Photopharmacology is a growing field where light is coupled to therapeutic action, enabling noninvasive, biological control. Many photopharmacological agents are irreversible, however, dynamic control is possible through the introduction of a photowitchable ligand, or molecule that can be interconverted between two structural states by absorption of a photon. The prototypical, well-studied, photoswitch is an azobenzene; however, azoheteroarenes have shown excellent properties such as absorbances in the visible range and high thermal barriers, ideal for biological applications. The discovery of new photoswitches is a time-consuming process when using individual density functional theory (DFT) calculations to predict the absorption. Traditional methods utilize high-throughput screenings of thousands of molecules but with computations still ranging from hours to days, it could take months before an ideal photoswitch is found, if at all. Rather than individually computing azoheteroarenes, we utilize machine learning (ML) to select the most promising molecules, then use DFT to confirm its properties. We have implemented an active search ML method, based on human feedback, to expedite the discovery of azoheteroarenes for biological applications. After training the model with a molecular library of 231,000 photoswitches, the algorithm selected 75% more molecules that absorbed between 450-549nm when compared to randomly selecting molecules. By performing 1500 DFT calculations over 30 iterations, we were able to effectively train a machine-learning algorithm to select molecules that absorb visible light. The active search has proven to be a valuable tool to search for new molecules with desired properties by substantially decreasing the time to discover new photoswitches.