative opportunities.
- Pivot-opportunities: Iteratively increases the novelty of opportunities by using the outcome of one generative rule as the input for the next, exploring new areas of an opportunity space.
- Classes-of-needs: Defines opportunity spaces based on unmet needs for recognized classes, guiding the discovery of relevant opportunities.
- Atypical-opportunities: Identifies more creative opportunities by seeking those with higher semantic distances from the most common examples within a space, thus increasing novelty.
- Creative-qualities: Incorporates predefined qualities associated with creative outcomes (e.g., ‘more informative’ or ‘playful’ for digital products) to direct the discovery of opportunities with these attributes.
The computational implementation of INSIGHTS combines different AI technologies, including generative AI, to automate a four-step procedure:
- Process information assets: Extracts and cleans text from various file types (PDF, MS Word, PowerPoint, HTML), translating non-English content.
- Discover spaces of opportunities: Applies topic modeling (using BERTopic) to identify clusters of similar words, defining distinct opportunity spaces.
- Describe each opportunity space: Uses GPT4o to generate natural language descriptions and short labels for each discovered space, aiding consultants in selection.
- Discover multiple opportunities in spaces: Applies bespoke generative rules, leveraging GPT4o prompts, to generate candidate opportunities. These rules incorporate the generative-rules, pivot-opportunities, atypical-opportunities, and creative-qualities functions, allowing for control over novelty settings and custom text inputs.
Evaluation and Key Findings
The INSIGHTS model was rigorously evaluated by generating over 2000 innovation opportunities for the UK hospitality sector and comparing them against opportunities generated by Google’s Notebook LM and OpenAI’s ChatGPT4o. The opportunities were rated for novelty and usefulness on a 1-7 scale by GPT4o using customized prompts.
The evaluation revealed several significant findings:
- INSIGHTS vs. Notebook LM: INSIGHTS generated business and technical design opportunities that were both significantly more novel and more useful than those from Notebook LM. For policy opportunities, INSIGHTS generated more novel but not significantly more useful outcomes.
- INSIGHTS vs. ChatGPT4o: Opportunities generated by INSIGHTS were significantly more useful than those from ChatGPT4o, even though both tools produced opportunities with comparable novelty ratings. This suggests that while ChatGPT4o can generate novel ideas, INSIGHTS excels in ensuring their practical applicability.
- Impact of INSIGHTS Functions: The ‘atypical-opportunities’ function was crucial for generating more novel and equally useful opportunities. The ‘pivot-opportunities’ function also successfully increased the novelty of subsequent opportunities without compromising usefulness. However, the ‘creative-qualities’ function did not consistently lead to more novel opportunities across all types, though it did for policy opportunities.
This research highlights that by integrating structured knowledge about creativity and offering explicit control over opportunity spaces, computational models like INSIGHTS can overcome the common novelty-usefulness trade-off often observed in ideas generated by general-purpose Large Language Models. The model provides a human-centered AI approach, giving consultants greater control over the creative thinking process.
Also Read:
- Navigating the Research Landscape: An Automated Framework for AI and Robotics
- Simplifying LLM Interaction: A New Approach to Adaptive Prompt Generation
Future Directions
Future work includes further refining the ‘atypical-opportunities’ function to explore the boundaries of opportunity spaces, investigating why subsequent pivots didn’t further increase novelty, and exploring different manipulations of topic terms. The INSIGHTS tool will also undergo more extensive evaluations by consultants in real-world innovation projects, such as exploring policy innovations related to violence in society.
For more detailed information, you can read the full research paper here.
