Group: information
Topic: attribute-value pairs as information
Topic: database queries, joins, and relational algebra
Topic: default value
Topic: expert systems
Topic: frame problem
Topic: fundamental concepts such as type, attributes, relationships are all the same
Topic: hierarchical structures
Topic: metaphysics and epistemology
Topic: models of reality
Topic: representing a relationship
Topic: knowledge as interrelated facts
Topic: knowledge representation by frames
Topic: relational database
Topic: science as measurement
Topic: semantic networks
Topic: skepticism about knowledge
Topic: software models of reality
Topic: thought is computational
Topic: using annotations in hypertext
Topic: using a world model in robotics
| |
Summary
Artificial intelligence requires a knowledge representation of the world. Knowledge is as important as intelligence. You need to understand the world, to understand goals and how a plan implements these goals.
Knowledge can be represented by linked representations, rows in a relational database, and conceptual dependencies. A microworld or planbox simplifies the task. The world itself can be sensed and manipulated.
Dennett argues that AI can decompose knowledge into simpler and simpler data structures. Eventually the data is simple enough for a computer to manipulate, thus avoiding an infinite regress of internal homunculi. (cbb 6/06)
Subtopic: AI and knowledge representation
Quote: a research program for AI is to hand-code a broad knowledge base, then acquire knowledge through reading, and finally learn by discovery [»lenaDB1_1991]
| Quote: artificial intelligence requires a representation of the world; relates to traditional problems of philosophy [»mccaJ_1969]
| Quote: artificial intelligence requires knowledge, lots of it [»mostJ11_1985]
| Quote: artificial intelligence represents the task environment with a symbolic structure and a systematic correspondence
| Quote: knowledge programming gives the programmer program-level control over problem-level knowledge [»abboRJ8_1987]
| Quote: the knowledge principle: the power of AI depends on the program's knowledge of its task domain instead of the program's reasoning processes [»feigEA5_1996]
| Quote: AI's data structures are primitive examples of representations that understand themselves; can conceivably replace the homunculus with an army of mechanical homunculi
| Subtopic: maintaining knowledge representation
Quote: a goal for expert-systems is for users to create and maintain the knowledge base; currently high maintenance cost [»bobrDG9_1986]
| Quote: the intelligence of a system is frozen in content and structure of its database [»mccrDL10_1984]
| Subtopic: importance of goals
Quote: if we identify the theme that a person is operating under, we can predict his pattern of goals; e.g., garabageman collects garbage because of job [»schaRC_1981]
| Quote: the key problem it determining plans is understanding the kinds of goals that people ordinarily pursue [»schaRC_1981]
| Subtopic: linked representation
Quote: any collection of knowledge about something can be modeled to any degree of detail by an appropriate plex; nothing simpler; foundation of the Computer-Aided Design System [»rossDT_1963]
| Quote: although IS-A links are widely used for classification, they have different meanings in different knowledge-representation systems [»bracRJ10_1983]
| Subtopic: AND/OR tree
Quote: an attack tree is an AND/OR threat model; OR nodes are alternative attacks, AND nodes are steps to implement the attack [»schnB_2000]
| Quote: evaluate a system's vulnerabilities by propagating leaf nodes to the attack tree's root; e.g., PGP
| Subtopic: world as its own model
Quote: for simple level intelligence, it is better to use the world as its own model [»brooRA1_1991]
| Quote: the real world is its own best model; up to date, detailed, symbolic representation irrelevant; need to sense it appropriately and often enough
| Subtopic: knowledge as description
Quote: KRL organizes knowledge by conceptual entities with associated descriptions
| Quote: the main operations for KRL are augmenting descriptions for new knowledge, matching descriptions, and searching for matching descriptions [»bobrD1_1977]
| Subtopic: knowledge as a relational database
Quote: each row of a relation represents an assertion; like a knowledge base; e.g., person A is an employee of the company with these properties [»coddEF_1990]
| Quote: if allow duplicate rows, then all users need to agree on its meaning; no precise, context-independent interpretation [»coddEF_1990]
| Quote: the sharing of data requires the sharing of its meaning; a single, simple, explicit description of every row of every relation
| Quote: the database description is like a knowledge base
| Subtopic: conceptual dependency
Quote: the heart of meaning representation is the representation of events; events at core of CD (conceptual dependency)
| Quote: in CD, events are represented independent of sentences and words; these representations use predefined schemas with rule governed slots [»schaRC_1981]
| Quote: in 'John went to New York', the CD object is also 'John' because of a rule attached to the verb 'go', i.e., 'John went John to New York' [»schaRC_1981]
| Quote: CD needs a set of meaning primitives to work in the most general way possible without duplicating information [»schaRC_1981]
| Quote: slot filling is a major role of CD descriptions, whether for parsing sentences or making inferences; unknown elements from sentence or general knowledge [»schaRC_1981]
| Subtopic: microworld
Quote: a microworld is a well-defined, limited world with interesting events and educational ideas; e.g., blocks world [»goldEP8_1982]
| Subtopic: planbox or script for behavior
Quote: a planbox is a stereotypical method of attempting a goal; it has preconditions, an action, and a goal; e.g., ASK whose action is MTRANS [»schaRC_1981]
| Quote: a script gives the appropriate behavior for a standard situation; e.g., asking a waitress for food in a restaurant script [»schaRC_1981]
| Quote: a new situation invokes a frame from memory; includes expectations and assumptions; modify as needed [»winoT_1986]
| Quote: attached to each frame is how to use it, what will happen next, what to do if expectations fail [»minsM_1981]
| Quote: every script includes roles; the actors in a story take on the roles of the script on instantiation; unfilled roles are assumed [»schaRC_1981]
| Quote: in stories without available scripts, we reconstruct the planning process of each character to determine their plans and goals [»schaRC_1981]
| Subtopic: limits of knowledge representation
Quote: skeptical about knowledge representation even though it is fundamental to cognitive science, linguistics and artificial intelligence [»winoT_1986]
| Quote: what matters is the nature of ideas; they will always be hard to develop, organize, and transmit [»nelsTH_1974]
| Quote: general AI methods failed when the specialized knowledge needed for real-world problems swamped their heuristic methods and data encodings
| Quote: early AI work did not appreciate the sheer amount and variety of specialized knowledge underlying intelligent behavior [»lindRK6_1993]
| Quote: psychology must posit internal representations, but something is a representation only for or to someone; i.e., a homunculus that leads to an infinite regress [»dennDC_1978a]
| Quote: current knowledge representation systems are a house of cards built on a low-level foundation with many middle layers [»bobrD1_1977]
|
Related Topics
Group: information (46 topics, 1160 quotes)
Topic: attribute-value pairs as information (57 items)
Topic: database queries, joins, and relational algebra (34 items)
Topic: default value (8 items)
Topic: expert systems (8 items)
Topic: frame problem (13 items)
Topic: fundamental concepts such as type, attributes, relationships are all the same (37 items)
Topic: hierarchical structures (46 items)
Topic: metaphysics and epistemology (99 items)
Topic: models of reality (33 items)
Topic: representing a relationship (28 items)
Topic: knowledge as interrelated facts (23 items)
Topic: knowledge representation by frames (18 items)
Topic: relational database (35 items)
Topic: science as measurement (36 items)
Topic: semantic networks (42 items)
Topic: skepticism about knowledge (34 items)
Topic: software models of reality (24 items)
Topic: thought is computational (60 items)
Topic: using annotations in hypertext (13 items)
Topic: using a world model in robotics (12 items)
|