AI Product Ideation With Large-Language Models: an Experiment in Multi-Lingual Innovation in Japanese, Thai, and Burmese

Large Language Models (LLMs) such as ChatGPT have recently been used for ideation of new products. Given that LLMs are primarily trained on English, and a small number of other high-resource languages, this raises the question: is the ideation opportunity skewed or limited by this language constraint?

This research explored how the quality of generated product ideas varies when using an LLM in a high (Japanese), medium (Thai), or low (Burmese) resource language. The research utilised several research domains and was, therefore, necessarily exploratory. The experiment’s outcomes will inform further research in ideation, product development, and innovation within cross-cultural/multi-lingual working environments.

The basic task proposed in this project was to generate and evaluate ideas for a new product that would sell to university students in Kumamoto, Japan for JPY 5,000, an ideation test for LLMs which has been conducted previously in English with American students (Girotra, et. al. 2023). For this project the ideas were generated through ChatGPT, using a common set of prompts provided to native Japanese, Thai, and Burmese speakers. The Japanese students directly represented the target customers for the proposed products in this research and were a baseline for comparing the other two teams.

The presentation outlines an ongoing collaboration between two university professors, one based in Japan, the other in Thailand. The collaboration also involves native Japanese-speaking students from a private university in Kumamoto, Japan, and native Thai and Burmese-speaking students from a private university in Bangkok, Thailand.