The reliability of embedded AI systems

An artificial intelligence (AI) software reliability model, which accounts for the trade-off between correctness and response time, fuzzy rather than binary output, faults in fundamental techniques used in AI programs, and the evolving database that AI programs rely on, is discussed. The model can be reduced to a simpler form for most applications, such as stand-alone expert systems and game-playing programs. In its full generality, it is applicable to real-time planning systems involving hardware and software.<
>