MIRPUR UNIVERSITY OF SCIENCE AND TECHNOLOGY

DEPARTMENT OF SOFTWARE ENGINEERING

 Artificial Intelligence
(Introduction to Artificial Intelligence)

(lecture #2)

Engr. Abdul Qadir Khan

(Lecturer)

Monday, March 24, 2025

 LECTURE CONTENTS

• History of AI
• The State of the Art
• Intelligent agents
• The concept of Rationality

Artificial Intelligence Engr. Abdul Qadir Khan 3

 History of AI

1943-1955: Early Work (McCulloch & Pitts, Turing)McCulloch & Pitts (1943):
Developed the first artificial neuron model, laying the foundation for neural
networks.
They showed that any computable function could be performed by a network of
neurons, and all logical operations (like AND, OR, NOT) could be done by simple
neuron networks.
They also suggested that these networks
could learn over time, which laid the foundation
for Hebbian learning (a rule for modifying the
strength of connections between neurons),
introduced by Donald Hebb in 1949.

Artificial Intelligence Engr. Abdul Qadir Khan 4

 History of AI

• Alan Turing (1950): Proposed the Turing Test to measure a machine's ability to
exhibit intelligent behavior. His work inspired early AI thinking
• He also proposed important concepts like machine learning, genetic algorithms,
and reinforcement learning.
• One of his ideas was the Child Program, where he suggested that instead of
mimicking the adult human mind, it might be better to try creating a program that
learns like a child, developing over time.

Artificial Intelligence Engr. Abdul Qadir Khan 5

 History of AI

The Birth of Artificial Intelligence (1956)

• In 1956, John McCarthy played a key role in starting Artificial Intelligence (AI)
as an official field. After working at Princeton and Stanford, McCarthy moved to
Dartmouth College and organized a workshop in the summer of 1956. This
workshop is where AI was officially born.
Dartmouth Conference (1956)
• McCarthy invited top scientists, including Marvin Minsky, Claude Shannon, and
Nathaniel Rochester, to work together on AI. The main idea was to see if
machines could learn, use language, solve problems, and improve themselves, just
like humans. They believed that by working together for a summer, they could
make real progress on these big challenges.

Artificial Intelligence Engr. Abdul Qadir Khan 6

 History of AI
Key Developments in Early AI (1952–1969)
1.General Problem Solver (GPS):
1. Newell and Simon created a program called GPS in the 1960s. It was designed to mimic human
problem-solving. The GPS was able to solve problems in a similar way to humans, showing that
machines could think logically like humans.
John McCarthy's Contributions (1958):
• LISP: McCarthy created the LISP programming language in 1958, which became very popular in AI
development for many years.
• Advice Taker: McCarthy also introduced the Advice Taker, a program that could use general
knowledge to solve problems and improve itself by adding new information.
Microworlds and AI Applications:
In the 1960s, AI researchers started focusing on microworlds—small, controlled environments where AI
could solve specific tasks:
• SAINT Program: Solved simple calculus problems.
• STUDENT Program: Solved algebra problems.

Artificial Intelligence Engr. Abdul Qadir Khan 7

 History of AI

Knowledge-Based Systems: Key Developments (1969–1979)

• AI shifted focus to knowledge-based systems that used specialized knowledge to solve
problems more effectively. This period saw the rise of knowledge-based systems using
expert rules and reasoning to improve problem-solving.
DENDRAL:
• AI program that identified molecular structures using mass spectrometer data.
• Used expert knowledge (rules) for faster, more accurate solutions.
• First successful knowledge-based system
MYCIN:
• Medical AI system for diagnosing blood infections.
• Used 450 rules to diagnose and performed as well as expert doctors.
• Incorporated uncertainty (certainty factors) in diagnoses.
Artificial Intelligence Engr. Abdul Qadir Khan 8
 History of AI

AI Becomes an Industry (1980–Present)

• R1 Expert System (1982): The first successful commercial expert system, R1,
helped configure orders for new computer systems at Digital Equipment
Corporation (DEC). By 1986, it saved the company $40 million annually.
• Japan's Fifth Generation Project (1981): Japan's 10-year plan to create
intelligent computers using Prolog but these ambitious projects didn't meet their
goals.
• AI Boom: By 1988, the AI industry grew from millions to billions of dollars, with
companies developing expert systems, vision systems, robots, and specialized
software. However, this success was followed by the "AI Winter," where many
companies failed to meet expectations.

