Artificial Intelligence

Course Learning Outcomes

At the end of this course:
 Knowledge and understanding
You should have a knowledge and understanding of the basic concepts of Artificial
Intelligence including Search, Game Playing, Knowledge Based Systems (KBS),
Planning and Machine Learning.
 Intellectual skills
You should be able to use this knowledge and understanding of appropriate
principles and guidelines to synthesise solutions to tasks in AI and to critically
evaluate alternatives.
 Transferable Skills
You should be able to solve problems and evaluate outcomes and alternatives
Areas of AI and Some Dependencies
Knowledge
Search Logic Representation

Machine
Planning
Learning

Expert
NLP Vision Robotics Systems
 What is Artificial Intelligence ?
 Making computers that think?
 The automation of activities we associate with human thinking, like
decision making, learning ... ?
 The art of creating machines that perform functions that require
intelligence when performed by people ?
 The study of mental faculties through the use of computational models ?
What is Artificial Intelligence ?
 The study of computations that make it possible to perceive,
reason and act ?
 A field of study that seeks to explain and emulate intelligent
behaviour in terms of computational processes ?
 A branch of computer science that is concerned with the
automation of intelligent behaviour ?
 Anything in Computing Science that we don't yet know how to do
properly ? (!)
 AI Definitions
 The study of how to make programs/computers do things
that people do better Thinking
 The study of how to make computers solve problems which machines or
require knowledge and intelligence machine
 The exciting new effort to make computers think … intelligence
machines with minds
 The automation of activities that we associate with human
thinking (e.g., decision-making, learning…) Studying
 The art of creating machines that perform functions that cognitive
require intelligence when performed by people
faculties
 The study of mental faculties using computational models
 A field of study that seeks to explain and emulate intelligent
behavior in terms of computational processes
 The branch of computer science that is concerned with the Problem
automation of intelligent behavior Solving and
CS
 So, What Is AI?
 AI as a field of study
 Computer Science
 Cognitive Science
 Psychology
 Philosophy
 Linguistics
 Neuroscience
 AI is part science, part engineering
 AI often must study other domains in order to implement systems
 e.g., medicine and medical practices for a medical diagnostic system, engineering and
chemistry to monitor a chemical processing plant
 AI is a belief that the brain is a form of biological computer and that the
mind is computational
 AI has had a concrete impact on society but unlike other areas of CS, the
impact is often
 felt only tangentially (that is, people are not aware that system X has AI)
 felt years after the initial investment in the technology
 What is Intelligence?
 Is there a “holistic” definition for intelligence?
 Here are some definitions:
 the ability to comprehend; to understand and profit from experience
 a general mental capability that involves the ability to reason, plan, solve problems, think
abstractly, comprehend ideas and language, and learn
 is effectively perceiving, interpreting and responding to the environment
 None of these tells us what intelligence is, so instead, maybe we can
enumerate a list of elements that an intelligence must be able to
perform:
 perceive, reason and infer, solve problems, learn and adapt, apply common sense,
apply analogy, recall, apply intuition, reach emotional states, achieve self-awareness
 Which of these are necessary for intelligence? Which are sufficient?
 Artificial Intelligence – should we define this in terms of human
intelligence?
 does AI have to really be intelligent?
 what is the difference between being intelligent and demonstrating intelligent
behavior?
What is Artificial Intelligence ?

THOUGHT Systems that thinkSystems that think

like humans rationally

Systems that act Systems that act

BEHAVIOUR like humans rationally

HUMAN RATIONAL
 Systems that act like humans:
Turing Test
 “The art of creating machines that perform functions that
require intelligence when performed by people.” (Kurzweil)
 “The study of how to make computers do things at which, at
the moment, people are better.” (Rich and Knight)
 Systems that act like humans

?
 You enter a room which has a computer terminal.You have a
fixed period of time to type what you want into the terminal,
and study the replies. At the other end of the line is either a
human being or a computer system.
 If it is a computer system, and at the end of the period you
cannot reliably determine whether it is a system or a human,
then the system is deemed to be intelligent.
 Systems that act like humans

 The Turing Test approach

 a human questioner cannot tell if
 there is a computer or a human answering his question, via teletype
(remote communication)
 The computer must behave intelligently
 Intelligent behavior
 to achieve human-level performance in all cognitive tasks
 Systems that act like humans
 These cognitive tasks include:
 Natural language processing
 for communication with human
 Knowledge representation
 to store information effectively & efficiently
 Automated reasoning
 to retrieve & answer questions using the stored information
 Machine learning
 to adapt to new circumstances
The total Turing Test
 Includes two more issues:
 Computer vision
 to perceive objects (seeing)
 Robotics
 to move objects (acting)

*** This six disciplines compose most of the AI

What is Artificial Intelligence ?

