Traditional Information technology models are based on conventions established outside the models themselves: meanings and relationships are predefined, they exist primarily in the mind of the designer, and must be embedded into data formats and program code at design time. As a consequence, it is difficult, if not impossible, to automate the interpretation of the models in a flexible manner: when something changes in the model conventions, corresponding changes must be introduced manually by a human being.
Semantic models encode meanings as part of the modeling paradigm, so that models can be interpreted by the automatic processes without human intervention. It is also possible to encode different kinds of knowledge (domain, heuristic, algorithmic) using the same formalism: this makes it possible to ensure interoperability among different kinds of systems.