AMOEBA is a molecular mechanics force field that addresses some of the shortcomings of a fixed partial charge model, by including permanent atomic point multipoles through quadrupoles, as well as many-body polarization through the use of point inducible dipoles. In this work, we investigate how well AMOEBA formulates its non-bonded interactions, and how it implicitly incorporates quantum mechanical effects such as charge penetration (CP) and charge transfer (CT), for water-water and water-ion interactions. We find that AMOEBA's total interaction energies, as a function of distance and over angular scans for the water dimer and for a range of water-monovalent cations, agree well with an advanced density functional theory (DFT) model, whereas the water-halides and water-divalent cations show significant disagreement with the DFT result, especially in the compressed region when the two fragments overlap. We use a second-generation energy decomposition analysis (EDA) scheme based on absolutely localized molecular orbitals (ALMOs) to show that in the best cases AMOEBA relies on cancellation of errors by softening of the van der Waals (vdW) wall to balance permanent electrostatics that are too unfavorable, thereby compensating for the missing CP effect. CT, as another important stabilizing effect not explicitly taken into account in AMOEBA, is also found to be incorporated by the softened vdW interaction. For the water-halides and water-divalent cations, this compensatory approach is not as well executed by AMOEBA over all distances and angles, wherein permanent electrostatics remains too unfavorable and polarization is overdamped in the former while overestimated in the latter. We conclude that the DFT-based EDA approach can help refine a next-generation AMOEBA model that either realizes a better cancellation of errors for problematic cases like those illustrated here, or serves to guide the parametrization of explicit functional forms for short-range contributions from CP and/or CT.