Abstract: With the development of deep learning network in recent years, the writing of poems by artificial intelligent (AI) is prospering. Now, AI poem-writing is facing four problems. First of all, AI is unable to appreciate beauty, therefore it fails to differentiate good poems from bad ones, and also, it is difficult for AI to give meaning to a poem. Secondly, AI, which has no subjective consciousness, lacks intention to write a poem, so its works have no significance and metaphors, and cannot be understood by humans essentially. Thirdly, AI writes poems through imitation, which is opposite to the characteristic of poetry, that is, creation. Considering this point, it is difficult for AI to master poetry writing, because poetry has few stable and clear standards. Fourthly, poetry lacks tool value, so there is no necessity to develop AI's poem-writing function alone. Before the emergence of strong AI, these problems mentioned above are less likely to be solved. Nevertheless, we can still learn something from the current AI production of poetry, which is helpful for us to rethink the quality and standards of poetry. Since poetry is mainly aesthetic rather than pragmatic, humanities should be more involved in the development of AI "poets". The prospects of AI are unlimited, so we should be optimistic about them.