To integrate existing AI techniques into a consistent system, an intelligent core is needed, which is general and flexible, and can use the other techniques as tools to solve concrete problems. Such a system, NARS, is introduced. It is a general-purpose reasoning system developed to be adaptive and capable of working with insufficient knowledge and resources. Compared to traditional reasoning system, NARS is different in all major components (language, semantics, inference rules, memory structure, and control mechanism).
NARS is an intelligent reasoning system, whose interaction with its environment can be described as a stream of input sentences in a formally defined language and a stream of output sentences in the same language. These two streams are called the system's "experience" and "responses", respectively (Wang, 1994; Wang, 1995a; Wang, 1995b). Each sentence in the language represents an inheritance relation between two terms. By definition, a sentence "S C P" indicates that the subject term "S" is in the extension of the predicate term "P", and "P" is in the intension of "S". Because the relation "C" is defined to be reflexive and transitive, "S C P" also indicates that "S" inherits the intension of "P", and "P" inherits the extension of "S".
Multi-relational networks are used extensively to structure knowledge. Perhaps the most popular instance, due to the widespread adoption of the Semantic Web, is the Resource Description Framework (RDF). One of the primary purposes of a knowledge network is to reason; that is, to alter the topology of the network according to an algorithm that uses the existing topological structure as its input. There exist many such reasoning algorithms. With respect to the Semantic Web, the bivalent, monotonic reasoners of the RDF Schema (RDFS) and the Web Ontology Language (OWL) are the most prevalent.
In a probability-based reasoning system, Bayes' theorem and its variations are often used to revise the system's beliefs. However, if the explicit conditions and the implicit conditions of probability assignments `me properly distinguished, it follows that Bayes' theorem is not a generally applicable revision rule. Upon properly distinguishing belief revision from belief updating, we see that Jeffrey's rule and its variations are not revision rules, either. Without these distinctions, the limitation of the Bayesian approach is often ignored or underestimated. Revision, in its general form, cannot be done in the Bayesian approach, because a probability distribution function alone does not contain the information needed by the operation.