Human divergent exploration capacity for material design: A comparison with artificial intelligence

Abstract

Applications of artificial intelligence (AI) to material design have attracted increasing attention in recent years. Although AI-aided material design holds great promise for some applications, whether it has surpassed human creativity remains uncertain. The aim of the current study was to compare the divergent exploration capacity of AI with that of humans on a material design task. Human participants were asked to find a high-performance lubricant molecule under searching conditions comparable to a state-of-the-art AI system. Results indicated that, on average, AI was able to find significantly better lubricant molecules. However, the best molecule found by AI fell short of the best molecule found by a human participant. Furthermore, the structural characteristics of the molecules found by AI and human participants differed significantly. These findings suggest that a state-of-the-art AI system is capable of surpassing human divergent exploration capacity in material design, as in other fields in which AI has advanced. Nevertheless, our results also demonstrate that human intelligence and AI can play complementary roles in covering a broader search space. This investigation opens up new possibilities for collaborative systems involving both AI agents and humans in material design.