Artificial Intelligence Engr. Abdul Qadir Khan 9

 History of AI

The Return of Neural Networks (1986–Present)

• Back-Propagation: In the 1980s, the back-propagation algorithm was revived,
becoming a key tool for training neural networks and solving many learning
problems.
• Modern Neural Networks: Research split into improving architectures and
algorithms, and modeling biological neurons, contributing to areas like pattern
recognition and machine learning.

Artificial Intelligence Engr. Abdul Qadir Khan10

 History of AI
The Emergence of Intelligent Agents (1995–Present)
• Intelligent Agents: Focus shifted to creating agents like SOAR that solve broader problems.
The internet became a key area for AI, with applications like search engines and
recommender systems.
• AI and Other Disciplines: AI integrated with fields like control theory and economics,
leading to advances in robotics, such as self-driving cars.
1.3.10 The Availability of Very Large Data Sets (2001–Present)
• Big Data: With massive datasets available, AI moved from focusing on algorithms to
leveraging large data for better performance in tasks like word-sense disambiguation and
image filling.
• Data-Driven Learning: AI shows improved performance with more data, even using basic
algorithms.
• AI Spring: With advancements in machine learning, AI is now integrated into industries
like healthcare, finance, and consumer services, marking a new era after the "AI Winter."

Artificial Intelligence Engr. Abdul Qadir Khan11

 Class Activity

• Create groups of maximum 3 students and have each group research and
present how AI is applied in different industries today (e.g., healthcare, finance,
robotics, etc.).
• You should include examples such as AI in diagnostics, fraud detection, or
autonomous driving.

Artificial Intelligence Engr. Abdul Qadir Khan 12

 The State of the Art

Robotic Vehicles: Autonomous vehicles like STANLEY and BOSS demonstrated advanced
capabilities in navigating complex terrains.
Speech Recognition: Automated systems, like those used by United Airlines, enabled natural
language interaction.
Autonomous Planning and Scheduling: NASA's Remote Agent program demonstrated autonomous
planning for spacecraft operations.
Natural Language Processing (NLP):
Language Understanding: AI systems, like chatbots and virtual assistants, have improved natural
language understanding, enabling more sophisticated interactions with users.
Language Translation: Advanced language translation models, such as those using transformer
architectures, have achieved remarkable accuracy.

Artificial Intelligence Engr. Abdul Qadir Khan 13

 The State of the Art

Computer Vision:
Object Recognition: Deep learning models have significantly improved object recognition in images
and videos, contributing to applications in autonomous vehicles, security systems, and more.
Facial Recognition: Facial recognition technology has become more prevalent in security,
authentication, and personalized user experiences.
Machine Learning and Deep Learning:
Deep Learning Advances: Breakthroughs in deep learning, especially with convolutional neural
networks (CNNs) and recurrent neural networks (RNNs), have led to improved performance in
various tasks, from image recognition to natural language processing.
Transfer Learning: Techniques like transfer learning have enabled models to leverage knowledge
gained from one task to perform well in related tasks, reducing the need for extensive training data.

Artificial Intelligence Engr. Abdul Qadir Khan 14

 The State of the Art
Reinforcement Learning:
Game Playing: Reinforcement learning algorithms have demonstrated exceptional performance in playing
complex games, including board games like Go and video games.
AI in Healthcare:
Medical Diagnosis: AI systems, particularly deep learning models, are making strides in medical image
analysis for diagnosing conditions such as cancer from radiological images.
Drug Discovery: AI is being used to accelerate drug discovery processes by predicting potential drug
candidates and their effects.
Autonomous Systems:
Drones and Robotics: AI-powered drones and robots are increasingly used for tasks such as surveillance,
delivery, and exploration.

