Towards an Artificial Scientific Muse: Suggesting High-Impact Research Collaborations in Physics with Semantic Networks and Machine Learning

The exponential growth in scientific publications poses a severe challenge for human researchers. It forces the attention to more narrow sub-fields, which makes it challenging to discover new impactful research ideas and collaborations outside one's own field. To address these challenges, we introduce SciMuse, our first step towards an artificial scientific muse. SciMuse can suggest high-impact cross-disciplinary ideas and collaborations for scientists. It uses a dynamic semantic network built from millions of scientific papers from the past decades, incorporating both semantic relationships between scientific concepts and a metric of their historical impact. We show that by using machine learning, SciMuse can predict the impact of new, never-investigated research ideas with high quality. This allows us to suggest new high-impact research collaborations between researchers who have never collaborated before. With the advent of GPT-4, we can follow up and generate these high-impact research topic seeds into full research topics and objectives. Our work isn't merely about predicting impactful research; it aims to catalyze unforeseen collaborations and novel avenues in the natural sciences. Given SciMuse's focus on natural sciences, we hope it will be interesting for the attendees and likewise expect many useful comments and ideas from researchers at the interface between physics and AI.