KNOWLEDGE

UNIVERSITI MALAYSIA SABAH

  • FACULTY OF COMPUTING & INFORMATICS
    • Course: KT24202 Artificial Intelligence Knowledge Representation & Logic & RL in AI
    • Instructor: Dr. Mohammed Ahmed
    • Date: 2026-04-12

LECTURE AGENDA

  • Knowledge Representation & Logic
  • Reinforcement Learning
  • Faculty of Computing & Informatics
  • Universiti Malaysia Sabah
  • Lecture Series 2026

THINKING QUESTIONS

  • How does a robot know what an “object” is without ever seeing one before?
  • If I say “bird,” what exactly should an AI understand about it?
  • Can a machine “understand” knowledge, or does it only store data?
  • Why can humans learn from mistakes, but some AI systems struggle to do so?

SECTION 01: FOUNDATIONS

What is Knowledge Representation?

  • Core Definition:

    • Knowledge Representation (KR) is a fundamental AI field dedicated to structuring and formatting information about the world in a manner that machines can comprehend and process.

  • The Knowledge Base:

    • Structured information resides here, serving as the central repository for an agent's understanding.

  • Functional Capabilities:

    • Reasoning:
    • Drawing logical conclusions from known facts.
    • Decision-Making:
    • Evaluating options based on contextual knowledge.
    • Problem Solving:
    • Actively resolving complex, real-world tasks.

  • Importance of KR:

    • Without a KR framework, AI systems lack the contextual understanding necessary to interpret their environments.
    • Example: "It is raining." How would you format this so a computer knows to bring an umbrella?

DATA, INFORMATION, AND KNOWLEDGE

Transformation Process

  • STAGE 1: Data

    • Raw, unorganized facts and figures lacking inherent meaning or context.
    • Example: "12012012"

  • STAGE 2: Information

    • Contextualized, categorized, and condensed data that becomes meaningful.
    • Answers: Who, What, When, Where

  • STAGE 3: Knowledge

    • Perceptions, skills, and experience applied to information to achieve goals.
    • Answers: How

  • In AI, Knowledge Representation focuses on structuring this "Knowledge" tier so machines can reason and solve problems.

TYPES OF KNOWLEDGE

Categories of Knowledge

  • Declarative Knowledge:

    • Knowing what (Facts & Assertions)
    • Includes object relationship and value (e.g., Rises in)

  • Procedural Knowledge:

    • Knowing How (Rules & Methods)

  • Structural Knowledge:

    • Knowing Why (Hierarchies & Taxonomies)

Examples

  • Hierarchical Relationships:
    • Class and Sub-class examples:
    • For a computer system:
      • [Computer System]
      • Class → Sub-Class relationships such as
      • [The Sun]
      • [Hardware]
      • [Software]
    • Semantic Networks and implicit propositions.

EXPLICIT VS. TACIT KNOWLEDGE

Definitions and Differences

  • Explicit Knowledge:

    • CODIFIED & OBJECTIVE
    • Easily recorded, articulated, and shared.
    • Stored in manuals, databases, and policies.
    • Readily processed by LLMs and NLP systems.

  • Tacit Knowledge:

    • INTUITIVE & SUBJECTIVE
    • Internal traits like individual skills and expertise.
    • Deeply ingrained and difficult to codify or transfer, it includes mental models and "gut feelings."

Formalization Challenge

  • Tacit knowledge is captured through problem-solving sessions with experts, translating intuitive skills into explicit rules for AI Knowledge Bases.
  • Methods include mentorship and storytelling.

KNOWLEDGE ACQUISITION PROCESS

Steps for Gathering Knowledge

  1. Literature Review:
    • Analyzing academic papers, reports, and documented sources to extract foundational facts and domain logic.
  2. Expert Interviews:
    • Engaging with subject matter experts to elicit nuanced insights, decision paths, and tacit experiences.
  3. Direct Observation:
    • Studying real-world scenarios and operational interactions to understand practical knowledge application.
  • Outcome:
    • The acquired knowledge is translated into structured formats (Rules, Frames, or Logic) that are machine-readable and computationally processable.

APPROACHES TO KNOWLEDGE REPRESENTATION

Methodologies

  1. Relational Knowledge:
    • Facts + relations (like a DB).
    • Example: Parent(Anna, Bob) → Anna is Bob’s parent.
  2. Inheritable Knowledge:
    • Hierarchies + inheritance.
    • Example: Bird → Fly; Penguin → Bird (but overrides Fly).
  3. Inferential Knowledge:
    • Logical: Likes(x, pizza) ∧ Serves(restaurant, pizza) → Will Visit(x, restaurant).
  4. Procedural Knowledge:
    • Condition-action rules.
    • Example: IF overdue(book) THEN send_reminder(email).

