Quality Measures and Assurance for AI Software

This report is concerned with the application of software quality and assur-ance techniques to AI software. It describes work performed for the NationalAeronautics and Space Administration under contract NAS1 17067 (Task5) and is also available as a NASA Contractor Report.The report is divided into three parts. In Part I we review existing softwarequality assurance measures and techniques{those that have been developedfor, and applied to, conventional software. This part, which provides a fairlycomprehensive overview of software reliability and metrics, static and dy-namic testing, and formal speci cation and veri cation, may be of interestto those unconcerned with AI software. In Part II, we consider the char-acteristics of AI-based software, the applicability and potential utility ofmeasures and techniques identi ed in the rst part, and we review thosefew methods that have been developed speci cally for AI-based software. InPart III of this report, we present our assessment and recommendations forthe further exploration of this important area. An extensive bibliographywith 194 entries is provided. Quality Measures and Assurance for AI Software1John RushbyComputer Science LaboratorySRI International333 Ravenswood AvenueMenlo Park, CA 94025Technical Report CSL-88-7R, September 1988(Also available as NASA Contractor Report 4187)1This work was performed for the National Aeronautics and Space Administra-tion under contract NAS1 17067 (Task 5).