Jump to the MAX Overview page.
Jump to the Comparative Capabilities page.
Jump to the Home page.
Capabilities
Overview:
MAX is meant to be a standard implementation of metareasoning through a
dedicated architectural component.
The other capabilites of MAX are implemented through its knowledge.
Because MAX is meant to be a general architecture, where the capabilites that
are needed by one user or another can be added into the system, the capabilites
of MAX are presented as potential capabilites for the most part. In this
analysis we only present the capabilites which are specifically in the
architecture.
The authors
seem to think that any capability which can be expressed through declarative
knowledge can be supported in MAX. This would rule out uncertainty reasoning
and non-monotonic logic however.
Meta-Reasoning
Metareasoning needless to say is MAX's raison-d'etre. The interpreter, which
is a procedural production system interpreter, uses a rule-based forward
chaining engine to guide the process. It matches its knowledge with the
declarative knowledge already in the system, and assembles and modifies
information as it deems necessary. This presumably would allow the system
to decide when it does not have enough knowledge to solve a problem, to
recognize problems that it has already solved, and prioritize behaviors in
order to maximize system efficiency.
Reactivity
The monitor mechanisms, which can run in the background, can allow for and
react quickly to unexpected events. The forward chaining of the procedural
knowledge in the kernel dictates that an event which has been deemed
crucial to the system can be recognized on one pass.
Taskability
The knowledge of the system is uniform enough so that we can update a list
of commands for a MAX agent to execute, and it should be able to handle them
in a reasonable amount of time. The metalevel nature of the knowledge should
also be able to inform the user if the task is impossible to execute.
Learning
The basic learning mechanism in MAX is the ability to modidy its knowledge
through the metalevel kernel of the system. However, this is a limited
definition of learning; rather it should be considered a stepping stone for
learing. This has been demonstrated by the implementation of other learning
systems in Max agents, such as explanantion based learning. Therefore the
learning mechanism could be considered one of the capabilities that Max
supports, but does not implement in a broad sense.
Perception
A perception mechanism is implied in the definition of the MAX architecture.
The system was set up for a real time robotic environment from the
description. Even though no
mechanism is explicitly defined, the
perceptions/actions cycle was constructed to handle perception.
Other
It should be mentioned that many of the other capabilites that are not
mentioned here such as interruptability, planning, and prediction have been
implemented using a MAX architecture. Because of the general nature of this
architecture, the capabilites included here are those that follow from the
definitions of the architecture itself, and not from its sum total
possibilities.