The wireless spectrum is a scarce resource nowadays, and traditional static fixed-frequency access is no longer suitable for modern wireless technology. To meet the growing demand for spectrum access, dynamic access techniques have emerged. One of the main challenges is to enhance dynamic spectrum access (DSA) efficiency while ensuring a good quality-of-service (QOS) for primary users. Recently, artificial intelligence (AI) based DSA methods have attracted interest because they facilitate smart decision-making. Also, integrating Deep Learning into real DSA systems is not as easy as it looks like. These models spend huge amounts of energy in reliance alongside the processing time, due to the immense amount of training data they need. The problem is much worse when energy efficiency and low latency is the goal. As we look into incorporating AI into DSA for this article, we will look in dept. at different considered approaches, new trends, and still unsolved prominent problems.