KNOWLEDGE GOAL AND TYPES

Representation Needs

  • Goal: Common sense reasoning
    • Need to represent knowledge about the world including:
    • Objects
    • Events
    • Procedures
    • Relations
    • Mental states
    • Meta knowledge

PROPERTIES OF REPRESENTATION SYSTEMS

Key Attributes

  • Representational Adequacy:
    • Ability to represent the required knowledge.
  • Inferential Adequacy:
    • Ability to manipulate knowledge and produce new knowledge.
  • Inferential Efficiency:
    • Ability to direct inference methods productively and use limited resources efficiently (e.g., time, storage).
  • Acquisitional Efficiency:
    • Ability to acquire new knowledge, ideally automatically.

CATEGORIES AND OBJECTS

Understanding Categories

  • Specific objects (e.g., my basketball BB9) and general categories (e.g., Basketballs).
  • Categories as Relationships:
    • Basketballs(BB9) = True
  • Reification of predicates:
    • Basketballs → use in other predicates
    • Member(BB9, Basketballs) = True
    • Abbreviated to BB9 ∈ Basketballs
  • Subcategories for Instance:
    • Subset(Basketballs, Ball) = True
    • Abbreviated as Basketballs ⊂ Ball
  • Taxonomy: System of categories and subcategories.

BASIC RELATIONS FOR CATEGORIES

  • Disjoint Relationships:

    • Disjoint({Animals, Vegetables})

  • Exhaustive Decomposition:

    • ExhaustiveDecomposition({Americans, Canadians, Mexicans}, NorthAmericans)

  • Partitioning:

    • Partition({Males, Females}, Animals)

  • These properties can be defined using first-order logic.

PHYSICAL COMPOSITION

Basics

  • Basic relations such as PartOf

    • PartOf(Bucharest, Romania)
    • PartOf(Romania, EasternEurope)
    • PartOf(EasternEurope, Europe)
    • PartOf(Europe, Earth)

  • Can be used to define composite objects:

    • Biped(a) ⇒ ∃l1, l2, b
    • Leg(l1) ∧ Leg(l2) ∧ Body(b)
    • ∧ PartOf(l1, a) ∧ PartOf(l2, a)
    • ∧ PartOf(b, a)
    • ∧ Attached(l1, b) ∧ Attached(l2, b)
    • ∧ l1 ≠ l2
    • ∧ [∀l3Leg(l3) ∧ PartOf(l3, a) ⇒ (l3 = l1 ∨ l3 = l2)]

PROTOTYPES

Understanding Prototypes

  • Natural categories are hard to define; there is no set of features applicable to all instances.
  • Prototypes have typical properties.
    • Example: Select typical members of categories
    • ∃b ∈ Typical(Bird) ⇒ CanFly(b)

HIERARCHIES AND INHERITANCE

Taxonomies

  • Hierarchies are a natural way to structure categories.
  • Groups of things that share properties in the world do not require definitions to be repeated.
  • Example:
    • Saying “elephants are mammals” conveys extensive information.
  • Questions about inheritance:
    • “Does A inherit from B?” is the same as “Is B in the transitive closure of IS-A (or subsumption) from A?”

GRAPHICAL REPRESENTATION OF INHERITANCE

Reasoning with Inheritance

  • IS Relations:
    • Clyde ↓ Elephant (category) ↓ Gray (property)
    • Clyde is an Elephant, Elephant is Gray.
  • Transitive Relations:
    • Transitive closure on all paths rather than just direct relationships.
    • Example:
    • Clyde is an Elephant, Elephant is Gray ⇒ Clyde is Gray

STRICT INHERITANCE

Conclusions and Paths

  • Conclusions produced through complete transitive closure on all paths (any traversal procedure will do).
  • All reachable nodes imply associated properties.

LATTICE STRUCTURE WITH STRICT INHERITANCE

Multiple Inheritance

  • Multiple AND (∧) parents (i.e., multiple inheritance).
  • Trees: All conclusions reachable by any paths are supported.

DEFEASIBLE INHERITANCE

Property Overriding

  • Inherited properties can be overridden or defeated.
  • Conclusions determined by upward search from focus node, selecting the first version of property desired.

SHORTEST PATH HEURISTIC

Path Navigation

  • Links have polarity (positive or negative).
  • Utilization of shortest path heuristic to determine which polarity matters in conclusions.
  • Some paths may be preempted, while others are considered admissible.

PROBLEM: AMBIGUITY

Handling Ambiguity

  • Lack of a single shortest path necessitates explicit handling of ambiguous reasoning chains.
  • Distinguish between ambiguous and unambiguous reasoning chains, preferring some extensions over others (default logic).
  • Difference between credulous vs. skeptical reasoning.

ONTOLOGIES

Knowledge Organization

  • Categories include:

    • Anything
    • Abstract Objects
    • Generalized Events
    • Sets
    • Numbers
    • Physical Objects
    • Processes
    • Places
    • Animals
    • Humans

  • Organization of knowledge within a single taxonomy.

