A Glossary of Applied Artificial Intelligence Terms

Digital Equi C : A Glossary of Applied Artificial Intelligence Terms Preface This glossary is designed to acquaint you with terms that are found in concepts of artificial intelligence technology —including terms that are specific to Digital, and to provide a perspective on how these techniques can add to your ability to solve complex problems. Most of the definitions in this glossary have been adapted from the following sources Clayton, Bruce D. ART Programming Tutorial. (Los Angeles, CA:) Inference Corporation, 1985 Kinnucan, Paul. “Computers That Think Like Experts.” High Technology. (January 1984): p. 36 Scown, Susan J. The Artifictal Intelligence Experience: An Introduction. (Maynard, MA:) Digital Equipment Corporation, 1985 The definitions contained in this glossary are provided to assist individuals unfamiliar with the terms of the subject matter, and are not intended to create any obligations on the part of Digital Equipment Corporation. The Digital logo, Microvax,vax,vax-11,vax Calling Standard, vax Lisp and vMs are trademarks of Digital Equipment Corporation. GOLDEN COMMON LisP and GCLIsP are trademarks of Gold Hill Computers. UNIX is a trademark of at&t Bell Laboratories. ant and Inference are trade- marks of Inference Corporation ae Digital Equipment Corporation A Glossary of Applied Artificial Intelligence Terms absolute rule An absolute rule is a purely deduc- tive and assertive rule that is used to evaluate a situation. For instance, an absolute rule might detect that the diameter of a circle has become known in the database, and as a result, act to calculate the area and circumference of the circle. algorithm Anexplicit, finite set of instruc- tions that is guaranteed to find a solution to a particular problem, although not necessarily by the best or fastest route. Most conven- tional programming tools were designed to develop applications that use explicit instructions. By definition, however, this approach assumes that the path by which a problem can be solved can be determined in advance. See explor- alory programming. ART™ The automated reasoning tool, an expert system software develop- ment environment from Inference ~ Corporation. art provides knowl- edge engineers with a comprehen- sive set of knowledge representation and storage techniques and graphics capabilities for building expert systems. artificial intelligence Broadly speaking, a growing set of computer problem-solving tech- niques that are being developed to imitate human thought or decision making processes, or to produce the same results as those processes. An effort to create machines and computer programs that have AO A i Chi 41 47 4 LOTTA tatatatatata some of the characteristics of intel- ligence typically associated only with human beings. These include the ability to understand language, reason, learn, and solve complex problems. backtracking An element of a search process that involves returning the database or conditions in a system to a previous state, in order to try an alternative solution path. backward chaining A wpe of system activity that attempts to solve a problem by stating a goal and looking into the database for the conditions that would cause that goal to come about, then reiterating this process, using those conditions as the goals and searching for their precondi- tions, etc. For example: in an expert sys- tem that uses backward chaining, it is possible to solve problems such as ‘How can I end the fiscal year at 10% under budget?” In this case, the program begins with a goal— being 10% under budget—and works backward to a set of ante- cedent conditions which would bring that goal about. Backward chaining also pro- vide. the foundation for the expla- nation facility that is used in many expert systems to retrace a solution by recalling the rules that were used in the reasoning process. Because it is used to build a strategy for reaching a known goal, an expert system that employs backward chaining techniques ts commonly called a goal-driven system. blackboard A structured workspace on which a system can post information about the internal states of objects or system registers, for consulta- tion and appropriate action by operators in the system. A black- board is the mechanism by which multiple knowledge sources in the system can communicate. compatibility The capability of using an instruc- tion, program, or component on more than one computer with the same result. constraint rule A type of rule that applies limits to a search, by specifying that its associated pattern is never allowed to occur in a valid solution. Con- straint rules provide a means to reduce the number of unproduc- tive searches. Common LISP An implementation of the Lise programming language that incor- porates features that are common to several implementations of Lisp. See LISP. deduction The process of reaching a conclu- sion by logical means. domain The problem area about which a system has knowledge EMACS An editing tool developed at the Massachusetts Institute of Tech- nology (mir), well known in the UNIX™ environment that enables programmers to perform many editing operations with only a sin- gle keystroke. expert system A computer system that embodies the specialized knowledge of one or more human experts and uses that knowledge to solve problems. By capturing in software the best knowledge and judgment availa- ble, it is possible to distribute expertise on a wider scale An expert system generally con- sists of a knowledge base and an inference engine, both of which are continually modified and evalu- ated, It may also include a natural language interface that facilitates user communication with the sys- tem, an explanation facility, and a knowledge acquisition subsystem that is used to enhance the knowl- edge base. A major strength of an expert system is that it can take the best insights of several human experts and apply them to the same prob- Jem simultaneously. When human experts see the mistakes that an evolving expert system is making, they can determine what knowl- edge the system is