Enhancing Creativity in Large Language Models through Associative Thinking Strategies

TL;DR

The paper investigates whether prompting a large language model (vGPT-4) to form associations between unrelated concepts can enhance creativity. It introduces two associative-thinking prompting strategies and evaluates them across Product Design, Storytelling, and Marketing using human judgments of originality and usefulness, with one-hop prompts tested in Product Design. Findings show that random association substantially increases originality across domains (notably ~40% in Product Design and ~28% in Storytelling), though usefulness gains are domain-dependent and sometimes diminished due to incongruity; the one-hop strategy often resembles the control. The work highlights both the potential and limitations of associative thinking as a practical tool for AI-assisted ideation, and points to context-switching and diversity of associations as important directions for future research.

Abstract

This paper explores the enhancement of creativity in Large Language Models (LLMs) like vGPT-4 through associative thinking, a cognitive process where creative ideas emerge from linking seemingly unrelated concepts. Associative thinking strategies have been found to effectively help humans boost creativity. However, whether the same strategies can help LLMs become more creative remains under-explored. In this work, we investigate whether prompting LLMs to connect disparate concepts can augment their creative outputs. Focusing on three domains -- Product Design, Storytelling, and Marketing -- we introduce creativity tasks designed to assess vGPT-4's ability to generate original and useful content. By challenging the models to form novel associations, we evaluate the potential of associative thinking to enhance the creative capabilities of LLMs. Our findings show that leveraging associative thinking techniques can significantly improve the originality of vGPT-4's responses.

Figures (8)

• Figure 1: The prompt used for applying associative thinking strategy in the Product Design Domain. The prompt begins and ends with the base prompt and query respectively, while we vary the associative thinking strategy based on the condition.
• Figure 2: Originality scores for Product, Storytelling and Marketing Domains. Overall, the Random Association Associative thinking strategy boosts originality across all domains.
• Figure 3: Responses from vGPT-4 for the Marketing Domain. For many shop types, we found the results to be quite persuasive and useful, like the left one. However, sometimes the LLM would output images where the random object did not integrate in a useful way, like the example on the right.
• Figure 4: Prompts used for the Storytelling and Marketing domains. In the control condition, no associative thinking strategy was introduced. In the Story Telling Domain (left), the story type and plot twist were provided and in the Marketing Domain (right) the shop type was provided. In both cases, we fix a random object and manipulate the associative thinking strategy.
• Figure 5: Instances where vGPT-4 generated novel and useful ideas in the Product Domain. These product ideas were scored only for the textual outputs, but we include the images here for reference. We observe that some images make the idea look unworkable, for example, the screw-like groove may prohibit sharpening the pencil. However, a simple tweak like treating the screw-like groove as a pencil sleeve could easily solve the issue.