Autonomous Vehicles: Ongoing advancements in self-driving car technology have the potential to transform
transportation

Artificial Intelligence Engr. Abdul Qadir Khan 15

 AI Applications Shaping the Modern World

Search Engines and Information Retrieval (Google): Google, founded in 1998, transformed online
search using AI algorithms to deliver precise and relevant search results.
E-commerce Recommendations (Amazon, Netflix): AI-powered recommendation systems,
implemented by platforms like Amazon and Netflix, revolutionized online shopping and content
consumption by providing personalized suggestions.
Speech Recognition and Virtual Assistants (Siri, Alexa, Google Assistant): The advent of virtual
assistants, including Siri, Alexa, and Google Assistant, introduced natural language processing and
voice recognition into everyday devices for user interaction.
Image and Facial Recognition (Facebook, Security Systems): AI-driven image and facial recognition
technologies, exemplified by Facebook's photo tagging and security applications, became integral for
identity verification and personalized experiences.
Autonomous Vehicles (Tesla): The automotive industry embraced AI for self-driving cars, with Tesla
incorporating machine learning algorithms to enable advanced driver-assistance features, paving the
way for autonomous vehicles.

Artificial Intelligence Engr. Abdul Qadir Khan

 Intelligent Agents
•An agent is anything that can be viewed as perceiving its
environment through sensors and acting upon that environment
through actuators.
•The human agent has eyes, ears, and other organs for sensors
hands, legs, and vocal tract for actuators.
•A robotic agent has cameras and infrared range finders for
sensors- various motors for actuators
•A software agent receives keystrokes, file contents, and network
packets as sensory inputs and acts on the environment by
displaying on the screen, writing files, and sending network
packets

Artificial Intelligence Engr. Abdul Qadir Khan 17

 Intelligent Agents

Percept :
A percept is what an agent perceives or senses at a specific moment. For example, if
a robot has a camera and detects a red ball in its view, the visual data of the red
ball is its percept.
Percept Sequence
A percept sequence is the entire history of an agent's sensory inputs. Consider a
smart vacuum cleaner. If, over time, it has sensed and recorded data about different
rooms, objects, and obstacles in its environment, the collective record of all these
observations forms its percept sequence.

Artificial Intelligence Engr. Abdul Qadir Khan 18

 Intelligent Agents

Percept :
A percept is what an agent perceives or senses at a specific moment. For example, if
a robot has a camera and detects a red ball in its view, the visual data of the red
ball is its percept.
Percept Sequence
A percept sequence is the entire history of an agent's sensory inputs. Consider a
smart vacuum cleaner. If, over time, it has sensed and recorded data about different
rooms, objects, and obstacles in its environment, the collective record of all these
observations forms its percept sequence.

Artificial Intelligence Engr. Abdul Qadir Khan 19

 Agent Function Vs Agent Program
Agent Function:
An agent's behavior is described by the agent function. This function maps any given
sequence of sensory inputs (percept sequence) to a specific action.
Example: Consider a thermostat in a room. The agent function could be a set of rules
specifying how the thermostat adjusts the temperature based on the current temperature and
the desired setting.
Agent Program:
The artificial agent's function is implemented by an agent program. The program is a
tangible, executable set of instructions that runs within a physical system.
Example: The agent program is the embedded software in the thermostat device that reads
the current temperature, compares it to the desired setting, and controls the heating or
cooling system accordingly.
In summary, the agent function is an abstract, mathematical concept, while the agent
program is the concrete implementation of that concept within a physical system.

Artificial Intelligence Engr. Abdul Qadir Khan 20

 Agent Function

Intelligent Agents

Agent Program

21
Artificial Intelligence Engr. Abdul Qadir Khan
 The concept of Rationality

Information Gathering:
The process of performing actions to modify future percepts, is crucial for rational decision-
making. The process of collecting data or knowledge to understand a particular topic or
situation better.
Example: A weather prediction model gathers data from various sensors to improve its
forecasts. Groundwork for neural networks.
Exploration:
The act of navigating and investigating an initially unknown environment to gain information.
Involves actively searching, investigating, or discovering new places, ideas, or possibilities.
Example: A vacuum-cleaning robot exploring a new room to understand its layout and
identify the obstacle
.

