chapter 13 artificial intelligence. 2 thinking machines a computer can do some things better --and...

Chapter 13

Artificial Intelligence

2

Thinking Machines A computer can do some things better --and

certainly faster--than a human can: Adding a thousand four-digit numbers Counting the distribution of letters in a book Searching a list of 1,000,000 numbers for

duplicates Matching finger prints

3

Thinking Machines BUT a computer would

have difficulty pointing out the cat in this picture, which is easy for a human.

Artificial intelligence (AI) The study of computer systems that attempt to model and apply the intelligence of the human mind.

Figure 13.1 A computer might have trouble identifying the cat in this picture.

4

Thinking MachinesArtificial: humanly contrived often on a natural

model

Intelligence: the ability to apply knowledge to manipulate one's environment or to think abstractly as measured by objective criteria

Clearly, intelligence is an internal characteristic. How can it be identified?

5

In the beginning… In 1950 Alan Turing wrote a paper titled

Computing Machinery And Intelligence, in which he proposed to consider the question “Can machines think?”

But the question is “loaded” so he proposed to replace it with what has since become known as the Turing Test.

6

The Turing Test The important question for Turing wasn’t

“Can machines think?” but“How will we know if they can?”

The Turing test is used to empirically determine whether a computer has achieved intelligence.

7

The Imitation GameThe 'imitation game' is played with three people:

a man (A) a woman (B) and an interrogator (C) who may be of either sex.

The interrogator stays in a room apart from the other two.

The object of the game for the interrogator is to determine which of the other two is the man and which is the woman.

8

The Imitation GameThe interrogator is allowed to put questions to A

and B.

It is A's object in the game to try and cause C to make the wrong identification.

The object of the game for the third player (B) is to help the interrogator.

9

The Imitation GameWe now ask the question:

'What will happen when a machine takes the part of B in this game?'

Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman?

10

The Turing TestFigure 13.2 In a Turing test, the interrogator must determine which respondent is the computer and which is the human

11

The Turing Test There are authors who question the validity of

the Turing test . The objections tend to be of 2 types.

The first is an attempt to distinguish degrees, or types of equivalence.

12

The Turing Test Weak equivalence: Two systems (human and

computer) are equivalent in results (output), but they do not arrive at those results in the same way.

Strong equivalence: Two systems (human and computer) use the same internal processes to produce results.

13

The Turing Test The Turing Test, they argue, can demonstrate

weak equivalence, but not strong. So even if a computer passes the test we won’t be able to say that it thinks like a human.

Of course, neither they nor anyone else can explain how humans think!

So strong equivalence is a nice theoretical construction, but impossible in reality.

14

The Turing Test The other objection is that a computer might seem to be

behaving in an intelligent manner, while it’s really just imitating behaviour.

This might be true, but notice that when a parrot talks, or a horse counts, or a pet obeys our instructions, or a child imitates its parents we take all of these things to be signs of intelligence.

If a parrot mimicking human sounds can be considered intelligent (at least to some small degree) then why wouldn’t a computer be considered intelligent (at least to some small degree) for imitating other human bevaviour?

15

Turing’s View “I believe that in about fifty years time it will

be possible to programme computers with a storage capacity of about 109 to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning.”

16

Can Machines Think? No machine has passed the Turing Test, yet.

Loebner Prize established in 1990 $100,000 and a gold medal will be awarded to the

first computer whose responses are indistinguishable from a human's

Let’s look at one of the more difficult tasks for AI…

17

Aspects of AI Knowledge Representation

Semantic Networks Search Trees

Expert Systems Neural Networks Natural Language Processing Robotics

18

Knowledge Representation The knowledge needed to represent an object

or event depends on the situation. There are many ways to represent knowledge.

One is natural language. Even though natural language is very descriptive,

it doesn’t lend itself to efficient processing.

19

Semantic Networks Semantic network: A knowledge

representation technique that focuses on the relationships between objects.

A directed graph is used to represent a semantic network (net).

20

Semantic Networks

Figure 13.3 A semantic network

21

Semantic Networks The relationships that we represent are

completely our choice, based on the information we need to answer the kinds of questions that we will face.

The types of relationships represented determine which questions are easily answered, which are more difficult to answer, and which cannot be answered.

eos

22

Search Trees Search tree: a structure that represents all

possible moves in a game, for both you and your opponent.

The paths down a search tree represent a series of decisions made by the players.

