Block II, Unit I, Fundamentals of Symbolic AI

Block II, Unit I, Fundamentals of Symbolic AI

The starting point is: expressions that both humans and computers can reason about

This idea led to PSSH Symbolic representation Symbolic reasoning

1


This unit has nine sections Introduction What is intelligence Examples of intelligent behavior The physical symbol system hypothesis Representation and reasoning Developing representations Examples of symbol representations Representation Reasoning

2


What is intelligence Are intelligent actions only ones that are learnt or

could an instinctive? Are intelligent tasks only the ones that can only be

done competently by a subset of ‘intelligent’ people, or do tasks that almost anyone can manage, such as speaking, planning or manipulating objects, require intelligence too?

Is intelligence in some way related to expertise?

3


What is intelligence

Are intelligent actions only ones that are learnt or could an instinctive?

AI has traditionally tended to focus on a number of behaviors that have come to be labelled as, or are assumed to be, ‘intelligent’. These include:

natural language processing: the creation of machines capable of understanding and responding to human languages such as English;

expert problem solving: for instance, diagnosing medical conditions from complex sets of symptoms;

planning and scheduling tasks such as airline scheduling or planning the layout of a factory floor;

logical reasoning, often under conditions of uncertainty.

Intelligence comes from knowledge and the use of knowledge.

4


Examples of intelligent behavior:

Interior design:

Using the space efficiently

Finding the drawers

…

5


Examples of intelligent behavior:

Route planning

Maps are symbolic

representations

6


The PSSH: Many kinds of intelligent behaviour that AI scientists have tried to replicate on

machines, including: natural language processing expert problem solving logical reasoning planning and scheduling.

This list is by no means exhaustive. One might add, for example: visual processing and recognition controlling physical movement

7


The PSSH: One conventional view of intelligence is that it is a process of:

taking in information from the world; processing it some sophisticated way and identifying a sensible

solution; doing something with the solution.

Activity 1.1 Page 18, the paper “computer science and the empirical inquiry: symbols and search” by Newell and Simon, the course DVD

8


The PSSH: Following this article, the aim of all science is to pin down the

essential facts of some aspect of nature. Newell and Simon propose just such a model of an intelligent

system, which they call a symbolic system or a physical symbol system.

9


The PSSH: Symbols can be gathered together into symbol structures or expressions the two terms

seem to be interchangeable.

Expressions (which can be single symbols) designate things in the world, which Simon and Newell term objects.

Some expressions specify processes that operate on other expressions within the physical symbol system.

The task of identifying a process, applying it to an expression and so changing the expression in some way is termed interpretation.

physical symbol systems are capable of manipulating not only the designations of objects, but also their own processes.

10


The PSSH: symbol, defining this simply as some entity that exists in the real world and obeys the laws

of physics.

a physical symbol system is a collection of symbols and symbol structures.

Certain symbols or symbol structures designate objects in the real world;

others represent processes that can occur.

The combination of all the symbol expressions within a physical symbol system determines the system’s state.

The state evolves over time as processes within the system (also expressed as symbols and symbol structures) operate on it.

The set of states that a physical symbol system can be in is termed the system’s state space,

11


Representation and reasoning:

Example one: the water jar problem

Example two, the Missionaries and cannibals problem.

12


13


Developing Representation:

Exercises 1.10, 1.11, 1.12, 1.13 P.29

The 32 dominos covering two squares on the chessboard

14


Example of symbol representations: Chess is the classic problem of AI.

15


Example of symbol representations: Chess is the classic problem of AI. Assigning symbols: w for white and b for black, and the positions:

on(a,1,_) on(b,1,_) on(c,1,w(K)) on(d,1,w(R)) ... on(b,6,_) on(c,6,b(N)) on(d,6,b(P))

16

Block II, Unit I, Fundamentals of Symbolic AI Each square is represented by an expression consisting of a predicate

name (in this case on) and three arguments (the symbols inside the brackets, separated by commas).

The first argument represents the square’s column (known as the file in chess jargon); the second represents the row (known as the rank);

The third gives the type of piece there. The piece is itself represented by a predicate name (w or b) and one

argument (the kind of piece), with _ representing an empty square.

17


Processes: the moves that pieces are legally capable of making. Each chess piece moves in its own particular way. The rook moves horizontally or vertically as far as you like, but must

stop when reaching another piece (it can never leap over an occupied square).

