TM366: Artificial

Intelligence

1
Credit and mark distribution
8 credits course, one semester
Pre-requisite course: (M269)
One TMA (20%), one MTA(30%) and one final exam (50%)
To pass the course you have to get:
 A Minimum of 40% on the CA (TMA and MTA)
 A Minimum of 40% on the final exam
 A Minimum of 50% for the average of the CA and the final

2
Course structure
The course covers the following items:
 intelligent machines
 Symbolic intelligence
 Natural intelligence
 Neural networks
 Evolutionary computation
 Reflections

3
What is AI
Artificial intelligence attempts not just to understand but also to build
intelligent entities.
Work started after World War II, and the name itself was coined in 1956.
AI currently encompasses a huge variety of subfields, ranging from the general
(learning and perception) to the specific, such as playing chess, proving
mathematical theorems, writing
poetry, driving a car on a crowded street, and diagnosing diseases.
AI is relevant to any intellectual task; it is truly a universal field.

4
What is AI cont.

5
What is AI cont.
Acting Humanly:
The Turing Test approach: proposed by Alan Turing (1950), was designed to
provide a satisfactory operational definition of intelligence.
A computer passes the test if a human interrogator, after posing some written
questions, cannot tell whether the written responses come from a person or
from a computer.

6
What is AI cont.
Thinking Humanly:
The interdisciplinary field of cognitive science brings together computer models
from AI and experimental techniques from psychology to construct precise
and testable theories of the human mind.

7
What is AI cont.
Thinking rationally:
The Greek philosopher Aristotle was one of the first to attempt to codify “right thinking,” that
SYLLOGISM is, irrefutable reasoning processes.
His syllogisms provided patterns for argument structures that always yielded correct
conclusions when given correct premises.
for example, “Socrates is a man; all men are mortal; therefore, Socrates is mortal.”
These laws of thought were LOGIC supposed to govern the operation of the mind; their study
initiated the field called logic.

8
What is AI cont.
Acting rationally:
An agent is just something that acts (agent comes from the Latin agere, to do).
A rational agent is one that acts so as to achieve the
best outcome or, when there is uncertainty, the best expected outcome.

9
Disciplines contributing to AI
1. Philosophy
2. Mathematics
3. Economics
4. Psychology
5. Computer engineering
6. …

10
Disciplines contributing to AI
1. Philosophy
Can formal rules be used to draw valid conclusions?
How does the mind arise from a physical brain?
 Where does knowledge come from?
 How does knowledge lead to action?

11
Disciplines contributing to AI
2. Mathematics
What are the formal rules to draw valid conclusions?
 What can be computed?
How do we reason with uncertain information?

12
Disciplines contributing to AI
3. Economics
How should we make decisions so as to maximize payoff?
How should we do this when others may not go along?
How should we do this when the payoff may be far in the future?

13
Disciplines contributing to AI
4. Neuroscience
How do brains process information?

14
Disciplines contributing to AI
5. Psychology
How do humans and animals think and act?

15
Disciplines contributing to AI
6. Computer engineering
How can we build an efficient computer?

16
Disciplines contributing to AI
7. Control theory and
cybernetics
How can artifacts operate under their own control?

17
Applications of AI
Robotic vehicles
Speech recognition
Autonomous planning and scheduling
Game playing
Spam fighting
Robotics
Machine Translation

18
19
 What are the Types of Artificial
Intelligence?
Artificial intelligence technologies are categorised by their
capacity to mimic human characteristics, the technology they use
to do this, their real-world applications, and the theory of mind.
There are 3 types of artificial intelligence.
[Link] narrow intelligence (ANI)
[Link] general intelligence (AGI)
[Link] superintelligence (ASI)

20
Types of Artificial Intelligence?

21
 Types of Artificial Intelligence?
1. Artificial Narrow Intelligence (ANI)
 Artificial Narrow Intelligence (ANI), also known as weak AI or narrow AI, is
the only type of artificial intelligence we have successfully realized to date.
 Narrow AI is goal-oriented, designed to perform singular tasks such as
facial recognition, speech recognition, voice assistants, driving a car, or
searching the internet and is very intelligent at completing the specific task
it is programmed to do.
 Narrow AI has experienced numerous breakthroughs in the last decade,
powered by achievements in machine learning and deep learning.
 For example, AI systems today are used in medicine to diagnose cancer and
other diseases with extreme accuracy through replication of human-Sequa
cognition and reasoning.
22
 Types of Artificial Intelligence?
2. Artificial General Intelligence (AGI)
 Artificial General Intelligence (AGI), also known as strong AI or deep AI, is the idea
of a computer with general intelligence that can learn and use its knowledge to
solve any issue.
 In every given circumstance, AGI can think, comprehend, and behave in a manner
that is indistinguishable from that of a human. Researchers and scientists working
on artificial intelligence have yet to create robust AI.
 To succeed, they'd have to figure out how to make robots aware and program them
with a comprehensive set of cognitive skills.
 Strong AI uses a theory of mind AI framework, which refers to the ability to
discern needs, emotions, beliefs and thought processes of other intelligent
entitles.
 The theory of mind-level AI is not about replication or simulation, it’s about
training machines to truly understand humans.
23
 Types of Artificial Intelligence?
3. Artificial Super-Intelligent (ASI)
 Artificial Super-Intelligent (ASI) is a hypothetical AI that does more than replicate
or understand human intelligence and behaviour;
 ASI is when robots become self-aware and exceed human intelligence and ability.
 Artificial superintelligence is the idea that AI will grow to be so similar to human
emotions and experiences that it will not only understand them, but will also elicit
emotions, wants, beliefs, and goals of its own.
 ASI would potentially be superior at everything humans do, including math,
science, athletics, art, medicine, hobbies, emotional connections, and everything
else, in addition to mimicking human intellect.
 ASI would have a better memory and be able to process and analyse information
and stimuli more quickly. As a result, super-intelligent species' decision-making
and problem-solving abilities would be considered superior to those of humans. In
other words, Artificial Super Intelligence can learn on its own.
24
Weak versus strong AI
Weak AI: computer value is that it gives us a very
powerful tool.

