As the core communication technology in the 6G era, terahertz technology can effectively address the challenge of increasingly diminishing frequency band resources. It caters to the rapidly growing demand for traffic and connections while enabling large transmission bandwidth. Machine learning algorithms, such as deep neural networks, convolutional neural networks, and short-term memory networks, play a pivotal role in mitigating strong nonlinear effects within 6G transmission systems and are crucial tools for realizing 6G terahertz wireless communication. This review delves into diverse deep learning paradigms implemented in the photonic millimeter wave and terahertz wireless transmission systems. It highlights the notable strides made in leveraging photonic technology for generating ultra-high-speed terahertz wave wireless signals on domestic and international fronts. The paper provides a comparative analysis of different technical approaches. Additionally, the review offers a comprehensive view of traditional and emerging artificial intelligence technologies applied to terahertz communication systems. Finally, it outlines future development directions for terahertz communication technology, focusing on achieving high-speed and high-capacity performance.