Jeffrey S. Rosenschein and Vineet Singh Heuristic Programming Project Computer Science Department Stanford University Stanford, CA 94305 Abstract Meta-level control, in an Artificial Intelligence system, can provide increased capabilities This improvement, however, is achieved at the cost of the meta-level effort itself. This paper outlines a formalization of the costs involved in choosing between independent problem-solving methods: the cost of meta-level control is explicitly included. It is often desirable for Artificial Intelligence systems to make use of explicit knowledge about what they know; this tneta-level knowledge allows a program to direct its own activities in an informed and efficient manner [I] [21. The use of meta-level knowledge by a system to control its own actions is called'new-level confrol. If we are to gain efficiene; thi-migh the use of meta-level effort, \'.e must be sure that %'.hat is aved at the base level is not canceled by what is expended at the rnota-level.