Depression is one of the most prevalent psychological and psychiatric disorders today. Traditional diagnostic approaches have subjective limitations that rely on doctors’ experience; artificial intelligence applied to the detection of depression can perform specific data analysis, thus assisting doctors engaged in the diagnosis of psychological and psychiatric disorders to improve diagnostic efficiency. The first generation of knowledge-driven depression detection can only solve the problem of certainty in a fully informed and structured environment, where the choice of depression-related features directly affects the recognition results. The second generation of data-driven depression detection needs to be driven by a large amount of data, and the results obtained under black box models alone are not credible enough. The third generation of depression detection ensures the processing of information through information fusion, combining heterogeneous information from different sources high quality, combining the respective advantages of the first two generations of depression analysis systems, manually extracted features combined with depth features can better tap into depression feature information, decision fusion is highly fault tolerant, and the fused model can increase the reliability of the results and provide a more comprehensive analysis of depression data. This paper examines a large amount of relevant literature, summarises and analyses existing research results, and points out the future development direction of AI depression diagnosis research.