Issues

• Where does the knowledge come from?

  The domain supervisors precode the knowledge in the global database. Medical knowledge is usually amenable to conceptual graph representations, but other domains may not, due to either the nature of the domain or the ignorance of the supervisors. The designers claim the system can learn by experience according to some utility function, but the exact nature of this method, as well as its successfullness, is unclear.

• Is speed really independent of knowledge?

  Although, there is a professed speed-knowledge independence, the actual LISP machine implementation shows significant performance degradation, even before completion of a two-hour (real time) simulation. The designers attribute this to memory management, but it is not clear that other platforms may avoid this problem, given the possibility of data overload.

• How does it handle control plan interaction?

  In pursuing multiple, simultaneous goals, an AIS may have more than one active control plan. Operations from both plans may appear on the agenda and may be scheduled on successive cycles. It is possible that each operation may undo the result of the other, resulting (at least temporarily) in incoherent behavior.