THOUGHT Systems that thinkSystems that think

like humans rationally

Systems that act Systems that act

BEHAVIOUR like humans rationally

HUMAN RATIONAL
Systems that think like humans:
cognitive modeling
 Humans as observed from ‘inside’
 How do we know how humans think?
 Introspection vs. psychological experiments
 Cognitive Science
 “The exciting new effort to make computers think …
machines with minds in the full and literal sense”
(Haugeland)
 “[The automation of] activities that we associate with
human thinking, activities such as decision-making,
problem solving, learning …” (Bellman)
What is Artificial Intelligence ?

THOUGHT Systems that thinkSystems that think

like humans rationally

Systems that act Systems that act

BEHAVIOUR like humans rationally

HUMAN RATIONAL
Systems that think ‘rationally’
"laws of thought"
 Humans are not always ‘rational’
 Rational - defined in terms of logic?
 Logic can’t express everything (e.g. uncertainty)
 Logical approach is often not feasible in terms of
computation time (needs ‘guidance’)
 “The study of mental facilities through the use of
computational models” (Charniak and McDermott)
 “The study of the computations that make it possible to
perceive, reason, and act” (Winston)
What is Artificial Intelligence ?

THOUGHT Systems that thinkSystems that think

like humans rationally

Systems that act Systems that act

BEHAVIOUR like humans rationally

HUMAN RATIONAL
Systems that act rationally:
“Rational agent”
 Rational behavior: doing the right thing
 The right thing: that which is expected to maximize goal
achievement, given the available information
 Giving answers to questions is ‘acting’.
 I don't care whether a system:
 replicates human thought processes
 makes the same decisions as humans
 uses purely logical reasoning
Systems that act rationally
 Logic → only part of a rational agent, not all of
rationality
 Sometimes logic cannot reason a correct conclusion
 At that time, some specific (in domain) human knowledge or
information is used
 Thus, it covers more generally different situations of
problems
 Compensate the incorrectly reasoned conclusion
Systems that act rationally
 Study AI as rational agent –
2 advantages:
 It is more general than using logic only
 Because: LOGIC + Domain knowledge
 It allows extension of the approach with more scientific
methodologies
 Rational agents
 An agent is an entity that perceives and acts

 This course is about designing rational agents

 Abstractly, an agent is a function from percept histories to

actions:
[f: P* → A]
 For any given class of environments and tasks, we seek the
agent (or class of agents) with the best performance
 Caveat: computational limitations make perfect rationality
unachievable
 → design best program for given machine resources
 Artificial
 Produced by human art or effort, rather than originating
naturally.
 Intelligence
 is the ability to acquire knowledge and use it" [Pigford
and Baur]
 So, AI was defined as:
 AI is the study of ideas that enable computers to be intelligent.
 AI is the part of computer science concerned with design of
computer systems that exhibit human intelligence(From the
Concise Oxford Dictionary)
From the above two definitions, we can see that AI has two
major roles:
 Study the intelligent part concerned with humans.
 Represent those actions using computers.
Goals of AI
 To make computers more useful by letting them take over
dangerous or tedious tasks from human
 Understand principles of human intelligence
The Foundation of AI
 Philosophy
 At that time, the study of human intelligence began with no
formal expression
 Initiate the idea of mind as a machine and its internal operations
The Foundation of AI
 Mathematics formalizes the three main area of AI:
computation, logic, and probability
 Computation leads to analysis of the problems that can be
computed
 complexity theory
 Probability contributes the “degree of belief ” to handle uncertainty
in AI
 Decision theory combines probability theory and utility theory (bias)
The Foundation of AI
 Psychology
 How do humans think and act?
 The study of human reasoning and acting
 Provides reasoning models for AI
 Strengthen the ideas
 humans and other animals can be considered as information processing
machines
The Foundation of AI
 Computer Engineering
 How to build an efficient computer?
 Provides the artifact that makes AI application possible
 The power of computer makes computation of large and
difficult problems more easily
 AI has also contributed its own work to computer science,
including: time-sharing, the linked list data type, OOP, etc.
The Foundation of AI
 Control theory and Cybernetics
 How can artifacts operate under their own control?
 The artifacts adjust their actions
 To do better for the environment over time
 Based on an objective function and feedback from the environment
 Not limited only to linear systems but also other problems
 as language, vision, and planning, etc.
The Foundation of AI
 Linguistics
 For understanding natural languages
 different approaches has been adopted from the linguistic work
 Formal languages
 Syntactic and semantic analysis
 Knowledge representation
The main topics in AI
Artificial intelligence can be considered under a number of headings:
 Search (includes Game Playing).
 Representing Knowledge and Reasoning with it.
 Planning.
 Learning.
 Natural language processing.
 Expert Systems.
 Interacting with the Environment
(e.g. Vision, Speech recognition, Robotics)
We won’t have time in this course to consider all of these.
 Some Advantages of Artificial
Intelligence

