Study of project risk in system development implementing artificial intelligence

Abstract

Recently, as the work life reform, the demand of artificial intelligence (AI) systems is significantly increasing. However, it is difficult to apply conventional system development methodology: deciding the specifications from business logics, to AI systems development. It is because that since the output of AI depends on train data, features, and models, it is unknown whether artificial intelligence is adequate to the objective or not without verification. Although the risk can be reduced with Proof of Concept (PoC) in pre-development stage, the effectiveness of the system cannot be guaranteed because of the difference between PoC and actual operation. This paper discuss how to deal with the project risk of AI implementation comparing the actual conventional system development and AI system development.