Membranes made of polymer are being developed and applied to carbon dioxide (CO2) separation in carbon capture at industrial scale. Computational discovery of new membrane materials relies on data collection and machine learning algorithms for automatically creating new monomers in the process. The physical performance validation of generated monomer candidates within the actual membrane, however, is complicated, time consuming and computationally expensive. In this contribution, we have selected CO2-Permeability and Cohesive Energy Density (CED) as figures-of-merit in our AI inverse molecular design workflow. We compute CED according to group contribution based Fedors-type cohesive energy and the molar volume from SMILES representations. Within the generated results, two representative monomers with similar structures but with different polarities were selected for physical validation. Specifically, we have performed Constant Pressure Drop Molecular Dynamics simulations of separation performance for membranes made of the two monomers, leading to high permeability results. our approach could be extended from homopolymer to copolymer membrane discovery to cover a broader range of materials applications.