 more powerful and more useful computers

 new and improved interfaces
 solving new problems
 better handling of information
 relieves information overload
 conversion of information into knowledge
The Disadvantages

 increased costs
 difficulty with software development - slow and expensive
 few experienced programmers
 few practical products have reached the market as yet.
Search
 Search is the fundamental technique of AI.
 Possible answers, decisions or courses of action are structured into an abstract
space, which we then search.
 Search is either "blind" or “uninformed":
 blind
 we move through the space without worrying about what is coming next, but
recognising the answer if we see it
 informed
 we guess what is ahead, and use that information to decide where to look
next.
 We may want to search for the first answer that satisfies our goal, or we may want
to keep searching until we find the best answer.
 What is Search?
 We define the state of the problem being solved as the values of the
active variables
 this will include any partial solutions, previous conclusions, user answers
to questions, etc
 while humans are often able
to make intuitive leaps, or
recall solutions with little
thought, the computer must
search through various
combinations to find a
solution
 To the right is a search space
for a tic-tac-toe game
Search Spaces and Types of Search
 The search space consists of all possible states of the problem as it is
being solved
 A search space is often viewed as a tree and can very well consist of an
exponential number of nodes making the search process intractable
 Search spaces might be pre-enumerated or generated during the search
process
 Some search algorithms may search the entire space until a solution is
found, others will only search parts of the space, possibly selecting
where to search through a heuristic
 Search spaces include
 Game trees like the tic-tac-toe game
 Decision trees (see next slides)
 Combinations of rules to select in a production system
 Networks of various forms (see next slides)
 Other types of spaces
Search Algorithms and Representations
 Breadth-first  We will study various forms of
 Depth-first
representation and uncertainty
handling in the next class period
 Best-first (Heuristic Search)
 Knowledge needs to be represented
 A*
 Production systems of some form
 Hill Climbing are very common
 Limiting the number of Plies  If-then rules
 Predicate calculus rules
 Minimax
 Operators
 Alpha-Beta Pruning  Other general forms include
 Adding Constraints semantic networks, frames, scripts
 Genetic Algorithms  Knowledge groups
 Forward vs Backward Chaining  Models, cases
 Agents
 Ontologies
Knowledge Representation & Reasoning
 The second most important concept in AI
 If we are going to act rationally in our environment, then we must have some way of
describing that environment and drawing inferences from that representation.
 how do we describe what we know about the world ?
 how do we describe it concisely ?
 how do we describe it so that we can get hold of the right piece of knowledge
when we need it ?
 how do we generate new pieces of knowledge ?
 how do we deal with uncertain knowledge ?
 Knowledge

