1.

Introduction to Artificial Intelligence:

This is social networking age. We all spend a considerable amount of time online depending on our
interests and requirements. When we browse through social websites and online stores, a lot of
suggestions pop-up or slide-in in our way. Where have they come from? A programmed component of the
web site or mobile app is doing it. We are tracked right from the moment we logon to any online platform
until we logoff.
Websites and apps are programmed to track us in various ways which is called our browsing signature or
browser fingerprinting.
The analysis of enormous amount of data to produce useful patterns of visitor's browsing
habits, interests and buying preferences is called analytics.
Al is the field of conceiving, designing and developing machines which should perform tasks that usually
require human intelligence. Al is the art and science of developing machines running on intelligent
algorithms that make them capable of thinking, acting and learning like human beings.
This very nature of Al field makes it a huge umbrella of technology which covers all the domains and
application areas like Medical and health care, research and development, manufacturing, sales, travel,
education, defence etc.
The term Artificial Intelligence was first coined by Stanford researcher John McCarthy in 1956. In
plain and simple words, the ability of a machine to think and learn is called artificial intelligence.
AI is the science and engineering of making intelligent machines.
The AI field refers to the study of the principles, concepts and technology for building such machines and
systems that should think, act and learn like humans.
Machines possessing Al should be able to interact with their environment and perceive it through various
stimuli such as visual perception, speech recognition, language comprehension etc. in the form of received
data and respond to them, based on gathered intelligence.
Al Challenges:
1. Retain the facts as knowledge.
2. Recall the knowledge in a situation.
3. Think, analyse and apply logic.
4. Make useful and accurate predictions.
5. Make decisions and upgrade their intelligence algorithm themselves.
So, the biggest challenge is to develop a machine or a computer that can store knowledge and
improve its own program to solve new problems with its evolved or improved intelligence

Human Intelligence and Machines

What makes humans intelligent is their ability to reason. But what triggers reasoning? The answer is
sensing. We sense, we perceive, we receive a variety of stimuli from our surroundings and then we process
that input. This processing of what we sense is called reasoning.
A human brain senses, reasons and then, finally, acts upon it. For example, we come across an old
friend, recognise him or her and greet him or her.
We sense through our receptive organs. How should a machine sense?
A machine should first know what it is supposed to sense and then it should be able to sense (input)
images, patterns, faces, signatures, prints, textures, audio, moving images, numbers etc.
What should it sense from these? For instance, in an image of a group of people, is it supposed to sense
entire image, a face or just the background? So, sensing can be achieved by scanners and sensors.
Camera, Image scanner, sensor, speech recognition engine, fingerprint recognition program, audio sensor,
motion sensor, thermal sensor, light sensors, proximity (distance) sensor, chemical sensors, barometric
sensors are the equipment used to receive various stimuli from its surroundings.
After sensing, what to do with the stimulus (input) is entirely the problem domain of artificial
intelligence. Comparing facts and making decisions like in an Expert System. Then comes the action,
responding with voice or moving in a particular direction
Human intelligence is the combination of the following traits:
Stimuli - Sensing – Reasoning – Response
Learning:
Humans learn by - guidance and training by others or by self-paced trial and error method. They retain the
learning by practice, remembering and applying it in various situations. Machines are being developed to
learn by trial and error. Generalized learning is difficult as it demands application of learning in various
situations by using previous knowledge and experience.
Problem Solving:
In a simple situation, a machine can be programmed into looking for, finding and applying the possible
steps of solution to achieve a set goal. Such machines are useful in a specific task-
oriented environment like bottling plant, loading/ unloading of items, counting, assembling parts etc. In a
generalized situation, a machine needs to be trained into selecting the best suited approach to achieve the
goal and then retain it for future use. This is what Al is trying to achieve.

