The current approach to deep learning involves training models from scratch every few months. However, the idea of evolutionary AI and lifelong learning aims to develop models that can be built upon previous iterations, creating a database of neural networks that can be refreshed with new weights as needed.