FRAMES

Definition and Usage

  • A frame is a collection of attributes or slots and associated values that describe a real-world entity.
  • Each frame represents either a class or an instance (an element of a class).

EXAMPLE FRAME

  • Lecture Example:
    • Specialization of: meeting
    • Context: large number of students
    • Course: Op. Systems
    • Level: Difficult
    • Actions triggered by attributes such as intolerance and behavior expectations for students.

KNOWLEDGE DISCOVERY

Information Retrieval

  • When facing new situations, information is stored in frames with slots.
  • Some slots trigger actions leading to new situations.
  • Frames serve as templates needing to be filled in.
  • Filling them allows agents to undertake actions and retrieve other frames.
  • Frames extend traditional record datatypes in databases, aligning closely with object-oriented processing.

FLEXIBILITY IN FRAMES

Slot Functionality

  • Slots in a frame can contain:
    • Information for selecting a frame in a situation.
    • Relationships to other frames.
    • Procedures to execute after slot completion.
    • Default information for missing input.
    • Blank slots that can remain unfilled unless required for a task.
    • Other frames to maintain hierarchy.

EVENTS

Defining Events

  • Events describe truths contingent on time.
  • Example for a flying event:
    • E ∈ Flyings
    • Flyer(E, Shankar)
    • Origin(E, SanFrancisco)
    • Destination(E, Baltimore)
  • Events may or may not be ongoing during a specific time frame, with facts true at specific points called fluents.
  • Example: At(Shankar, Baltimore)

PREDICATES OF EVENTS

  • Truth at Time: T(f , t)
  • Event Occurrence: Happens(e,i)
  • Initiates: Initiates(e, f , t)
  • Terminates: Terminates(e, f , t)
  • Fluents: Clipped(e,f , i), Restored(e,f , i)

TIME INTERVALS

Benefits of Representation

  • Time can be represented in intervals or moments of zero duration.
  • Allows definitions such as:
    • End(i1) = Start(I2)
    • Preceding intervals, subsets, overlaps, beginning, end, identity of time intervals.

SCRIPTS

Script Definition

  • A structured representation describing a stereotyped sequence of events in a context.
  • Scripts help organize events in knowledge bases and relate closely to frames.

COMPONENTS OF A SCRIPT

  • Entry Conditions: must be true for script activation.
  • Results: facts true upon script completion.
  • Props: Items within the script’s context.
  • Roles: Actions participants perform.
  • Scenes: Temporal aspects of the script.

CANONICAL EXAMPLE: RESTAURANT VISIT

  • Objects: tables, menu, food, check, money
  • Roles: customer, waiter, cook, cashier, owner
  • Entry Conditions:
    • Customer is hungry.
    • Customer has money.
  • Results:
    • Customer is no longer hungry.
    • Customer has less money; owner gains more.

SCRIPT ACTIONS

Symbols and Meanings

SymbolMeaningExample
ATRANStransfer a relationshipgive
PTRANStransfer location of objectgo
PROPELapply physical forcepush
MOVEmove body partkick
GRASPgrab an objectgrab
INGESTconsume an objecteat
EXPELexpel from bodycry
MBUILDmentally create infodecide
CONCconceptualizethink
SPEAKproduce soundsay
ATTENDfocus sense organlisten

DETAILED SCRIPT EXAMPLE

Restaurant Interaction

  1. Scene 1: Entering
    • Customer enters restaurant and looks at tables.
    • Decides where to sit.
  2. Scene 2: Ordering
    • Waiter brings menu.
  3. Scene 3: Eating
    • Picks food from the menu and consumes it.
  4. Scene 4: Exiting
    • Pays the check and leaves the restaurant.

CYC

Overview

  • Goal: Codify millions of knowledge pieces for common sense.
  • Origin:
    • Started in 1984 by Microelectronics and Computer Technology Corporation.
    • 1986: Estimated completion required 250,000 rules and 350 man-years.
  • Spinoff:
    • In 1994, spun off into Cycorp, Inc.
    • 2008: Links to Wikipedia added.
    • 2012: Publicly available through OpenCyc.

Structure

  1. Basic Facts:
    • Examples include “Every tree is a plant” and inference like “Trees die eventually.”
  2. CycL Language:
    • Utilizes predicate calculus similar to Lisp.
  3. Current Efforts:
    • Connection to natural language processing.

CYC ONTOLOGY

Levels of Knowledge

  1. Upper Level:
    • Contains broad abstract concepts and universal truths.
    • Smallest but widely referenced.
  2. Middle Level:
    • Not universal but widely used abstraction.
    • E.g., human interaction, geospatial relationships.
  3. Lower Level:
    • Specific knowledge areas like chemistry and biology.

Specific Examples

  • Upper Level:
    • (isa Event Collection)
    • (genls Event Situation)
  • Middle Level:
    • DisjointWith(SocialGathering, SingleDoerAction)
  • Lower Level:
    • Various facts about chemical reactions and their classifications.

End of Document