In this thesis, we explore how a classical potential can be constructed by fitting an artificial neural network to the potential energy surface of an ab initio calculation. A Hartree-Fock implementation is explained in detail and used to calculate the potential energy surface. Further, we provide details on how a molecular dynamics code is implemented. This is verified by a simulation of argon crystallization. Results from the Hartree-Fock and molecular dynamics implementations are well aligned with those found in the literature. To bridge these two implementations in a simulation of hydrogen dissociation, the potential energy surface of hydrogen is fitted with the Fast Artificial Neural Network Library and applied in molecular dynamics. The results are on par with a study using the Kohen-Tully-Stillinger potential. In comparison, the artificial neural network potential is parameterized without empirical data nor initial assumptions about the form of the potential function, but suffers a performance loss by a factor of 10 - 20. Finally, we discuss different techniques used to visualize molecules, including isosurface and volumetric rendering of electron densities, and billboard rendering of systems with millions of atoms.