Declarative Procedural

• Declarative knowledge deals with factoid questions (what is the capital of India?
Etc.)
• Procedural knowledge deals with “How”
• Procedural knowledge can be embedded in declarative knowledge
Planning
Given a set of goals, construct a sequence of actions that achieves those goals:
 often very large search space
 but most parts of the world are independent of most other parts
 often start with goals and connect them to actions
 no necessary connection between order of planning and order of execution
 what happens if the world changes as we execute the plan and/or our
actions don’t produce the expected results?
Learning
 If a system is going to act truly appropriately, then it must be
able to change its actions in the light of experience:
 how do we generate new facts from old ?
 how do we generate new concepts ?
 how do we learn to distinguish different situations in new environments ?
Interacting with the Environment
 In order to enable intelligent behaviour, we will have to
interact with our environment.
 Properly intelligent systems may be expected to:
 accept sensory input
 vision, sound, …
 interact with humans
 understand language, recognise speech,
generate text, speech and graphics, …
 modify the environment
 robotics
History of AI
 AI has a long history
 Ancient Greece
 Aristotle
 Historical Figures Contributed
 Ramon Lull
 Al Khowarazmi
 Leonardo da Vinci
 David Hume
 George Boole
 Charles Babbage
 John von Neuman
 As old as electronic computers themselves (c1940)
The ‘von Neuman’ Architecture
History of AI
 Origins
 The Dartmouth conference: 1956 [‘AI’ term is coined ]
 John McCarthy (Stanford)
 Marvin Minsky (MIT)
 Herbert Simon (CMU)
 Allen Newell (CMU)
 Arthur Samuel (IBM)
 The Turing Test (1950)
 “Machines who Think”
 By Pamela McCorckindale
 A Brief History of AI: 1950s
 Computers were thought of as an electronic brains
 Term “Artificial Intelligence” coined by John McCarthy
 John McCarthy also created Lisp in the late 1950s
 Alan Turing defines intelligence as passing the Imitation Game
(Turing Test)
 AI research largely revolves around toy domains
 Computers of the era didn’t have enough power or memory to solve
useful problems
 Problems being researched include
 games (e.g., checkers)
 primitive machine translation
 blocks world (planning and natural language understanding within the toy domain)
 early neural networks researched: the perceptron
 automated theorem proving and mathematics problem solving
 The 1960s
 AI attempts to move beyond toy domains
 Syntactic knowledge alone does not work, domain knowledge
required
 Early machine translation could translate English to Russian (“the spirit
is willing but the flesh is weak” becomes “the vodka is good but the
meat is spoiled”)
 Earliest expert system created: Dendral
 Perceptron research comes to a grinding halt when it is proved
that a perceptron cannot learn the XOR operator
 US sponsored research into AI targets specific areas – not
including machine translation
 Weizenbaum creates Eliza to demonstrate the futility of AI
 1970s
 AI researchers address real-world problems and solutions through expert
(knowledge-based) systems
 Medical diagnosis
 Speech recognition
 Planning
 Design
 Uncertainty handling implemented
 Fuzzy logic
 Certainty factors
 Bayesian probabilities
 AI begins to get noticed due to these successes
 AI research increased
 AI labs sprouting up everywhere
 AI shells (tools) created
 AI machines available for Lisp programming
 Criticism: AI systems are too brittle, AI systems take too much time and
effort to create, AI systems do not learn
 1980s: AI Winter
 Funding dries up leading to the AI Winter
 Too many expectations were not met
 Expert systems took too long to develop, too much money to invest,
the results did not pay off
 Neural Networks to the rescue!
 Expert systems took programming, and took dozens of man-years of
efforts to develop, but if we could get the computer to learn how to
solve the problem…
 Multi-layered back-propagation networks got around the problems of
perceptrons
 Neural network research heavily funded because it promised to solve
the problems that symbolic AI could not
 By 1990, funding for neural network research was slowly
disappearing as well
 Neural networks had their own problems and largely could not solve a
majority of the AI problems being investigated
 Panic! How can AI continue without funding?
 1990s: ALife
 The dumbest smart thing you can do is staying alive
 We start over – lets not create intelligence, lets just create “life” and
slowly build towards intelligence
 Alife is the lower bound of AI
 Alife includes
 evolutionary learning techniques (genetic algorithms)
 artificial neural networks for additional forms of learning
 perception and motor control
 adaptive systems
 modeling the environment
 Let’s disguise AI as something new, maybe we’ll get some
funding that way!
 Problems: genetic algorithms are useful in solving some optimization
problems and some search-based problems, but not very useful for
expert problems
 perceptual problems are among the most difficult being solved, very
slow progress
 Today: The New (Old) AI
 Look around, who is doing AI research?
 By their own admission, AI researchers are not doing “AI”, they are doing
 Intelligent agents, multi-agent systems/collaboration
 Ontologies
 Machine learning and data mining
 Adaptive and perceptual systems
 Robotics, path planning
 Search engines, filtering, recommendation systems
 Areas of current research interest:
 NLU/Information Retrieval, Speech Recognition
 Planning/Design, Diagnosis/Interpretation
 Sensor Interpretation, Perception, Visual Understanding
 Robotics
 Approaches
 Knowledge-based
 Ontologies
 Probabilistic (HMM, Bayesian Nets)
 Neural Networks, Fuzzy Logic, Genetic Algorithms
 Periods in AI
 Early period - 1950’s & 60’s
 Game playing
 brute force (calculate your way out)
 Theorem proving
 symbol manipulation
 Biological models
 neural nets
 Symbolic application period - 70’s
 Early expert systems, use of knowledge
 Commercial period - 80’s
 boom in knowledge/ rule bases
 Periods in AI cont’d
 ? period - 90’s and New Millenium
 Real-world applications, modelling, better evidence, use of
theory, ......?
 Topics: data mining, formal models, GA’s, fuzzy logic,
agents, neural nets, autonomous systems
 Applications
 visual recognition of traffic
 medical diagnosis
 directory enquiries
 power plant control
 automatic cars
 Fashions in AI
Progress goes in stages, following funding booms and crises: Some examples:
1. Machine translation of languages
1950’s to 1966 - Syntactic translators
1966 - all US funding cancelled
1980 - commercial translators available

