The discovery of new materials has been a time consuming and resource intensive effort. The following trial-and-error approach is typically employed: identifying known materials with properties similar to the new material’s target properties and then modifying or combining them for achieving the desired outcome. The approach is driven by a specialist’s knowledge, laboratory experimentation, and it can take years to yield results. The computer revolution has brought about powerful simulation techniques, such as the density functional theory (DFT) and high-throughput computational materials screening and design methods (HCMSD), have enabled substantial speed-up of the process. However, one limitation of HCMSD is that it usually relies on time consuming ab initio calculations, such as DFT simulations. However, the large number of computations required for probing the phase space or performing materials screening can render HCMSD impractical. In recent years, artificial intelligence (AI) powered material design attracted considerable attention. Machine learning (ML) uses existing experimental and simulation databases to achieve faster material screening and design. In particular, inverse materials design based on ML method shows great potential, which relies on an algorithm that creates optimized molecular structures based on a pre-defined feature vector containing a set of materials target properties. In this work, Ronaldo Giro et al. from IBM Research, employed reverse material design method based on ML to establish a fully automated silicon material design process and completed the design and physical validation of polymer membrane materials that filter carbon dioxide under realistic temperatures and pressures. The authors established a data set based on the monomer structure, polymer glass transition temperature, half-decomposition temperature and carbon dioxide permeability to implement graph-based design of optimized monomer units and conduct physical validation of polymer membrane gas permeation through molecular dynamics simulations. This work opens a pathway for advancing ML-generative design beyond small-molecule applications and will substantially accelerate the discovery of complex materials for scaled applications.