lacking and con- tinue to enhance it. See inference engine, knowledge base explanation facility A feature of many expert systems that tells what steps were involved in the process by which the system arrived ata solution. These facili- ties can be simple traces of steps, or they can be more complex, supply- ing encoded reasons why the solu- tion uses one alternative rather than another. See expert system exploratory programming A set of techniques developed by the artificial intelligence commu- nity to deal with problems for which the design of a solution can- not be known in advance. A major premise of this type of program- ming is that, by exploring various techniques and attempting to rap- idly prototype a solution to some small subset of the problem, it can be better understood. Supporting techniques include interactive edit- ing and debugging, integrated pro- gramming environments, and graphics-oriented user interfaces. forward chaining A type of system activity that applies Opcrators to a Current state in order to produce a new state, and so on, until the solution is reached. In an expert system, a forward-chaining rule detects cer- tain facts in the database and takes an action because of them. For example: in a system that uses forward chaining, it is possible to solve problems such as “What will be the cost of new office furni- ture if hire three more people?” In this case, the program synthe- sizes an answer from pieces of knowledge, An expert system that employs forward-chaining techniques is also called a data-driven system. frame A knowledge representation tech- nique based on the idea of a frame of reference. A frame carries with it asct of slots which can represent objects that are normally associated with the subject of the frame. The slots can then point to other slots or frames, a feature that gives frame- based systems the ability to allow one object to inherit characteristics from another, and to support inferences. framework system A type of artificial intelligence systems-building tool designed to reduce the amount of time required to develop an expert system. A framework system includes built-in knowledge representation and reasoning techniques, and may also include editors, translators, and debugging tools to simplify the coding of expert knowledge in a form that the computer can use. A knowledge engineer customizes a framework system for a specific application by building a knowl- edge base for the problem domain of interest. G GCLISP~ GOLDEN COMMON LIsp. A microcomputer-based artificial intelligence tutorial and develop- ment software system. goal A condition or set of conditions to which a valid solution must conform. ll heuristic A process, sometimes a rule of thumb, that may help in the solu- tion of a problem, but that does not guarantee the best solution, or indeed, any solution. Because the success of a heuristic is not guaran- teed, a problem that can be solved by one algorithm frequently requires many heuristics. The pri- mary effect of heuristics is to elimi- nate the need to examine every possible approach. See algorithm. I icon A symbol to which a computer user can point an interface device in order to select a function, such as “move window.” image processing The examination by a computer of digitized data about a scene and the features in it in order to extract information. inference A conclusion based on a premise. inference engine The part of a rule-based system that selects and executes rules. In contrast to algorithms embedded in traditional software programs, but like the human reasoning process, the conclusion that an inference engine will draw from a given set of facts is not known in advance. Interlisp A general-purpose environment for building and using artificial intelligence applications based on the Lisp programming language. Interlisp for vAx™ is an implemen- tation of Interlisp developed by the Information Sciences Institute of the University of Southern Califor- nia specifically for use on the vAx family of computers. integration 1 A software design concept that allows users to move easily between application programs, or to incorporate data from one pro- gram into another, such as moving data displayed in a graphics pro- gram into a text document. 2 A computing approach that allows an organization to match its communications and information needs across organizational levels to specific products (systems) through the use of common system and information architectures. intelligent system A system equipped with a know!- edge base that can be manipulated in order to make inferences. The distinction between a sys- tem that can perform intelligent operations and one that merely manipulates data can be character- ized ina highly simplified example. In atypical street address, 129 First Street, the house number (129) is a piece of numeric dafa—the type of data on which conventional systems can perform standard numeric computations The notion that 129 First Street is an address is a piece of énforma- tion. Many systems can support the use of information through such activities as data and database management, using traditional pro- grams written in such languages as COBOL and PASCAL. Understanding that First and Ist are equivalent ways to represent the ordinal number | ina series, and that all ordinal numbers have comparable ways of expressing the same notation is a relevant piece of knowledge—one that most auto- | mated development tools would find difficult to represent and work with. But such understanding is essential to any base of knowledge that could, for example, help a publishing company prevent the same magazine from being sent twice to the same subscriber: to John James at 129 First Street, Any- town, USA, and to John James at 129 Ist Street, Anytown, USA. K knowledge base The part of an artificial intelligence system that contains structured, codified knowledge and heuristics used to solve problems. Artificial intelligence systems using such a base are called knowledge-based systems. In an expert system, the knowledge base generally contains a model of the problem, know!