No piece, including the rook, may land on a square occupied by a piece of its own colour, but it is possible to move onto a square occupied by a piece of the opposite colour

18


Processes: the moves that pieces are legally capable of making.

19


Processes: the moves that pieces are legally capable of making. Taking the position shown in Figure 1.11, two of the expressions

representing the state of the board at this point would be:on( d , 1 , w ( R ))

on( d , 5 ,_)

Now suppose it is white’s turn to move and he decides to move his rook on d1 to d5

The move will then alter these two expressions to:

on( d , 1 ,_)

on( d , 5 , w ( R ))

20

Block II, Unit I, Fundamentals of Symbolic AI Exercise 1.15 P.34: amazing conclusion. There are around 1043 possible state of a chess game 32 pieces to store, 6 types, we need 194 bits=24 bytes to store each

position To hold all the reasonable games of chess, we need 2.4 1035 Gb This requires a hard disk of 1.2 1030 m2

To compare, the surface area of the earth is around 5 1014 m2

Possible solution: heuristic search.

21

Block II, Unit I, Fundamentals of Symbolic AI Known patterns: the pin and the fork.

22

Block II, Unit I, Fundamentals of Symbolic AI Expert systems: One particular type of intelligent systems is the expert system. Its design is as follows:

23

Block II, Unit I, Fundamentals of Symbolic AI Expert systems: example

Car maintenance

Doctor diagnostics

Production rules:

IF NOT engine-turns-over

AND NOT headlights-are-bright

THEN problem-with-battery

These rules can be used for inference, where new knowledge is discovered about a situation.

The rule above is able to determine whether the battery in a car is flat.

The rule is in two parts: the antecedent or premise (everything between the IF and the THEN) and the consequent or conclusion (everything after the THEN).

If the inference process can show that the antecedent is true, then the consequent is also true.

When a rule is used to prove its consequent, that rule is said to fire.

24

Block II, Unit I, Fundamentals of Symbolic AI Expert systems dialogue

25


Representations: facts and actions Declarative knowledge: can be explicitly state: my cup is empty Procedural knowledge: knowledge how to speak French, or how to

learning how to drive, or prepare a satisfying meal. Base knowledge: concerns the rules of the game Meta knowledge: can be used to guide the player’s reasoning to

explore the promising moves while quickly disregarding the bad ones. Meta reasoning: it makes possible for the system to reason about what

it is doing and how it is doing it. For example, a chess player may know that a particular move will set up an attack later.

26


Representations: features of good representation Coverage Parsimony (concise) Clarity Use of derived knowledge Specificity: advantage of using existing notations: predicate logic is

often used as a notation for expressing knowledge.

27


Examples of representations:

symbolic logic: propositional logic and predicate logic Propositional logic

R represents the proposition “it is raining” U represents the proposition “I am carrying an umbrella” W represents the proposition “I am wet”

(R U) W

28


29


30


Examples of representations:

symbolic logic: propositional logic and predicate logic Predicate logic: allows us to state sentence like “All men are mortal” Mortal(man), If man(turing) → Mortal(turing) Predicate logic use two quantifiers

Universal quantifier , x, man(x) → mortal(x) Existential quantifier

Predicate logic sees the world as collections of objects that have properties and relations with other objects; objects are different from each other because they have different properties (such as names).

31


32


Semantic networks: they represent a set of entities and relationship between them. The entities can be either objects, collections of objects or concepts. There exits two types

1- definitional networks (define concepts in term of other concepts)

2- assertional networks (describe sets of assertions or propositions)

33


Ovals represent concepts Asterisk marking the integer oval

indicates that this concept is of primitive type

Broad arrows represent subtype Arrows with a circle represents role

34


Reasoning: A goal can be a single goal: jugs problem It can be anyone of a huge number of states (chess) Satisfaction problem: if any valid solution is as good as any other. Optimization problem: if the objective is not to find ANY solution,

but to find a good (or even the best) solution (TSP) Forward chaining: start from the hypothesis Backward chaining: start from the potential goal

Example: car diagnosis problem.

35


Reasoning:

36

Block II, Unit I, Fundamentals of Symbolic AI

Documents

symbolic system

intelligent system

intelligent tasks

kinds of intelligent

subset of intelligent

scheduling tasks

symbol structures

airline scheduling