Reasoning
Logical reasoning is the distinct characteristic of human brain. Reasoning has broadly 2 types:
Deductive and Inductive. In deductive reasoning the facts are analysed and guarantee a conclusion.
For example: Raj is a non-vegetarian so he will also eat a vegetable if non-vegetarian dish is not available.
In inductive reasoning, facts only support the conclusion without any guarantee. For example:
Ram falls sick most often when he eats eggs. Ram must be allergic to eggs.
Hardest challenge in Al is to develop machines that are able to apply inductive reasoning which needs a
critical and intelligent analysis of the available facts in different scenarios or contexts on the basis of
previous experience.
Language
Learning any language is a complex Process. Language contains grammar and words — word with multiple
meanings, words with similar meanings etc. After learning the language, an endless variety of sentences
can be formed which is challenging for a machine to do. Al based voice response systems and chat bots etc.
are being developed in a restricted application area.

Keywords:
 Browsing signature/ browser fingerprinting: Pattern of a user's browsing habit.
 Analytics: Analysis of enormous amount of data to produce useful patterns.
 Algorithm: A process or logical set of rules to solve a problem or perform a calculation.
 Sensing: Perceiving an external stimulus.
 Reasoning: Thinking logically to reach a conclusion.
 Expert system: A system that compares facts and makes decisions.

2. TYPES & TECHNIQUES

AI Types: Complexity of Intelligence and Functionality
Al Types on the Basis of Complexity of lntelligence:
AI that works with limited functionality and needs some prior information to be fed in order to accomplish
a task is called weaker or narrow Al. Machines that exhibit a limited extent of intelligence to accomplish a
single, or certain simple tasks are said to have narrow/ weak intelligence. Such machines are deployed to
perform some repeated tasks.

Key features:
1. Performa dedicated assigned task.
2. Limited to a particular field of application.
3. Has a predefined set of functions.

Some examples are:

1. Performing web-wide search of content (e.g. Google search).
2. A website suggesting similar products depending on the items bought by the user earlier.
 3. Recognising face among several single image shots or group images.
4. Voice interface-based assistants such as Alexa and Siri.
5. Self-driven vehicles.
6. Any logic-based game such as a card game
7. A robot collecting empty food trays from the tables in a restaurant.

Strong or Generalized AI:

Machines with strong intelligence are capable of not only "think" like humans but also able to retain
the learning developed by the tasks accomplished. Such learned intelligence is used to solve the same
problem in different scenarios.
Key features:
Perform variety of tasks in changing context like responding to a question in different ways.
Smart like humans. Think and respond like humans do.
Capable of applying retained knowledge to solve new problems.
Examples:
1. A self-learning diagnostic system loaded with information regarding diseases and possible symptoms
along with rules to diagnose the diseases.
2. An airplane training system that functions without the help of a trainer.
3. An intelligent chat-bot that understands customer's needs & suggests solutions by its learned
intelligence.
4. A teaching robot that learns by answering students' queries and thereby enhancing its teaching skills.

Al Types on the Basis of Functionality: Classified into following types:

