Homer

• Heterogeneous knowledge representation

  Homer uses a heterogeneous knowledge representation. That is, the knowledge in Homer is not all in the same form, and thus, is not all usable by each different module, such as for natural language, planning, etc. This results in the duplication of knowledge in different module, so that each module has a copy of some of the same information, with each copy being in a form usable to that module.

• Low efficiency knowledge access

  The Homer system has the problem of a considerable decrease in response time as new episodic memory is added to the agent. Because the agent "remembers" everything that happens to it, navigating through its knowledge base to retrieve information about the past, or for any other task, takes longer as the agent exists for a longer and longer period of time.

• Inconsistent knowledge

  The Homer architecture is very gullible. The agent believes anything it is told, even if it contradicts its world knowledge beliefs.

• 99% theory

  Homer uses the "99% theory" when dealing with objects in the environment. It tries to classify things according to arbitrary characteristic that encompass 99% of the existing instances of any class of objects. For unusual cases, such as an apple two meters in diameter, Homer may be unable to classify the object. However, for 99% of the cases, Homer should be able to identify an object given sufficient knowledge.