Strong AI: computer is not only a tool, rather, the

25
 What is Self-learning?
Self-learning systems are artificial agents that can acquire and
renew knowledge on their own over time, without the need for
hard coding.
These are adaptive systems whose functionalities increase
through a learning process that is generally based on trial and
error, a learning model influenced by neurosciences.
A self-learning system first seeks to engage with its users or the
surrounding environment and then observes the changes
generated by its activities.

26
 What is Self-learning?
Self-learning AI systems, as they are now constructed, operate to meet pre-programmed
objectives.
When used in sensible human contexts, systems based on artificial neural network hardware
have proven the ability to outperform traditional digital operating systems.
As they are now built, these systems have demonstrated a capability to adapt to changing
environmental conditions that is sometimes superior to that of parametrically logical systems.
A self-learning system interacts with its users or surrounding environment initially by attempts
and observes the changes produced by its actions.
AI technologies like Reinforcement Learning, Inverse Reinforcement Learning, and Learning by
Demonstration are accelerating the development of such systems.
The use of this paradigm is now helping numerous application sectors such as gaming, finance,
banking, autonomous cars, healthcare, and robots.

27
What are the Types of Self-
learning AI?
There are two types of data in Machine Learning. The first is labelled data, whereas
the second is unlabelled data.
Labelled data includes both the input and output parameters in a machine-readable
form, but it takes a lot of human labour to label the data in the first place. Unlabelled
data contains only one or none of the parameters in machine-readable form. This
eliminates the need for human labour but also necessitates more sophisticated
solutions.

28
What are the Types of Self-
learning AI?

29
Models for natural systems
◦ Simulation: is a model that captures the functional connections
between inputs and outputs of the system (calculator)
◦ Replication: is a model that captures the functional connections
between inputs and outputs of the system and is based on processes
that are same as, or similar to, those of the real-world-
system(weather forecast)
◦ Emulation: is a model that captures the functional connections
between inputs and outputs of the system and is based on processes
that are same as, or similar to, those of the real-world-system and in
the same material as the natural system (ex primitive earth model)

30
Symbolic AI
is a class of AI that has the goal to construct machines that have the following
human intelligence features:
 Use of language
 Forming and using concepts
 Complex problem-solving, such as playing chess
 Learning
 Creativity

31
Symbolic AI components
Intelligent machines, two important keys of symbolic AI:
Representation: intelligent computer systems contain a model
in some logical or mathematical form, of the problem being
solved. These models are thus essentially symbolic, consisting of
logical expressions
Search: computer systems can find “intelligent” answers to
complex problems by searching among all possible answers for
the best one. The process of search will be governed by rules.

32
The Physical Symbol System
Hypothesis (PSSH)
is a position in the philosophy of artificial intelligence
formulated by Allen Newell and Herbert A. Simon.
They wrote: "A physical symbol system has the necessary and
sufficient means for general intelligent action."
Symbolic representation
Symbolic reasoning
The starting point is: expressions that both humans and computers can
reason about

33
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.

34
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.

35
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,

36
PSSH examples

37
PSSH examples

38
PSSH examples

39
Symbolic representation of
chess game
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))

40
Symbolic representation of
chess game
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.

41
Symbolic representation of
chess game
Processes: the moves that pieces are legally capable of making.

Each chess piece moves in its own particular way.

42
Symbolic representation of
chess game
Processes: the moves that pieces are legally capable of making.

43
Symbolic representation of
chess game
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 ))

44
 Symbolic AI- Expert system
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.

45
Example-expert system

46
Symbolic AI-representation
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.

47
Symbolic AI-representation
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.

48
Symbolic AI-representation
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

49
Symbolic AI-representation

50
Propositional logic Example 1
Having R, U, (R^U) ~W
Can we deduce ~W

51
Propositional logic Example 2
Having P ^ ¬ P
Can we deduce Q?

1. P & ¬ P Premise
2. P And elimination(using 1)
3. ¬P And elimination(using 1)
4. P v Q Addition(using 2)
5. Q Unit resolution(using 3,4)

52
Propositional logic Example 3
t (m v e)
s ¬e
t^s

Deduce m

53
Propositional logic Example 4
s r
(p v q)  ¬ r
¬s  (¬q r)
P

Deduce q

54
Symbolic logic
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).

55
Symbolic logic

56
Attributes of symbolic AI
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

57