Although Soar has an unlimited knowledge base, the organization of this knowledge into the productions of long-term memory and the concomitant use of the production matcher result in an inherent attentional mechanism. Only those productions that match to the current working memory need to be considered at any time. This contrasts to systems that use a first-order logic representation in which, outside of some architecturally-imposed modularity, all knowledge must be considered at each decision step. This attentional mechanism, since it depends only upon the contents of working memory, also leads to Soar's distractibility.
Go to a discussion of this property for multiple architectures.