A self-teaching intelligent-personal-agent interface (IPAI) for the VA
X/VMS operating system has been designed and tested Its performance is
based on (1) a priori knowledge about the operating system in use, an
d about the specific user environment, and (2) learning by observing a
user interact with the computer system. The interface combines a synt
ax-based approach, a memory-based approach, and an approach based on i
ntelligent personal agents. The implemented system was tested by users
who completed evaluation forms. The best and worst properties of the
implemented system are highlighted. Experiments show that the memory-b
ased intelligent-agent approach enables designing very flexible and ad
aptable systems that learn from past events without burdening the user
. Memorized events enable good performance even in situations that wer
e not anticipated by the designers. The classical rule based approach
achieves similar performance for expected tasks, yet is less adaptable
to changes in environment and demands more programming efforts. The i
ntelligent-personal agents approach has been shown to be a promising d
irection with the expected trust and competence problems. For mundane,
easy tasks, the intelligent-personal-agents approach is very close to
application maturity, thus enabling a step-up in man-machine communic
ation.