This is interesting because it probably most accurately describes the current approach to AI. Developing a fully realized "hard AI" system is much too complex a

This is interesting because it probably most accurately describes the current approach to AI. Developing a fully realized "hard AI" system is much too complex a goal to be completed in any reasonable timespan. Instead, most AI researchers are working on more manageable subproblems (machine learning, computer vision, semantic processing, etc) -- tools if you will. We can't be sure, but I personally like to think that these more approachable advances in building an AI toolset will help to work on the larger and more complex AI problems in the future.