In General

Each architecture is analyzed along the following dimensions:

This is the general structure of the architecture, including the major components, and the design methodology.

This is the environment where agents built using a particular architecture are intended to function, and are capable of functioning.

These are the capabilities of an agent built using a particular architecture; that is, what an agent can and cannot do. "Capabilities" spans a spectrum, rather than being always a yes or no issue:

  1. Capabilities directly supported by the architecture
  2. Capabilities that have been implemented in agents under the architecture
  3. Capabilities that are not in (1) or (2) above, but which does not seem to have any theoretical obstacles in the way
  4. Capabilities not in (1) or (2), where there are fundamental obstacles in the architecture preventing their natural implementation. Natural implementation is important, because many or all of the architectures surveyed are computationally universal.
So for the purposes of our analysis, (1) and (2) correspond to the architecture having the capability, and (3) and (4) do not.

These are properties of an agent built using a particular architecture, that do not fall under the category of capabilities.

These are the circumstances surrounding an architecture and the issues that are brought out by the researchers.