Chapter 13 Artificial Intelligence

1
Chapter 13Artificial Intelligence

• Division of labor
• Knowledge Representation
• Reasoning Tasks
• Intelligent Searching
• Expert Systems

2
AI

• AI is the part of computer science that attempts
  to make computers act like human beings.
• The Turing test allows a human to interrogate two
  entities both hidden from the interrogator. One
  of them is a human and the other is a computer.
  If the interrogator cannot tell between the
  responses then the computer has exhibited
  sufficient human intelligence to pass the Turing
  test.

3
Division of labor

• Humans perform a variety of tasks
• Main categories are given below
• Computational tasks (math, sorting, searching)
• Algorithms can be written for the computational
  tasks.
• Computers are better at performing computational
  tasks provided the algorithms written are
  correct.
• Recognition tasks (recognizing people,
  recollection, understanding the language you
  speak)
• Humans are better at this.
• It involves sensory and motor skills tasks too.
• An infant recognizes the mothers face.
• Difficult to write algorithms for this.
• Reasoning tasks (deduction, intuition, making
  decisions)
• Difficult to write algorithms here too.
• Is there a traditional step by step approach?
• The complex recognition and reasoning tasks that
  are so easy for us to do are very difficult for
  computers.

4
Knowledge representation

• Knowledge is a body of facts or truths.
• For a computer to make use of it we need to have
  some representational form in which the knowledge
  is stored in the computer.
• For computational tasks, the relevant knowledge
  is often numeric or text items.
• What about other complex knowledge?
• We have four possibilities.
• Natural language
• Formal language
• Pictorial
• graphical

5

• Natural language a para. in English or whatever
  language you are familiar with.
• The representation form is text.
• The meaning of the paragraphs is also important
• Formal language Sacrifices richness of
  expression for precision of expression.
• Short phrases.
• The term formal language means the language of
  formal logic usually expressed symbolically.
• D(x) ? L(x) could mean if x is a dog then x has
  four legs.
• All rules are in the form of if then else stmts.
• For all x (D(x) ? L(x))
• For all x that has the dog property also has the
  four legged property.

6

• We could use D(x) to symbolize that the symbolic
  entiry x has the attribute of being a dog.
• B(x) would be interpreted as meaning that x has
  the attributed of being brown.
• Below are given six formal language statements.
• Spot is a dog D(S)
• Spot is brown B(S)
• Every dog has four legs for all x (D(x)? F(x))
• Every dog has a tail for all x (D(x) ?T(x))
• Every dog is a mammal for all x (D(x) ? M(x))
• Every mammal is warm blooded for all x (M(x)
  ? W(x))

7

• Pictorial
• Information is stored in pictorial form as an
  image.
• A grid of pixels that have attributes of shading
  and color
• In our example it would show Spot our dog to be
  brown with four legs. And some labeling saying,
  Spot the brown dog.
• The visual representation would contain
  additional knowledge about Spots appearance that
  is not embodied in the natural language para or
  the formal language stmts.
• It would fail to capture the knowledge that Spot
  is a mammal and that mammals are warm blooded.
• It would fail to tell us that all dogs have four
  legs and a tail.
• (A photo of a three legged dog does not tell us
  that all dogs have three legs)

8

• Graphical
• Math sense of a graph which has nodes and arcs.
• Called a semantic net.

Mammal
Warm blooded
is
Is a
Four legs
has
Dog
Tail
has
instance
Spot
Brown
Is color
9
Features of any knowledge representation scheme

• Adequacy
• Capture all relevant knowledge.
• Natural language can capture a lot of knowledge
  but it may be difficult to extract it. There are
  various nuances in languages. Shades of grey!
• A formal language has the advantage of extracting
  the essentials.
• Efficiency
• The form should be minimalist.
• Avoid redundant information
• Allowing some knowledge that is not explicitly
  represented to be inferred from the knowledge
  that is explicitly represented.
• Eg. if spot is a dog and he has four legs then he
  has to be a mammal. No need to store again that a
  dog is a mammal. This info can be got from the
  info already there.
• Extendibility
• Should be easy to add new features
• Appropriateness
• The representation should be appropriate for the
  knowledge domain being represented. Eg. a
  pictorial scheme would appear to be the most
  appropriate way to represent the knowledge base
  for a problem dealing with recognition of visual
  images.
• The choice of the representational form for
  knowledge depends on the knowledge to be captured
  and on the type of task for which the knowledge
  is to be put to use.

10
Recognition Tasks

• We want to make the computer think like humans
• Mimic the human brain.
• We have neurons, dendrites and synapses and an
  axon.
• Figure 13.4
• Each neuron can be considered as a simple
  computational device with a single on-off output.
  
• Power of the brain comes from the
  interconnections between these neurons.
• Neurons operate on a time scale of milliseconds
  as opposed to nanoseconds in which computer
  operations are measured.
• Scientists have built artificial neural networks
  to simulate this.
• They have applied those networks to recognition
  tasks.
• Lots of training is required prior to making the
  neurons recognize things.
• Figure 13.5 and 13.6
• Neuron fires (on) if the sum of its inputs is
  greater than its threshold value.
• We can have various layers of neurons
• We need an input layer
• An output layer
• And the hidden layers if needed.

11

• Uses of neural networks
• Handwriting recognition
• Speech recognition
• Recognizing patterns indicative of credit card
  fraud
• Recognizing bad credit risks for loans
• Predicting the odds of cancer
• Finding the best routing algorithm is a large
  communications network.

