In General
The following catagories listed below represent general architectural
properties. Under each heading, a description of the specific
property is given, as is a partial list of the architectures containing that
property. A more complete description of that property as
implemented in any specific architecture can be found by following that
architecture's link.
Knowledge Base Properties
Knowledge Representation
One way in which the knowledge contained in the architecture's knowledge base can be
catagorized is by whether it is homogeneous or heterogeneous. A homogeneous knowledge base
would be one in which all the information is of the same form. An example could be
a knowledge base in which everything (facts, control knowledge, etc.) is represented in a
frame-based representation. A heterogeneous knowledge base is one in which the knowledge
appears in different forms, i.e., one form for data, another for control knowledge, etc.
Homogeneous knowledge representation
Heterogeneous knowledge representation
No knowledge representation
Knowledge Access Efficiency
As the knowledge bases of these architectures grow increasingly large, an efficient way of
accessing specific knowledge in the knowledge bases becomes increasingly important. While
this is an issues for almost all architectures, only those architectures that are of special
interest are considered below.
High knowledge access efficiency
Low knowledge access efficiency
Consistent Knowledge Base
In an uncertain environment, or an environment where an agent can only have imperfect
knowledge, it is desireable for the agent to be able to keep its knowledge consistent
by updating its beliefs based on new information from the world.
Consistent knowledge
Inconsistent knowledge
Declarative vs. Procedural Knowledge
Another way to classify some of the knowledge in the system is as either declarative
knowledge or procedural knowledge. Declarative knowledge, sometimes called Glass Box
knowledge, is open to inspection by all modules in the system. The derivation
of the knowledge is included along with the pre- and post-conditions associated with it.
Procedural knowledge, also known as Black Box knowledge, does not contain the derivation of
the knowledge. The pre- and post-conditions are given, but no information of the processes
that occured during the state change are included.
Declarative knowledge
Procedural knowledge
Meta-Knowledge
Meta-knowledge is any knowledge the system has about its own workings. This could include
probabilities of architectural module efficiencies or dependabilities. In general,
meta-knowledge is knowledge the system has about its own knowledge or operation.
Behavioral Properties
Reaction/Operational Speed
In dynamic environments, an agent often needs to have fast reaction speed in order to
respond to sudden changes in the environment. For an agent to operate in real-time,
the architecture must be able to react to changes in the same time scale that they occur.
Deliberative vs. Reflexive Learning
Deliberative vs. reflexive is one way to differentiate between learning mechanisms in
different architectures. With deliberative learning, the architecture can decide what to
learn, when to learn it, and why it learned it. Reflexive learning, on the other hand,
operates exactly as the name implies. Learning always happens under certain conditions.
The architecture has no choice in the matter.
Deliberative learning
Reflexive learning
Asychronous Operation
In modular architectures, the operation of the architecture can follow a set pattern, such
as plan then execute, or the modules can operate independantly of the other modules,
asynchronously. Below are a few of the architectures that exhibit asynchronous
operation.
Impasse Driven
Impasses occur when the agent is unsure of the appropriate action or knowledge. In
order to solve the problem, the agent creates a subgoal of resolving the impasse. Through
recursive calls such as this one, the architecture eventually reduces the problem to a
subproblem that it can operate on.