Artificial Intelligence Engr. Abdul Qadir Khan

 The concept of Rationality

Learning:
The ability of an agent to acquire knowledge or adapt based on experience and perceptions.
The process of acquiring new knowledge, skills, or understanding through study,
experience, or being taught.
Example: A language translation algorithm improves its accuracy over time by learning
from user corrections.
Environment:
The external surroundings or conditions in which an agent operates and perceives.
Example: An autonomous car responds to changes in traffic, weather, and road conditions
to navigate safely.

Artificial Intelligence Engr. Abdul Qadir Khan

 The concept of Rationality

Autonomy:
Autonomy is the ability to make independent decisions and take actions without constant
supervision or control from others.
Smart Home Devices:
Without Autonomy: Turning on lights, adjusting the thermostat, or locking doors typically
requires manual input from the homeowner.
With Autonomy: Smart home systems can learn from user preferences and automate tasks,
such as adjusting the temperature based on daily patterns or turning off lights when a room
is unoccupied

Artificial Intelligence Engr. Abdul Qadir Khan

 The concept of Rationality

We have learnt that we should design agents that "act rational”

• How do we define 'acting rationally ' so we can write programs?
• Should we consider the environment where the agent will be deployed?
Rational Agent: (in general)
A rational agent is one that does the right thing (not definition)
Every entry in the table for the agent function must be filled in correctly

Artificial Intelligence Engr. Abdul Qadir Khan 25

 The concept of Rationality

We are interested in environment states not agent states.

We should not define success in terms of the agent's opinion
The agent could achieve perfect rationality simply by deluding itself that its
performance was perfect
Human agents, for example, are notorious for sour grapes"- believing that they did not
want something after not getting it.
Performance Measure:
The consequence of an agent's behavior is assessed by observing the sequence of states
the environment goes through due to the agent's actions. The desirability of this
sequence is measured using a performance measure, which evaluates the effectiveness
or success of the agent in achieving its goals based on the resulting environment states.

Artificial Intelligence Engr. Abdul Qadir Khan 26

 Design Performance Measure with Environment Focus

It is better to design 'performance measures' according to what one

wants in the environment, rather than according to how one thinks the
agent should behave
For example, for the vacuum agent, we propose to measure the
performance by the 'amount of dirt cleaned up in a single eight-hour
shift.
"What is wrong with this performance measure?

Artificial Intelligence Engr. Abdul Qadir Khan 27

 Rationality

Rationality
What is rational at any given time depends on four things:
• The performance measure that defines the criterion of success.
• The agent’s prior knowledge of the environment.
• The actions that the agent can perform.
• The agent’s percept sequence to date.
Definition
For each possible percept sequence, a rational agent should select an action that is
expected to maximize its performance measure, given the evidence provided by the
percept sequence and whatever built-in knowledge the agent has.

Artificial Intelligence Engr. Abdul Qadir Khan

 Rationality

- Let's assume the following for the vacuum-cleaner example:

Performance measure - award one point for each clean square at each time step,
over a lifetime of1000 time steps
Prior knowledge - the agent knows that there are two squares but it does not know
the dirt distribution
Agent actions - available actions are left, right, and suck
Percept sequence - agent correctly perceives its location and whether that location
contains dirt-
Here is what the vacuum-cleaner agent does:- Cleans a square if it is dirty and
moves to the other square if not.
Is the vacuum cleaner agent rational? Is performance measure robust?

Artificial Intelligence Engr. Abdul Qadir Khan

 Omniscient agents

Omniscient agent
is like a super-smart being who knows everything, including the actual outcome of
its actions. In reality, achieving omniscience is impossible.
• For instance, consider a situation where you decide to cross the street, not
knowing that a cargo door will fall from a passing plane, causing an accident.
• An omniscient agent would have known this in advance and acted differently.
• The example illustrates that while rationality aims for the best decisions given
what's known, perfection would require predicting unforeseeable events, which is
not feasible. So, in practical terms, we aim for rational decisions, acknowledging
the inherent limitations in predicting the future.

Artificial Intelligence Engr. Abdul Qadir Khan 30

 References

• Stuart Russell & Peter Norvig, Artificial Intelligence: A Modern Approach, 3rd
Edition

Artificial Intelligence Engr. Abdul Qadir Khan 31

THANKS