Jump to the AIS Overview page.
Jump to the Comparative Capabilities page.
Jump to the Home page.

Capabilities


 

Planning

AIS does both complet and partial plans. The planning capabilities are implemented by the
dynamic control architecture, specifically in the agenda manager submodule. Planning takes into account both the interactions of the possible actions (as in regular planning) and the urgency of the situations. The types of plans generated will depend of the specific knowledge programmed into the system. In Guardian the plans are possible treatments for the patient.


Prediction

While the knowledge base in AIS can probably anticipate some of the possible happenings, the system was designed more as reactive rather than predictive. The
dynamic control architecture cycle is a simple plan-schedule-execute, where the plans are meant to handle situations that the system thinks need attention. Granted, the plans generated could include some predictive factors, but, in general, they are meant as simple courses of actions for handling past occurences. Our intuition is further confirmed by their implementation of an asynchronous I/O subsystem which can add new information from outside events to the knowledge base at any time. This is information that the system does not bother (or can not) predict.

We should mention, however, that all the plans (i.e. agendas) generated by the system are meant to achieve some goal. The system is, therefore, implicitly, predicting that the execution of its agenda will have the desired effects. This might or might not be the case.


Meta-Reasoning

The AIS system does not provide for meta-reasoning techiniques. However, its blackboard architecture does provide for the addition of new
cognitive skills. These could include the ability to do meta-reasoning, but it is not clear whether or not the basic blackboard architecture would have to be modified to allow for such modules.


Reactivity

One of the main objectives of the AIS architecutre was to build a system that could provide real-time respones to problems of different urgency. Reactivity was implemented by allowing the planning, scheduler and executor subsytems to be interruptable so that if something important happens in the world it would be dealt with promptly. The system determines which happennings are important by using the filters in its
dynamic I/O channels. These filters can be changed by the system itself, while it is working, so that it can focus on certain aspects of the environment that it deems more important at the moment. The filters can change the behavior of the system from, completly aware of every event in the environment, to completly oblivious to the outside world.


Taskability

The AIS architecture does not, as far as we could tell, provide the programmer the ability to give the system specific goals to be achieved, while the system is running. Any goals the system is to achive must be given prior to it starting its execution. After this all the actions performed will be dictated by the agenda manager which performs actions based solely on inut received from its perceptors and knowledge it has stored.


Learning

AIS provides no learning mechanism.


NLP

AIS has not been used to provide a natural language facility, altough it probably could be.


Interruptability

The AIS is a highly interruptable architecture, thanks to its
satirfycing reasoning cycle. This cycle, in fact, implements all the interrupts available to the system. Basically, it is the samve cycle used in the dynamic control architecture but expanded so that all the step in the cycle are interruptable and resumable.


Navigation/Manipulation

The AIS architecture is general enough to allow for any type of navigation and manipulation subsystems. In the
Guardian implementation the system does no navigation, it simply manipulates the environment (e.g. by giving advice or signalling that there is an emergency). Any new type of manipulation or navigation system can be used with AIS, as long as an appropiate dynamic I/O channel is implemented.


Coherent Behavior

An AIS architecture would show coherent behavior, as exemplified by the
Guardian implementation, because the actions are determined by the knowledge systems which has a particular set of goals that it tries to satisfy.


Perception

The AIS architecture is general enough that any type of perception substirem can be incorporated into it. This can be done as long as the proper
dynamic I/O channel is implemented.


Other

There are other capabilities that the AIS architecture stresses as very important. The authors were very concerned with the systems ability to satisfy some real-time constraints. AIS, therefore, provides this capability. Not only that but the system claims to degrade gracefully as the time boundaries become shorter. That is, the quality of the solution it finds to a problem decreases as the inverse (more or less) of the time it is allowed to think about it. AIS also provides the capability of choosing a focus of attention to which it can dedicate most of its resources. Finally, the system claims to have speed-knowledge independence, the response time of the system does not decrease as we add knowledge to it.