Recently, there has been a growing interest in using artificial intelligence (AI) technology to design metamaterial devices. This approach reduces the reliance on traditional design methods that require expertise in electromagnetics theory and simulation, resulting in a more efficient design cycle. Despite the progress made in this field, device design in the terahertz band remains relatively underdeveloped. This paper scrutinizes traditional design methods of terahertz metamaterial devices from the perspective of device function and focuses on the current research status of tunable multifunctional metamaterial devices. The article discusses how artificial intelligence techniques, such as machine learning, evolutionary algorithms, and deep learning, can aid in the optimization of metamaterial devices based on structural parameters and electromagnetic response. Finally, this section discusses future developments and challenges in the field, providing useful references for researchers engaged in related studies.