2. Neural Networks
1943 - first AI work by McCulloch & Pitts
1950’s & 60’s - Minsky’s book on “Perceptrons” stops nearly all work on nets
1986 - rediscovery of solutions leads to massive growth in neural nets research

The UK had its own funding freeze in 1973 when the Lighthill report reduced AI work severely -
Lesson: Don’t claim too much for your discipline!!!!
Look for similar stop/go effects in fields like genetic algorithms and evolutionary computing. This is a
very active modern area dating back to the work of Friedberg in 1958.
Symbolic and Sub-symbolic AI
 Symbolic AI is concerned with describing and manipulating our
knowledge of the world as explicit symbols, where these symbols
have clear relationships to entities in the real world.
 Sub-symbolic AI (e.g. neural-nets) is more concerned with
obtaining the correct response to an input stimulus without
‘looking inside the box’ to see if parts of the mechanism can be
associated with discrete real world objects.
 This course is concerned with symbolic AI.
AI Applications
 Autonomous Planning & Scheduling:
 Autonomous rovers.
AI Applications
 Autonomous Planning & Scheduling:
 Telescope scheduling
AI Applications
 Autonomous Planning & Scheduling:
 Analysis of data:
AI Applications
 Medicine:
 Image guided surgery
AI Applications
 Medicine:
 Image analysis and enhancement
AI Applications
 Transportation:
 Autonomous
vehicle control:
AI Applications
 Transportation:
 Pedestrian detection:
AI Applications
Games:
AI Applications
 Games:
 Game Playing
 Classic Moment: May, '97: Deep Blue vs. Kasparov
 First match won against world champion
 “Intelligent creative” play
 200 million board positions per second
 Humans understood 99.9 of Deep Blue's moves
 Can do about the same now with a PC cluster

 Open question:
 How does human cognition deal with the
search space explosion of chess?
 Or: how can humans compete with computers at all??

 1996: Kasparov Beats Deep Blue

“I could feel --- I could smell --- a new kind of intelligence across the table.”

 1997: Deep Blue Beats Kasparov

“Deep Blue hasn't proven anything.”

 Huge game-playing advances recently, e.g. in Go!

Text from Bart Selman, image from IBM’s Deep Blue pages
Decision Making
 Applied AI involves many kinds of automation
 Scheduling, e.g. airline routing, military
 Route planning, e.g. Google maps
 Medical diagnosis
 Web search engines
 Spam classifiers
 Automated help desks
 Fraud detection
 Product recommendations
 … Lots more!
 Designing Rational Agents

 An agent is an entity that perceives and acts.

 A rational agent selects actions that maximize its
(expected) utility.
 Characteristics of the percepts, environment, and
action space dictate techniques for selecting rational
actions
 This course is about:
 General AI techniques for a variety of problem types

Environment
 Learning to recognize when and how a new problem Sensors

Agent
Percepts
can be solved with an existing technique
?
Actuators
Actions
 Pac-Man as an Agent

Agent Environment
Sensors Percepts

?
Actuators Actions

Pac-Man is a registered trademark of Namco-Bandai Games, used here for educational purposes Demo1: pacman-l1.mp4 or L1D2
AI Applications
 Robotic toys:
AI Applications
Other application areas:
 Bioinformatics:
 Gene expression data analysis
 Prediction of protein structure
 Text classification, document sorting:
 Web pages, e-mails
 Articles in the news
 Video, image classification
 Music composition, picture drawing
 Natural Language Processing .
 Perception.
Homework
Read Pg (1 – 31) From the book