We present an innovative approach to the methodology of dynamical modelling, allowing practical reconstruction of the underlying dark matter mass without assuming both the density and anisotropy functions. With this, the mass-anisotropy degener-acy is reduced to simple model inference, incorporating the uncertainties inherent with observational data, statistically circumventing the mass-anisotropy degeneracy in spherical collisionless systems. We also tackle the inadequacy that the Jeans method of moments has on small datasets, with the aid of Generative Adversarial Networks: we leverage the power of artificial intelligence to reconstruct non-parametrically the projected line-of-sight velocity distribution. We show with realistic numerical simulations of dwarf spheroidal galaxies that we can distinguish between competing dark matter distributions and recover the anisotropy and mass profile of the system.