Frame (artificial intelligence)

• 1. Minsky's Foundational Proposal

  ›

  In June 1974, Marvin Minsky published a paper titled A Framework for Representing Knowledge at the MIT Artificial Intelligence Laboratory. The document introduced frames as a method to handle stereotyped situations that computers struggle to process. Before this proposal, researchers faced huge solution spaces even for simple tasks like extracting phonemes from audio or detecting object edges. Minsky argued that humans use stored knowledge to interpret new cognitive situations quickly. His work suggested that establishing context could automatically reduce possible search space significantly. This approach aimed to make computer systems more efficient by mimicking human thought patterns rather than relying solely on rigid logic.

  1

• 2. Structural Mechanics And Slots

  ›

  Each frame contains information about how to use it and what to expect next when expectations are not met. Information within these structures divides into two categories: unchanging data and terminals that usually change. Terminals function similarly to variables in programming languages. Top-level frames carry facts always true about a problem while terminals hold values that might shift with new information. Different frames can share the same terminals allowing for flexible connections between concepts. Every piece of information holds inside a slot which acts as a container for specific details. These slots store facts, data values called facets, procedures known as procedural attachments, default values, and other frames or subframes. An IF-NEEDED mechanism allows deferred evaluation while an IF-ADDED rule updates linked information automatically.

  345

• 3. Evolution Of Frame Languages

  ›

  The earliest frame-based languages were custom developed for specific research projects without being packaged as reusable tools. Researchers soon realized the benefits of extracting core infrastructure to create general-purpose frame languages independent of specific applications. One of the first such languages was KRL followed by KL-ONE which became one of the most influential early systems. Robert MacGregor at the Information Sciences Institute later developed Loom as a widely used successor to KL-ONE. During the 1980s business interest in artificial intelligence fueled by expert systems led to commercial products like Intellicorp's Knowledge Engineering Environment or KEE. This tool provided a full frame language with multiple inheritance slots triggers default values and a rule engine supporting backward and forward chaining. The Semantic Web research agenda later spawned renewed interest in automatic classification leading to standards like OWL or Web Ontology Language. Stanford University released Protege as open source software built on OWL though it ceased explicit frame support after version 3.5.

  914

• 4. Comparison With Object Orientation

  ›

  Frame languages share significant overlap with object-oriented languages despite different fundamental design goals. Developers moving from academic labs to the commercial world tended to ignore philosophical issues focusing instead on specific capabilities. Both paradigms arrived at representing primary software objects in taxonomies starting with very general types progressing to more specific ones. The primary difference lay in how much encapsulation each community considered critical for success. For object-oriented programming encapsulation remained one if not the most critical requirement to manage large complex systems. Frame language creators prioritized providing vast arrays of tools to represent rules constraints and logic over strict encapsulation. Multiple inheritance served as another main differentiator since frame languages required allowing frames or classes to have two or more superclasses. Many object-oriented languages eventually preferred single inheritance to maintain modularity even though early frame languages lacked message passing until developer demands forced changes.

  810

• 5. The Neats Versus Scruffies Divide

  ›

  Early work on frames drew inspiration from psychological research dating back to the 1930s showing people use stored stereotypical knowledge to interpret new situations. Researchers like Schank and Abelson used frames to illustrate how AI systems could process common human interactions such as ordering a meal at a restaurant. These scholars were less interested in mathematical formality believing formalisms did not model average human conceptualization well. Ron Brachman represented the opposing view seeking mathematical precision through First Order Logic and Set Theory. This conflict created a classic divide known as neats versus scruffies within artificial intelligence research. Neats valued mathematical precision while scruffies focused on intuitive psychologically meaningful representations. The KL-ONE language exemplified the formal approach by spawning subsequent frame languages with automated reasoning capabilities called classifiers. Classifiers analyze declarations defining sets subsets relations and automatically deduce additional relations detecting inconsistencies within models.

  1316

• 6. Legacy In The Semantic Web

  ›

  Classification technology originally developed for frame languages became a key enabler of the Semantic Web vision. The Internet contains highly informal unstructured data making it impossible to require all systems to standardize on one data model. Automatic classification provides developers with powerful tools to bring order and consistency to inconsistent collections of information found online. The goal involves organizing web pages not just by text keywords but by classification of concepts into ontologies. Standards like OWL provide semantic layers on top of the internet allowing users to specify concepts without worrying about homonyms or synonyms crowding search results. For instance searching for OWL today retrieves pages about birds rather than the Web Ontology Language standard. The Linking Open Data community emerged from this research focusing on exposing data on the web rather than modeling. Protege remains an active ontology editing tool though current versions prioritize OWL DL over explicit frames due to expressiveness requirements.

  718