Frame-Based Knowledge

Origin. Minsky's frame concept (1974) was extensively developed in Soviet AI at the Institute of Control Sciences and elsewhere, 1970s-1980s. Soviet researchers integrated frames with logic programming and semantic networks, emphasizing practical knowledge engineering over formal semantics.

Mechanism. Knowledge is represented as frames — structured records with slots for attributes, values, and relations. Frames support inheritance from parent frames, default values for typical cases, and attached procedures (demons) that trigger on slot access. The representation is closer to human conceptual structure than predicate logic, naturally supporting typical-case reasoning and exception handling. Frames can represent both classes (prototypes) and instances, taxonomic relations (is-a) and compositional relations (part-of).

Procedure. Design and implement frame-based knowledge systems: (1) Model the domain as frame hierarchy — identify the major concepts and their taxonomic relationships. Parent frames represent general categories; child frames specialize them. (2) Define slots — for each frame, what properties does it have? Define slot names, value types, constraints, and default values. Defaults capture typical cases; specific instances override when needed. (3) Define relations — is-a relations enable inheritance; part-of relations capture composition. Other relations (causes, requires, located-at) model domain structure. (4) Attach demons — if-needed demons compute slot values lazily on demand; if-added demons trigger when values are asserted to maintain consistency; if-removed demons handle deletion. Demons encode procedural knowledge. (5) Implement inheritance — child frames inherit slots from parents. Design the inheritance strategy: single inheritance is simple; multiple inheritance requires conflict resolution. (6) Query and reason — access knowledge by pattern matching or direct slot access. Use inheritance to avoid redundancy; specialize by overriding defaults.

Applies to. Expert systems. Knowledge bases. Semantic modeling. Ontology design. Any domain where structured knowledge with defaults and inheritance matches the domain structure.

Limitations. Multiple inheritance creates ambiguity when ancestors define conflicting values for the same slot. Default reasoning is non-monotonic: adding information can retract conclusions, complicating truth maintenance. Frame-logic integration was not fully formalized, so frame systems can behave unexpectedly in edge cases. Soviet work emphasized engineering practice over formal semantics, which limited theoretical guarantees.