12
Reasoning Tasks

• We reason based on a large body of facts and past
  experience to come to a conclusion.
• We want computers to emulate this.
• Intelligent searching
• Expert systems

13
Intelligent searching

• We saw the sequential search and binary search
  algorithms
• Sequential search ?(n)
• Binary search ?(lg n)
• A decision tree for a search algorithm
  illustrates the possible next choices of items to
  search if the current item is not the target.
• Figure 13.10 is the decision tree for sequential
  search
• Figure 13.11 is the decision tree for binary
  search

14
Classical search problem benefits from 2
simplifications

• The set of items being searched is a linear list.
  At each point in the search, if the target is not
  found, the choice of where to look next is highly
  constrained.
• We seek a perfect match so the comparison of the
  target against the list item results in a binary
  decision either they match or they do not.
• What if point 1 above does not hold? What is
  there are huge numbers of next choices to try
  from?
• Figure 3.12 ( A decision tree with exponential
  growth)
• Many problems fall in this category.
• Finding the shortest path through a network.
• Finding the winning move in a chess game
• Given that time and computing resources are
  limited we need an intelligent search
• An intelligent search narrows down the number of
  tree branches that must be tried and thereby puts
  a cap on the exponential growth.

15
2 aspects of intelligent searching

• Best first search pick a good starting node
  which to search.
• Recognize where you are and the desired
  destination.
• Means end analysis
• Making a wise choice of where to go next from the
  present node.
• Requires knowing what to do to reduce the
  differences you recognized in the previous step.

16

• Deep blue V/S Kasparov (1997)
• An intelligent chess playing strategy is one that
  makes an appropriate first move and that at each
  step makes a move more likely than others to lead
  to a winning game
• Brute force method try out all the possible next
  moves , all the possible moves after those moves
  and so on. No time to do this.
• Need intelligent searching
• Must be a storehouse of experience that can be
  consulted on the basis of the overall present
  configuration of the board.
• It is the size of the search space that is huge.

17

• What if we no longer want to find an exact match
• We are only looking for close matches.
• Internet search engines perform this kind of
  search.
• When you enter keywords in search engines it
  scans its database of web documents containing
  those words and return a list of possible hits
  usually ordered from closest to farthest away
  from the keywords you entered.
• This is still unintelligent searching
• Search engine knows nothing about user
  requesting the search.
• Search engine doesnt search the entire web every
  time you type in something to search.
• It already contains an database of web documents
  and does this collecting at night.
• Software that crawls through the web looking for
  new sites to add to their engine database.
• What if engine could sense users needs? Eg.
  search for mustang?
• Intelligent agents is a software technology
  designed to interact collaboratively with a user
  somewhat in the model of a Personal assistant.

18

• Agents
• Sense the environment and take appropriate
  action.
• Examples
• Desktop assistants
• Push technology to bring news to your mail box.
• Vote on an article on the web and then it
  dynamically adjusts what it send to you later.
• On line catalog sales company uses an agent that
  monitors incoming orders and makes suggestions
  eg. if a customer buys a certain book it can give
  options of other books bought by customers who
  bought the same book. (www.amazon.com does this)
• Travel agents(electronic)
• Buying selling stocks.

19
Expert Systems

• Agents still perform limited tasks.
• Consider managing a busy hospital room
• Draw on past experience
• Understand conditions and consequences
• Knowledge about the hospitals capabilities and
  resources in general
• Based on this knowledge base, a chain of
  reasoning is followed that leads, for eg. to a
  decision to treat patient A immediately in a
  particular fashion and to let patient B wait.
  (logical reasoning)

20

• AI simulates this kind of reasoning through the
  use of rule-based systems, which are also called
  expert systems or knowledge-based systems.
• A rule based system attempts to mimic the human
  brain to engage pertinent facts and string them
  together in a logical fashion to reach some
  conclusion.
• A rule based system must therefore contain these
  two components
• A knowledge base set of facts specific to your
  domain
• An inference engine mechanism to select the
  relevant facts and for reasoning from them in a
  logical way.

21

• Facts in knowledge base can be simply assertions
  like
• Lincoln was president during the Civil War
• Kennedy was president before Nixon
• FDR was president before Kennedy
• Another type of fact is a rule of the form if ..
  Then .
• If X was president before Y then X precedes Y
• If X was president before Z and Z precedes Y
  then X precedes Y
• From the above we get two more assertions
• Kennedy precedes Nixon
• FDR precedes Nixon

22

• Inference engines for rule based systems can
  proceed in several ways.
• Forward chaining begins with assertions and
  tries to match those assertions to the if clauses
  of rules to get new assertions.
• Backward chaining begins with a proposed
  conclusion and tries to match it with the then
  clause of rules and then looks at the
  corresponding if clauses and tries to match those
  with assertions or with the then clauses of other
  rules.
• This process continues until all if clauses that
  arise have been successfully matched with
  assertions in which case the proposed conclusion
  is justified or until no match is possible in
  which case the proposed conclusion is rejected.
• Eg. we would have said our Hypothesis FDR
  precedes Nixon (work backwards to justify
  hypothesis)

23

• Explanation facility allows user to see the
  assertions and rules used in arriving at a
  conclusion as sort of a check on the path of
  reasoning.
• A rule based system will work as long as its
  assertions and rules are good and cover all
  possibilities.
• Knowledge engineering The builder of such a
  system acquires the information for the knowledge
  base by consulting experts in the domain and
  mining their expertise.
• Requires interaction with the human expert.
• Rule based systems have been used in
• Medical diagnosis.
• Computer chip design
• Monitoring manufacturing work.
• Etc..

24

• The end of chapter 13