The recommendation of similar cases，the assistance of sentencing，and the warning of illegitimate departure are the most typical application models of the current legal artificial intelligence. However，several technical obstacles，for instance，over-reliance on manual intervention in maps construction，the lack of accuracy of human semantic recognition，incorrectness of similar cases’ identification，imperfect data quality of the training，the opacity of sentencing algorithms，and the paradox of granularity in the warning of illegitimate departure，still need to be addressed. Therefore，it is necessary to objectively analyze technical obstacles，and to promote the breakthrough technologies in the field of legal artificial intelligence，enhancing the role of law faculty and user-oriented approach in the research and development process.