Age hardening induced by the formation of (semi)-coherent precipitate phases is crucial for the processing and final properties of the widely used Al-6000 alloys despite the early stages of precipitation are still far from being fully understood. This crucial step in the technology of Al-based alloys is studied by means of multi-scale simulations that include first-principles atomistic modeling, surrogate models based on statistical learning, as well as kinetic Monte Carlo and continuum elasticity models to bridge time and length scales. We begin with an analysis of the energetics of nanometric precipitates of the meta-stable beta'' phases (that play a crucial role in this system) identifying the bulk, elastic strain and interface energies that contribute to the stability of a nucleating cluster. Results show that needle-shape precipitates are unstable to growth even at the smallest size beta'' formula unit. This study made it possible to develop a semi-quantitative classical nucleation theory model, including also elastic strain energy, that captures the trends in precipitate energy versus size and composition. This validates the use of mesoscale models to assess stability and interactions of beta'' precipitates. Studies of smaller 3D clusters also show stability relative to the solid solution state, indicating that the early stages of precipitation may be diffusion-limited. Our results thus point toward the need for a systematic study of the energetics of aggregates in the Guinier-Preston zone regime, and the interactions between those aggregates and vacancies and/or trace elements to understand and fine-tune the behavior of Al-6000 alloys in the early stages of precipitation. To enable full atomistic-level simulations of the whole precipitation sequence of this important alloy system, two Neural Network (NN) potentials have been created by representing just 2-body interactions and including also the 3-body interactions. For the latter, we developed an automatic scheme to determine the most appropriate representation of the structural features of this ternary alloy. Training of the NN uses an extensive database of energies and forces computed using Density Functional Theory, including complex precipitate phases. The NN potentials accurately reproduce most of the properties of pure Al which are relevant to the mechanical behavior and formation energies of small solute clusters and precipitates that are required for modeling the precipitation and mechanical strengthening. This success not only enables future detailed studies of Al-Mg-Si but also highlights the ability of machine learning methods to generate useful potentials in complex alloy systems. Finally, we used this NN potential to implement a kinetic Monte Carlo scheme to study the formation of pre-precipitation clusters. While quantitative accuracy will probably require further refinement of its training set, to achieve a more complete description of the interactions between solute atoms and vacancies, we could already observe some of the key mechanisms the determine the ultra-fast formation of aggregates. This work lays the foundations for a thorough investigation of the behavior of Al-6000 alloys over time and size scales that are technologically relevant and demonstrates a combination of atomistic modeling techniques that could be adapted to a large number of similar metallic alloys.