Reasons for Success (and Failure) in the Development and Deployment of AI Systems

Nestor Rychtyckyj, Alan Turski

The AI community annually gathers at conferences such as the Innovative Applications of Artificial Intelligence (IAAI) to celebrate the achievements of those AI applications that were successfully deployed in a production environment. However, the applications that do not get deployed successfully are mostly ignored and not discussed publicly. There has been a large amount of research done from both the software engineering aspect as well as the project planning view to determine why the failure rate of software projects remains so high (Standish 2004). The use of innovative technologies, such as AI, complicates the software development process even further. At Ford Motor Company we have had our share of success and failure with AI applications — in this paper we will describe both types of projects and try to draw conclusions of what factors are most relevant in the successful deployment of AI projects. The paper is organized as follows. Section 1 will discuss the history and eventual failure of an AI system due to implementation issues. The next section will describe the organizational challenges that need to be addressed by an organization that wants to utilize AI technology. This is followed by a discussion of the technical challenges that are faced in the development and maintenance of AI applications. In the last section, we will summarize the lessons that we have learned over time and discuss how they can be applied to future AI system development.

Subjects: 1. Applications; 1.7 Expert Systems

Submitted: May 8, 2008

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.