Building upon the groundwork laid by NAISS 2023/22-477, our research aims to delve deeper into the intricate relationship between optically relevant properties (such as dipoles and polarizabilities) and the structural dynamics of condensed materials. We are particularly intrigued by the behavior of ionic systems dissolved in water. It has been experimentally measure that the conductivity of certain ions in water changes when optical fields couple to the system [https://doi.org/10.1039/D3SC03364C]. Predicting such phenomena presents unique challenges due to the extreme anisotropy, local order, long-range disorder, and highly diffusive motion -- requiring the simulation of large numbers of atoms (large supercells). Predicting the optical characteristics of such systems is traditionally arduous and resource-intensive when relying solely on established density functional theory (DFT) methodologies. We anticipate that a well-trained Tensorial Neuro-Evolution Potential (TNEP) could resolve this challenge and deliver insight about the microscopic mechanism. In this project, we utilize a machine learning approach based on Tensorial Neuro-Evolution Potential (TNEP), a type of neural network-based machine learning model, to learn and predict the dipole moments and polarizability of a ions dissolved in water. In combination with nuclear forces obtained via the same NEP approach, it provides a way to estimate the interplay between ionic conductivity and optical characteristic. The success of NAISS 2023/22-477 illustrated the strength of the TNEP approach. In the previous iteration of this project, we have been able to train a satisfying TNEP model. We will be now required to refine this in order to estimate changes in conductivity sufficiently accurate. The required diffusion constants or auto-correlation functions of the dipole moment are quite sensitive to errors. Since our preliminary models exhibit a realistic dynamic, this will merely require a more refined training and better statistics on the inference. In conclusion, the project requires TNEP models for large ion-water systems and an active learning approach (iterative evaluation and retraining of the TNEP models). More resources are required to train meaningful models as the underlying system has highly an-isotropic and multiple species of ions should be considered. Initial models show good, but not accurate enough performance for conductivity measurements.