Entity that perceives, reasons and acts within some external environment (e.g., intelligent agent).

Alpha-Beta pruning

Method for improving the efficiency of minimax search used for game playing systems.


Able to function without external control or intervention (e.g., au­tonomous agent; autonomous mobile robot).


A learning algorithm used for neural networks.


Returning to a previous choice point in a search, in order to explore other alternatives.

Backward chaining

Method used in problem solving which involves starting with a goal or hypothesis and working backwards using rules to find what facts are necessary to prove the goal.

Bayes’ theorem

Rule for calculating the probability of a hypothesis given some evidence, based on other available probabilities.

Best first

Search strategy which uses heuristics to guide search, looking at the most promising nodes first.

Breadth first

Search strategy that involves exploring all nodes in a tree at a given depth (from root) before exploring nodes further down in tree.

Brute force

Search strategy not using any intelligence or heuristics.

Case-specific data

Data specific to a particular problem or case (e.g., data on a particular patient) in an expert system.

Certainty factors

Measure of the likelihood that a fact or conclusion is true, often used in rule-based expert systems.

Chinese room

Thought experiment proposed by Searle to demonstrate that a sys­tem could behave intelligently without being intelligent.


Group of items with similar characteristics.


Assigning an object to a particular category or class based on its features.

Combinatorial explosion

Term used to indicate a problem with exponential com­plexity, where increasing the size of a problem by a small amount causes an “explosion” in the number of possibilities to be considered when looking for a solution.

Multiple inheritance

Inheritance from several different sources (when a class may have multiple parents).

Natural language

A human language (like English) rather than a computer lan­guage.

Parse tree

Tree representation of syntactic structure of sentence.

Pattern recognition

Class of methods for identifying the category to which an object belongs based on its (often visual) pattern (e.g., identifying letters in handwriting from an image).


Simple neural network.


Basic unit of speech; part of word (e.g., “t”).


Point-like, basic element of a digital image.


Stage of language analysis that takes account of the context in which things are said.

Predicate logic

A logic widely used in Al for representing knowledge.


Representing how something should be done (procedures) rather than what is true (cf. Declarative).

Production rule

Term used for IF—THEN rules in rule-based expert systems. (Meaning of term slightly different in other areas of Computer Science.)

Proof theory

Theory stating what inferences are valid in a logic.

Reason maintenance

Recording the justifications for conclusions, so when a fact is withdrawn all the facts that are derived from it are also withdrawn.

Representational adequacy

Ability to represent complex facts.


A rule of inference and simple proof procedure based on that rule of inference. Used in the Prolog programming language.


System able to manipulate physical objects in the world, usually with the aid of sensors.

Rule-based system

Expert system based on using IF—THEN rules for represent­ing knowledge.

Search space

Set of all possible nodes to be considered in a given search problem (usually those reachable from some start node).

Search strategy

Strategy for controlling the search (of a graph or tree) for some target node or state.

Search tree

Tree representation of search space, showing how possible solutions may be reached from some initial state.

Semantic network

Knowledge representation scheme based on networks of nodes and links, normally representing objects and relationships between objects.


The meaning of a statement (whether a natural language sentence, statement in a programming language, or statement in a logic). Also used to refer to the stage of natural language understanding concerned with deriv­ing sentence meaning.

Speech act

Action performed by spoken utterance (e.g., command).

Speech recognition

The recognition of word sequences from a speech signal.

Speech understanding

Determining sentence meaning from a speech signal.

State space

Set of problem states reachable, given a problem to be solved using state space search.

State space search

Solving a problem by searching through the possible problem states that may be reached from some initial state.


Representation of knowledge such that there are no meaningful “symbol structures” (e.g., neural networks).

Symbol structures

Data structures composed of symbols denoting objects or con­cepts. Most Al knowledge is represented in this way.


Legal organizations of constituents in a language. For example, the syntax of English defines legal combinations of words in a sentence. Also used to refer to stage of natural language understanding concerned with ascertaining the structure of a sentence.

Turing test

Test proposed to check for human-like intelligence by comparing a computer program’s ability to answer arbitrary questions with that of a human.

Version space learning

Inductive learning method based on a particular algo­rithm for searching a space of hypotheses.

Working memory

Part of expert system used to represent facts that are currently believed true about the problem being worked on.