23

Search Trees

Figure 13.4 A search tree for a simplified version of Nim

24

Search Trees Search tree analysis can be applied nicely to

other, more complicated games such as chess. Because Chess trees are so large, only a

fraction of the tree can be analyzed in a reasonable time limit, even with modern computing power.

25

Search TreesTechniques for searching trees Depth-first: a technique that involves the analysis of

selected paths all the way down the tree. Breadth-first: a technique that involves the analysis

of all possible paths but only for a short distance down the tree.

Breadth-first tends to yield the best results.

26

Search Trees

Figure 13.5 Depth-first and breadth-first searches

27

Search TreesEven tough the breadth-first approach tends to

yield the best results, we can see that a depth-first search will get to the goal sooner – IF we choose the right branch.

Heuristics are guidelines that suggest taking one path rather than another one.

eos

28

Expert Systems Knowledge-based system: a software system that

embodies and uses a specific set of information from which it extracts and processes particular pieces.

Expert system: a software system based the knowledge of experts in a specialized field. An expert system uses a set of rules to guide its

processing. The inference engine is the part of the software that

determines how the rules are followed.

29

Expert Systems Example: What type of treatment should I put on

my lawn? NONE—apply no treatment at this time TURF—apply a turf-building treatment WEED—apply a weed-killing treatment BUG—apply a bug-killing treatment FEED—apply a basic fertilizer treatment WEEDFEED—apply a weed-killing and fertilizer

combination treatment

30

Expert Systems Boolean variables

BARE—the lawn has large, bare areas

SPARSE—the lawn is generally thin

WEEDS—the lawn contains many weeds

BUGS—the lawn shows evidence of bugs

31

Expert Systems Some rules

if (CURRENT – LAST < 30) then NONE

if (SEASON = winter) then not BUGS

if (BARE) then TURF

if (SPARSE and not WEEDS) then FEED

if (BUGS and not SPARSE) then BUG

if (WEEDS and not SPARSE) then WEED

if (WEEDS and SPARSE) then WEEDFEED

32

Expert Systems An execution of our inference engine

System: Does the lawn have large, bare areas? User: No System: Does the lawn show evidence of bugs? User: No System: Is the lawn generally thin? User: Yes System: Does the lawn contain significant weeds? User: Yes System: You should apply a weed-killing and fertilizer

combination treatment.eos

33

Artificial Neural Networks Attempt to mimic the actions of the neural

networks of the human body. Let’s first look at how a biological neural

network works: A neuron is a single cell that conducts a

chemically-based electronic signal. At any point in time a neuron is in either an

excited or inhibited state.

34

Artificial Neural Networks A series of connected neurons forms a pathway. A series of excited neurons creates a strong

pathway. A biological neuron has multiple input tentacles

called dendrites and one primary output tentacle called an axon.

The gap between an axon and a dendrite is called a synapse.

35

Artificial Neural Networks

Figure 13.6 A biological neuron

36

Artificial Neural Networks A neuron accepts multiple input signals and

then controls the contribution of each signal based on the “importance” the corresponding synapse gives to it.

The pathways along the neural nets are in a constant state of flux.

As we learn new things, new strong neural pathways are formed.

37

Artificial Neural Networks Each processing element in an artificial neural

net is analogous to a biological neuron. An element accepts a certain number of input

values and produces a single output value of either 0 or 1.

Associated with each input value is a numeric weight.

38

Sample “Neuron”

Artificial “neurons” can be represented as elements. Inputs are labelled v1, v2 Weights are labelled w1, w2 The threshold value is represented by T O is the output

39

Artificial Neural Networks The effective weight of the element is defined to

be the sum of the weights multiplied by their respective input values:

v1*w1 + v2*w2

If the effective weight meets the threshold, the unit produces an output value of 1.

If it does not meet the threshold, it produces an output value of 0.

40

Artificial Neural Networks The process of adjusting the weights and

threshold values in a neural net is called training.

A neural net can be trained to produce whatever results are required.

41

Sample “Neuron”

If the input Weights and the Threshold are set to the above values, how does the neuron act?

Try a Truth Table…

42

Sample “Neuron”

v1 v2 v1*w1 v2*w2 ∑ O0 0 0 0 0 0

0 1 0 .5 .5 0

1 0 .5 0 .5 0

1 1 .5 .5 1 1

w1=.5, w2=.5, T=1With the weights set to .5 this neuron behaves like an

AND gate.

43

Sample “Neuron”

How about now?

