This study investigates the development of efficient and effective bandwidth slices through the collaboration of AI-based methods across multiple Internet Service Providers (ISPs). A novel approach, Self-Adaptive Learning (SAL), is introduced, utilizing a one-step process to determine the quantity and location of bandwidth resources for each slice, catering to a wide range of use cases on shared infrastructure. The SAL method integrates a new performance index, the MCTS-level, which combines decision tree and sampling methodologies to approximate optimal solutions. A higher MCTS-level indicates a more comprehensive exploration of solutions by the Monte Carlo Tree Search (MCTS) and, consequently, higher throughput. Simulation results reveal that the SAL method achieves an MCTS-200 level, surpassing the previous benchmark of MCTS-100, demonstrating the potential of the SAL approach for enhancing network performance and efficiency.