Soar - Knowledge-Rich Environments

Description

• Some architectures attempt to compensate for environments with large amounts of data by attempting to learn only that information deemed useful. SOAR, on the other hand, internalizes reflexively any conclusion that it reaches and never discards any piece of knowledge.
• While it is possible that SOAR's learning mechanism, chunking, will produce useless chunks, SOAR deals with this problem by matching its rules so quickly that excess chunks do not slow down the system appreciably.
• This allows SOAR to operate in environments which are rich in knowledge, but also those in which it is difficult or impossible to judge the usefulness of any piece of knowledge a priori.
• It should also be noted that the same mechanisms that make SOAR very useful in environments containing large amounts of data make it very inappropriate for use in environments with little data. SOAR is intended to learn from every action that it takes, and if there is nothing to learn then this very fundamental capability of SOAR is wasted. For example, while SOAR can be applied to tasks with clearly defined algorithms, it is much more appropriate to apply SOAR to tasks without such algorithms and let SOAR derive its own algorithm.