With high-resolution imaging and an expansive field of view, near-infrared sensitivity, precise pointing control, and high survey speed, the Nancy Grace Roman Space Telescope will enable unprecedented capabilities to address key cosmological questions. To explore the expansion and structure of the universe, the telescope will provide wavefront stability of <1 nm and utilize a Wide Field Instrument comprising 18 4k × 4k near-IR detectors. With the Roman mission gathering data from millions of galaxies, artificial intelligence will be a crucial asset in processing an ultra-deep field. In this work, we discuss how both statistical and machine-learning-based modeling can lead to novel discovery on this front. Machine learning approaches, such as convolutional neural networks for large quantities of imagery, are particularly suited to analyzing large cosmological databases efficiently, but the explainability of the results is a potential limitation. As the data collected will be made open-access through the Mikulski Archives for Space Telescopes (MAST), the cosmological and astrophysical communities will be able to collaborate across institutions and disciplines to perform state-of-the-art analyses, improving upon developed benchmarks.