In 2024, MEXT will fund a project (“Development and sharing of generative AI models for scientific research”) at RIKEN to develop generative AI that forms hypotheses through dialogue with researchers. The project aims to train RIKEN’s experimental data and published data to generate hypotheses, experimental plans, etc., to help researchers brush up their hypotheses by interacting with AI. The plan is to first make a prototype available to university researchers in 2024-2025.
In the life and medical science fields, data such as genome, proteomics, and transcriptome are used for learning. RIKEN’s Biodynamics Research Center (BDR) will provide data and collaborate with RIKEN’s Computational Science Research Center.
The goal of developing generative AI is to search for large-scale, exhaustive hypotheses that exceed human cognitive abilities. Furthermore, by combining the generated hypotheses with experiments using automatic experimental robots, it will be possible to test hypotheses with higher throughput than ever before.
Up until now, experiments have been carried out manually over hundreds to thousands of conditions, but by using automated experiment robots, it is now possible to conduct experiments under several million conditions in about one to one and a half months. According to RIKEN’s estimates, by combining generative AI and automated experiment robots, it is possible to shorten the time from ideation to publication of drug discovery research from about two years to about two months.