Although artificial intelligence technology with its core mechanism, machine learning or deep learning, becomes an influential and trending topic in the field of architecture design, there exist obstacles to the application of automated AI design to the professional architecture practice. The thesis analysis the essential elements required for the machine learning algorithm to accomplish the task of conceptual design involves the capabilities of knowledge, perception, and creativity that are seemingly only possessed by human designers. However, machine learning agents associated with Convolution Neural Networks, Deep-Q Learning, and Generative Adversarial Networks can be proved to achieve the capabilities mentioned above by their mechanism. Although there exists certain constraints, setbacks, and bias, machine learning agent, particularly VQGAN + CLIP, has revealed notable potential in architectural conceptual design where its aesthetic creativity and spatial perception can match with professional human architects because of its remarkable mechanism relating visual objects to abstract texts backed by computing power and big-data era.