44

Sample “Neuron”

v1 v2 v1*w1 v2*w2 ∑ O0 0 0 0 0 0

0 1 0 1 1 1

1 0 1 0 1 1

1 1 1 1 1 1

w1=1, w2=1, T=1With the weights set to 1 this neuron behaves like an

OR gate.

eos

45

Natural Language Processing There are three basic types of processing going on

during human/computer voice interaction: Voice recognition — recognizing human words Natural language comprehension — interpreting human

communication Voice synthesis — recreating human speech

Common to all of these problems is the fact that we are using a natural language, which can be any language that humans use to communicate.

46

Voice Synthesis There are two basic approaches to the solution:

Dynamic voice generation Recorded speech

Dynamic voice generation: A computer examines the letters that make up a word and produces the sequence of sounds that correspond to those letters in an attempt to vocalize the word.

Phonemes: The sound units into which human speech has been categorized.

47

Voice Synthesis

Figure 13.7 Phonemes for American English

48

Voice Synthesis Recorded speech: A large collection of words

is recorded digitally and individual words are selected to make up a message.

Telephone voice mail systems often use this approach:

“Press 1 to leave a message for Nell Dale; press 2 to leave a message for John Lewis.”

49

Voice Synthesis Each word or phrase needed must be recorded

separately. Furthermore, since words are pronounced

differently in different contexts, some words may have to be recorded multiple times. For example, a word at the end of a question rises

in pitch compared to its use in the middle of a sentence.

eos

50

Voice Recognition The sounds that each person makes when speaking

are unique. We each have a unique shape to our mouth, tongue,

throat, and nasal cavities that affect the pitch and resonance of our spoken voice.

Speech impediments, mumbling, volume, regional accents, and the health of the speaker further complicate this problem.

51

Voice Recognition Furthermore, humans speak in a continuous, flowing manner.

Words are strung together into sentences. Sometimes it’s difficult to distinguish between phrases like “ice

cream” and “I scream”. Also, homonyms such as “I” and “eye” or “see” and “sea”.

Humans can often clarify these situations by the context of the sentence, but that processing requires another level of comprehension.

Modern voice-recognition systems still do not do well with continuous, conversational speech.

eos

52

Natural Language Comprehension Even if a computer recognizes the words that

are spoken, it is another task entirely to understand the meaning of those words. Natural language is inherently ambiguous,

meaning that the same syntactic structure could have multiple valid interpretations.

53

Natural Language Comprehension (syntax)

Syntax is the study of the rules whereby words or other elements of sentence structure are combined to form grammatical sentences.

So a syntactical analysis identifies the various parts of speech in which a word can serve, and which combinations of these can be assembled into sensible sentences.

54

Natural Language Comprehension (syntax)A single word can represent multiple parts of

speech.

Consider an example:

Time flies like an arrow.To determine what it means we first parse the

sentence into its parts.

55

Natural Language Comprehension (syntax)‘time’ can be used as a noun, a verb, or an adjective.

‘flies’ can be used as a noun, or a verb.

‘like’ can be used as a noun, a verb, a preposition, an adjective, or an adverb.

‘an’ can be used only as an indefinite article.

‘arrow’ can be used as a noun, or a verb.

56

Natural Language Comprehension (syntax)The table below summarises the various possibilities.

The a priori total number of syntactical interpretations is:

3 * 2 * 5 * 1 * 2 = 60

adjective adverb article noun preposition verb

time

flies

like

an

arrow

57

Natural Language Comprehension (syntax)This number can be reduced by applying syntactical rules. For

example, articles are used to “indicate nouns and to specify their application” so ‘arrow’ must be a noun, not a verb. This cuts the number of possible sentences in half.


time

flies

like

an

arrow

58

Natural Language Comprehension (syntax)Other grammar rules will reduce this number further, but how

far?

Rather than eliminate combinations, it may be easier to build a list of reasonable possibilities.

Consider the ways in which ‘time’ can be used: Adjective Noun Verb (intr)

59

Natural Language Comprehension (syntax)An adjective is “the part of speech that modifies a noun” so when

‘time’ is an adjective, ‘flies’ must be a noun.

A sentence needs a verb, so if ‘flies’ is a noun, ‘like’ must be the verb.

If ‘time’ is an adjective there is only ONE possible meaning.


time

flies

like

an

arrow

skip

60

Natural Language Comprehension (syntax)


time

flies

like

an

arrow

When ‘time’ is a noun there appear to be 10 possibilities, but there can only be one verb so there are only 5.


time

flies

like

an

arrow

61

Natural Language Comprehension (syntax)A sentence can have only one verb, so when ‘time’ is a verb there

are only 4 possible combinations of the other parts.


time

flies

like

an

arrow

The syntactical analysis reveals a total of 10 possible ways these words can form a sentence.