- edge about the behavior and inter- actions of objects in the problem domain, and a level of general- purpose knowledge. knowledge engineer A person who implements an expert system. A knowledge engi- neer interviews experts, to obtain the raw knowledge from which to structure the knowledge base and formulate the rule base, and pro- grams raw knowledge into a form that the computer can understand. knowledge representation A structure in which knowledge can be stored in a way that allows the system to understand the rela- tionships among pieces of knowl- edge and to manipulate those relationships The primary methods used to represent knowledge in expert systems are: procedural representation, which combines a number of items to form a solution. From all possible combinations of system options, for example, xcon, Digital’s configura- tion system for vAx and ppp-11 systems, selects and combines the appropriate components to meet a customer's system configuration requirements. rule-based representation, a two- part representation that specifies both a pattern and an action to be taken when real-world data matches that pattern. Complex patterns may be structured by link- ing clauses together with connec- tives such as AND and or. For example: a typical rule might be, “IF the patient’s temperature is greater than 100 degrees AND the patient has a runny nose, THEN conclude that the patient has a cold.” frame or schema representation, in which objects are represented by “frames” that define the object in terms of its relationship to other objects. For example: the standard properties of a mouse might include its biological parts, color, and habitat. A mouse can also be defined in terms of its relationship to other objects: perhaps as natural prey to a cat. A frame-based system is essen- tially a semantic network (see defi- nition below) in which objects are represented using frames rather than basic symbols. Frame-based systems can store an immense amount of knowledge about prop- erties and relationships concisely. semantic networks, which repre- sent abstract relationships among objects in the system's knowledge domain. Objects are linked together by the relationships between them. A typical relation- ship might be expressed as: golden retriever Is A dog. Dog 15 A mammal, where Is A represents the link. first-order logic, which formally represents specific propositions and the relationships between them. Rules of logic can be applied to these representations to derive any fact that follows logically from the propositions they represent. For example: on the basis of the two propositions ALL MEN ARE MORTAL and ARISTOTLE IS A MAN, a first-order logic system can infer that ARISTOTLE [S MORTAL. A first- order logic system supplies a means for explicitly representing virtually any type of knowledge. IL LISP A programming language (List Processing) designed specifically to manipulate symbols rather than numeric data. A Lisp data element is a list of symbols that may repre- sent any object, including its own list processing functions. A LisP program essentially consists of col- lections of independent proce- dures called functions. See symbolic processing. LISP Interpreter A part of many Lisp-based solt- ware tools that allows specitic list- processing operations such as match, join, and substitute, to exe- cute on a general-purpose com- puter rather than a special-purpose Lisp machine. LISP machine A single-user workstation with an architecture dedicated to the effi- cient writing and execution of applications using the Lisp pro- gramming language list Ina list processing software lan- guage, an ordered sequence of ele- ments, usually found within a pair of matching parentheses: for exam- ple: (A B C) is a list composed of the elements A, B, and C. In the LISP programming language, the first clement in the list is com- monly a program function and the others are arguments to be acted upon by that function. M macro function A uisp function which serves as a template for translating a Lisp form (language structure). When a macro is called, a new form is substituted for it and evaluated in place of the macro call. MicroVAX™ A family of 16- and 32-bit super microcomputers based on a single- chip implementation of the vax architecture. natural language A person’s native tongue. Natural language systems attempt to make computers capable of processing language the way people normally speak it instead of in specialized programming languages, thereby making it easier and more efficient for both inexperienced and sophis- ticated users to work with comput- ers. Natural language systems are particularly well suited for environ- ments that include many non- technical users or users who do not spend much time working with computers; for database inquiry systems; and for computer-assisted instruction systems. Today, most natural language systems are implemented in English O object-oriented programming Programming that focuses on indi- vidual program units (objects) consisting of instructions and data, rather than on procedures. OPSS5 for VAX An advanced programming lan- guage designed to facilitate the use of production rules. ops) for VAX Is a high-performance implementa- tion of the ors5 programming language. pattern The description of something for which a system should search, either in a knowledge base or a rule base. a re pattern matching A process performed by an expert system during a search through its knowledge base. The objective of the search is to match real world data—such as questions, problem statements, etc., against knowledge stored in the knowledge base predicate A predicate is a function that returns a truth value. Predicates are used to select among condi- tional alternatives. pretty-printing The style of printing implemented by a specific Lisp function which arranges LisP forms on indented lines to make them easier to read. production rule A procedural response triggered by a pattern. Rules are commonly structured in an if...then...format (1F the pattern is matched, THEN schedule a procedure for execu- tion.) When the condition is