SOFTWARE TECHNOLOGY TO SUPPORT REAL-TIME EMBEDDED ARTIFICIAL INTELLIGENCE BASED APPLICATIONS

THE CONSTRAINTS OF REAL-TIME APPLICATIONS CANNOT BE SATIFIED SIMPLY BY CHOOSING A MORE EFFICIENT IMPLEMENTATION OF "CLASSICAL" AI TECHNIQUES. THE APPLICATION MUST BE EXPLICITLY ANALYYED, DESIGNED AND BUILT FOR REAL-TIME PERFORMANCE. THE USE OF GENERALIZED AI STRUCTURES AND REASONING PARADIGMS CANNOT MEET REAL-TIME PERFORMANCE REQUIREMENTS, ESPECIALLY ON CONVENTIONAL, STANDARD MILITARY HARDWARE. THE SCOPE OF THE PROPOSED EFFORT IS TO EXPLORE THE FEASIBILITY OF ALTERNATIVE SOFTWARE ARCHITECTURES AND STRATEGIES FOR IMPLEMENTATION OF REAL-TIMME SYMBOLIC COMPUTATIONS. THE CLASS OF COMPUTATION TO BE ADDRESSED ARE THOSE WHICH ARE IMPLEMENTED AS SEARCH-INTENSIVE, NP-COMPLETE ALGORITHMS. THE NECESSITY TO USE ALGORITHMS OF THIS TYPE RAISES THE QUESTION OF WHETHER OR NOT REAL-TIME PERFORMANCE CAN EVER BE RELIABLY REALIZED BY AI APPLICATIONS.