Automated scientific discovery has the potential to enhance progress across various scientific fields significantly. However, assessing an AI agent’s ability to use comprehensive scientific reasoning is challenging due to the high costs and impracticalities of conducting real-world experiments. While recent neural techniques have led to successful discovery systems for specific problems like protein folding, mathematics, and materials science, these systems typically circumvent the complete discovery process, focusing instead on systematic searches within a predefined hypothesis space. This raises the question of how much more can be accomplished if AI is utilized throughout the scientific process, including creativity, hypothesis generation, and