I think trying to express terms using levels of behavior is a bit
confusing. I think an easier way of putting it would be to say that
there exists an "agent" which controls the "behavior" of an entity. This
agent runs on a single system somewhere in the network. This agent could
be a user, a complex state machine, a neural-net, etc. It "controls" the
"behavior" of the entity by manipulating the representation of the
entity within the virtual environment through a series of network
"Light-weight interactions" (to borrow the term from the NPSNET
"Internetwrok Infrastructure Requirements for Virtual Environments").
This light-weight interactions get translated by the clients as direct
actions such as physical movement, sounds emitted, change of appearance,
etc.
It is further conceivable, under this model, to have several agents
controlling a single entity using negotiation techniques. This gives
a coherent representation of the entity to the entire network, which might
not necessarily be the case if a behavior script ran on every client's
system.
I'm continuing this thread in the VRML list, but this is precisely the
type of discussion I would like to see transfered to the distributed-vr
list.
-Eric Van Hensbergen