62

Natural Language Comprehension (semantics)A syntactical analysis provides a structure, a semantic analysis

adds meaning.

Consider the first case in the syntactical analysis.


time

flies

like

an

arrow

What would the words mean under this syntactical interpretation?

63

Natural Language Comprehension (semantics)

‘time’ adjective“of, relating to, or measuring time”

‘flies’ noun“two-winged insect”

‘like’ verb“find pleasant or attractive”

‘an’ article

‘arrow’ noun“missile having a straight thin shaft with a pointed head at one end and often flight-stabilizing vanes at the other”

time flies (not house flies)

enjoy

an arrow(watching it, chasing it, eating it)

64





‘an’ article



enjoy


65





‘an’ article



enjoy


66





‘an’ article



enjoy


67





‘an’ article



enjoy

an arrow

(watching it, chasing it, eating it?)

68


This interpretation sounds absurd to us, but as our analysis showed, it’s a perfectly logical interpretation of the words.

This is problem is referred to as a lexical ambiguity.

It arises because words have multiple syntactical and semantic associations.

In this case there are many syntactically and semantically valid interpretations.

skip

69


Consider the cases in which ‘time’ is a verb.

The Free Dictionary lists 5 meanings: 1. To set the time for (an event or occasion). 2. To adjust to keep accurate time. 3. To adjust so that a force is applied or an action occurs at the desired

time: timed his swing so as to hit the ball squarely. 4. To record the speed or duration of: time a runner. 5. To set or maintain the tempo, speed, or duration of: time a

manufacturing process.

70


Not only are there 4 syntactical interpretations of the sentence when ‘time’ is a verb, each of those must be analysed under 5 semantic interpretations.

Let’s explore a few: 4. To record the speed or duration of: time a runner.

As a transitive verb, ‘time’ requires an object, so ‘flies’ is a noun. We’ve seen one meaning, here’s another of the 12 listed in The Free Dictionary :

5. Baseball A fly ball.

71


So, when you measure the length of time it takes a fly ball to travel its route, use the same techniques that you use when you time an arrow.

This interpretation isn’t as odd as it might seem.

We know that the paths taken by fly balls and arrows are all parabolic arcs, so it makes sense to time them the same way.

72


Consider the interpretations in which ‘time’ is a noun and ‘flies’ is a verb. (These are the parts of speech we would usually map onto them in this context.)

The Free Dictionary lists 29 distinct meanings for the noun ‘time’. Consider the first one:

1. a. A nonspatial continuum in which events occur in apparently irreversible succession from the past through the present to the future.

This is a normal sense of what we mean by ‘time’ but which meaning of the verb ‘flies’ should be used?

73


The Free Dictionary lists only one definition:“Third person singular present tense of fly.”

In this context, ‘fly’ can only be an intransitive verb, for which there are 14 different nuances. The most commonly used of these is “To engage in flight.”

But can that be the meaning intended in our saying? How does a computer program decide?

It’s easy to see why NLC is difficult for computers.

eos

74

Natural Language Comprehension A natural language sentence can also have a

syntactic ambiguity because phrases can be put together in various ways.

I saw the Grand Canyon flying to New York.

Is it possible that the Grand Canyon flew to New York?

75

Natural Language Comprehension Referential ambiguity can occur with the use of

pronouns.

The brick fell on the computer but it is not broken.

Since ‘it’ is the subject of the second clause, its referent is the subject of the first clause – ‘the brick’!

Are you happy knowing that no harm came to the brick?

eos

76

Robotics Mobile robotics: The study of robots that move relative to their environment, while

exhibiting a degree of autonomy. In the sense-plan-act (SPA) paradigm the world of the robot is represented in a complex

semantic net in which the sensors on the robot are used to capture the data to build up the net.

Figure 13.8 The sense-plan-act (SPA) paradigm

77

Subsumption Architecture Rather than trying to model the entire world all the time, the

robot is given a simple set of behaviours each associated with the part of the world necessary for that behaviour

Figure 13.9 The new control paradigm

78

Subsumption Architecture

Figure 13.10 Asimov’s laws of robotics are ordered.

eof

chapter 13 artificial intelligence. 2 thinking machines a computer can do some things better --and...

Documents

human slide

ai slide

artificial intelligence

turing test figure

turing test weak equivalence

imitation game

alan turing

turing wasnt