met, the rule causes the s an assertion (the patient has a fever) to its knowledge base. In practice, a rule which has been activated by a pattern match may be in competition with other activated rules. The stem to add stem’s infer- ence engine decides which of the activated rules should be executed, and in what order. Production rules simplify the generation of prompts (rules can easily be turned into questions) and the return of explanations to a user (rules can be modified to produce answers to questions). production system A computer program consisting entirely of if-then statements called productions. Production systems maintain two databases called working memory and production memory. Working memory con- tains a model of the current state of the problem and production mem- ory stores the productions. Pro- ductions are structured such that if aset of conditions about working memory are true simultaneously, some specified set of actions should be executed. Production system languages are non-procedural -that ts, the order of the rules in the program does not affect its operation. Because the programmer never needs to know the order of execution, the rules in a production system can be located for ease of maintenance. PROLOG A programming language (PRO- gramming in LoGic) designed pri- marily to manipulate symbols rather than numeric data. PROLOG differs from Lisp, another symbolic processing language, mainly in approach: PROLOG programs use assertions about objects and rela- tionships to handle queries about them. The program answers inquir- ies by consulting its knowledge base of relations. PROLOG is well suited as a base for relational data- base and natural language systems, and for applications that require simultaneous execution of differ- ent parts of the program. real-time Taking place during the actual occurrence of an event. Real-time refers to computer systems or pro- grams that perform a computation during the actual time that a related physical process transpires, in order that the results of the com- putation can be recorded or used to guide the physical process. Examples of real-time systems include use of computers to guide airplane landings or to monitor assembly line processes. search The process of trying different actions in a system until a sequence of actions is discovered that will achieve a goal state. symbol A isp data object used to name a variable, a functional definition, or a Lisp object with properties. symbolic processing A type of processing that primarily uses symbols rather than numeric representations of data. Although all computers process symbols, the symbols of traditional software programs primarily repre- sent numbers and numeric func- tions. In expert systems, symbols are not restricted to a numeric con- text, but may represent objects, concepts, processes, etc. Expert systems reason by processing these symbols. Artificial intelligence systems are used to simulate intelligent human behavior and reasoning. Since peo- ple do not think in numbers but symbolically, symbolic processing capabilities more readily meet the needs of these systems. Text manipulation on a word processor is one example of how symbolic processing works. Text manipulation is a function of mov- ing words, sentences, and para- graphs around, changing case, and highlighting, essentially without regard to the content of what is moved. VAX-11" Family Virtual Addressing EXtended 11. A Digital family of high performance, multi-user, multi-programming computer systems. A vAX com- puter combines a 32-bit architec- ture, MASSBUS clectrical cable, efficient memory management, and a virtual memory operating system, VAX/VMS. VAX/VMS VAX/VMS or VMs is the general- purpose, Virtual Memory Operat- ing System for the vax-11 family of systems. It provides demand pag- ing, working set management, and swapping as virtual memory ser- vices. VAX/VMs includes a full file management system (FMS) with record management uulities (RMS) VAX Calling Standard A standardized subroutine proto- col used across the vax-11 family Any software that supports the vax Calling Standard can call out to or be called by subroutines writ- ten in other languages, system ser- vices, and data management software. The ability to communi- cate with other software makes it possible to use languages and tools that are most efficient for an appli- cation of interest and link that application to others. VAX LISP An implementation of Common Lisp, a dialect of Lisp that runs on the vax family of computers using the vMs operating system. virtual memory A programming method that allows the operating system to pro- vide essentially unlimited program address space. In a vax-11 com- puter, the virtual memory design means that a vAx-11 program can address over 4 gigabytes (four bil- lion bytes) of address space W window An application software design concept that 1 allows several pro- grams to be run and displayed on the screen simultaneously, and 2 supports integration of data between application programs, e.g., spread sheet data displayed in one window may be included or merged with data stored in a word processing window. Use of multi- ple windows in a development environment permits system devel- opers to monitor multiple proc- esses or system states without the need to exit from one module to observe another. working memory The dynamic portion of a produc- tion system's memory. Working memory contains the database of the system, which changes as rules are executed. See production Syslem., workstation A system that 1 provides users with integrated, profession-specific functions, delivered through a single human interface (menu), 2 allows users to share work with other members of an organization by participating in a network of integrated functions located else- where in the organization, 3 has, at minimum, integrated graphics and word processing, plus terminal or file transfer communications, and 4 provides artificial intelli- gence developers with a single-user workstation. v paaisoy Saytry [[¥ UoresodioD wourdinby [eusiql CRI IYsAdoD O'¢z FI LO S8-S689I-dV VSN Ul pau