This paper reinterprets Ramón Yul's "Ars Combinatoria" as a conceptual foundation for building a modern research idea generation machine. By defining three constituent axes—topic (e.g., efficiency, adaptability), domain (e.g., question answering, machine translation), and method (e.g., adversarial training, linear attention)—we represent the motivations, problem formulations, and technical approaches commonly encountered in scientific work at a high level of abstraction. By extracting elements from expert or academic papers and curating their combinations, we prompt a large-scale language model (LLM) to generate diverse, relevant, and up-to-date research ideas. This modern thinking machine provides a lightweight, interpretable tool for enhancing scientific creativity and points the way toward collaborative idea generation between humans and AI.
