Model-Based Validation of Intelligence
Papers from the AAAI Spring Symposium
Lina Khatib and Charles Pecheur,Cochairs
The aim of this interdisciplinary symposium is to bring together researchers in software verification, formal methods, and intelligent systems to compare techniques, explore possible opportunities for cross-fertilization, and look for answers to the fundamental question of the nature of "validated intelligence." In particular, we are interested in the potential synergies between model checking and model-based reasoning. Artificial intelligence is finding many promising applications in safety-critical areas such as avionics, factory control or space exploration. To live up to these promises, intelligent software has to provide evidence that it meets the stringent quality control requirements imposed by such applications. However, we are still far from producing fully validated intelligent systems. In fact, it is not even clear that we know how to specify validation criteria. The need for analysis tools that guarantee coverage and robustness is apparent. Model checking is advertised as an efficient way to exhaustively search all behaviors of a program. It has been successfully put into practice for more traditional software. Another approach, model-based systems, has recently had great success in mission critical applications and also provides a formal basis to specify intelligent behavior. An important question arises: what is the relationship between model checkers and model-based reasoning systems? What is the relationship between their representation structures? What are the relationships between their reasoning mechanisms? Is it possible they are using different terminologies for similar concepts?