Abstract
Abstraction is a fundamental feature of human-level intelligence. But it is not clear how to combine knowledge on different levels of abstraction. This paper examines the use of hierarchical knowledge for heuristic problem-solving algorithms, regarding three options for integrating hierarchical knowledge into heuristic search: as a state evaluation heuristic, as a search-guiding heuristic and as a hierarchical search strategy. The Traveling Salesperson Problem serves as an example of a problem-solving task and the different strategies are evaluated with respect to tour length, robustness against misleading hierarchy assignments and acceptability of results by humans.
It turns out that the most effective and stable results can be achieved with hierarchical knowledge as a search-guiding heuristic combined with other heuristics.
