Computational Limitations

A real-world agent is limited in its computational abilities. This limitation becomes apparent in several ways. First, the agent does not have the resources to perform an infinite number of tasks or decisions in parallel. So the agent must choose what to do at a particular point in time. The time allowed to perform this selection is limited. Since the agent often does not have the ability to search through all of its knowledge and find the best possible action, some sort of search or decision strategy may be needed. Specific actions may have different levels of importance , so time may be an even more important factor. By focusing on a particular task or locale of observation, the agent can limit the amount of resources needed or the knowledge obtained. By using deliberative learning, the agent decides what knowledge it wants to learn, and can limit itself. Another way to reduce the knowledge that can be obtained is by limiting the agent to restricted or virtual environment. Meta-knowledge is knowledge about knowledge. This can often be helpful in making a decision about what to do or what to learn.

Architectures which ignore computational limitations:

Architectures which limit the amount or type of knowledge obtained:
Using focusing:

Using deliberative learning:

Using some other method:

Architectures which limit the planning/decision-making process:
Using meta-knowledge or meta-reasoning:

Using some other method:

Other characteristics of the Environment and Agent Body

Back to the Title Page