ABSTRACT
Within the previous chapters we have seen mention of a number of different ways in which AI does not mesh well with conventional software engineering both theory and practice. In general, the notion of a more or less linear progression from problem specification to tested software system does not look as though it will be readily applicable to AI software systems. In AI we usually have no abstract specification of the problem which is even close to complete or comprehensive, and the basis for comprehensive testing (let alone verification) also seems to be missing on many occasions.