Reactive machines:
A reactive machine knows how to respond to a particular stimulus (input) on the basis of a set of rules and
the logic to apply those rules in all possible scenarios. They show the most basic type of Al. They do not
store learning by previous problem solving. Ex. Computers that play strategy games against humans.
Limited Memory:
Machines are equipped with the logic of sensing the changes occurring around such as an object
changing its place or position at a distance. Machines with limited memory cannot retain the learning.
Next time, in the same situation, logic to resolve the problem is executed afresh. Self-driven cars, chat bots
and customer response systems are based on limited memory functionality.
Theory of mind:
Machines are capable of interacting socially. They are able to respond suitably to others by exhibiting
understanding of their emotions and gestures.
Self-awareness:
Machines are truly intelligent machines. Such machines, of course, do not exist but endeavours to develop
machines that exhibit intelligence exactly like us humans are ongoing.
There is another form of Al called Artificial Superintelligence which is the highest form of Al using which
machines will surpass human intelligence owing to their speed, capability to process astronomical amount
of data and self-evolving smarter algorithms.
How does Artificial Intelligence Work?
Working of Al is based on enormous amount of data (heart of Al technology). For example, if a machine is
to be taught to distinguish between a cat and a dog, it needs to be run through several thousand different
images of dogs and cats so that later it uses the "learning" from these images to identify a dog or a cat.
Then, there are filters based on algorithms that determine which data is useful for processing and which is
to be discarded. Third part of the functioning is the "learning brain" which is built on the complex
algorithms. It is able to identify patterns and trends in the data to develop learning out of it.
The fields of statistics and data analytics help in this process. There are various technologies, theories and
methods and subfields on which Al is based today.
Al and Neurons:
Our nervous system contains millions of neurons. Neurons are the microscopic cells that carry
information (sensory signals and responses) throughout the nervous system and into the brain. Millions of
neurons form a communication network. Study of neurons & research has made the basis of
Connectionism — neuron like computing. Al focuses on building a network of artificial neurons.
Artificial Neural Networks (ANN):
At Massachusetts Institute of Technology (MIT), Farley and Clark developed a 128-neurons system which
could memorise simple patterns and then distinguish one pattern among many. Each neuron was able to
hold the information taught (fed repeatedly) to it. Computer program repeatedly fed various combinations
of patterns to the artificial neurons system so that they retained possible combinations of various features
of patterns. ANN has paved the foundation for machine learning.
Machine Learning (ML):
It is a sub-set of Al. It enables a computer system to learn from experience without programming it further.
This technique is used to make computer perform accurate predictions after analysing the input given to it.
Machine learning has following approaches:
Supervised ML: In this approach, the computer system with ML ability is fed with the inputs and also
informed about what prediction/output is expected. After the prediction, the new findings are stored by
the machine for doing any new predictions in future. For example, a routing software learns to find the
fastest possible route by checking the patterns in traffic data and road conditions.
Unsupervised ML and Deep Learning: In this approach, machine is fed only with input data but not
the desired output details. It uses input data to analyse and find out any useful patterns. Such machines are
considered more intelligent, and their process of data analysis is called deep learning. After prediction, the
output is stored by the machine to use it for performing any future predictions. Deep learning machines are
helpful in finding new trends and patterns in data. For example, a machine can predict which students are
likely to show improved/declined performance in next exam by analysing performance data of the students
in entire school.
Reinforcement Learning: Is based on the concept of reward and punishment.
For each failure there is a punishment which helps the machine realize that the step was wrong
while a reward helps machine learn the right step. For example, an Al system is designed to find the
shortest and fastest route between two points. For every obstacle it encounters, it learns that the
movement or turn was wrong (punishment) so next time it tries the alternate path which lets it move
further distance until another obstacle. Eventually, the learning of correct route is reinforced.
Ex. Robotics, industrial automation, supply chains, games, automated vehicles

Keywords:
Weak Al: Al that works with limited functionality.
Strong Al: Al capable of "thinking" like humans and retain learning.
Supervised ML: Predictions done with the aid of the details of the task to be done.
Unsupervised ML: Machine figures out patterns on its own.
Reinforcement Learning: Learning based on reward-punishment approach.

3. APPLICATIONS OF ARTIFICIAL INTELLIGENCE

Task Based Classification of Al:
On the basis of task an Al system can perform, Al can be broadly classified into three
— ordinary tasks, formal tasks and expert level tasks.
Ordinary tasks: These are easy to learn and perform. Generally, they include voice recognition,
speech recognition, processing graphical inputs, audio-visual inputs, language translation, automated tasks
like factory robots do.
Formal tasks: Mathematical calculations, complex scientific calculations and derivations, strategy games.
Expert level tasks: Analysis of scientific data, processing enormous amount of data and predicting trends
and patterns in various industries etc.
A machine can be developed with algorithms to learn ordinary tasks. Then this learning can be used to
learn formal tasks and finally, machine can develop algorithms to perform expert level tasks.
Applications of Al:
Education and Training:
Key applications of Al in education are:
• Smart content generation.
• Customised learning.
• Adaptive learning.
• Data visualization of performance and feedback.
• Immersive learning environments
• Smart teaching.
Machine learning is used to create smart education systems which deliver adaptive (content that adapts
according to the capacity of the learner) educational content to the learners after analysing their response
and performance.
NLP capability is harnessed in making voice driven intelligent responsive learning systems. Voice
operated Question-Answer based systems are also the outcome of NLP in education.
Integration of AI with other technologies such as Cloud can make smart education reach in remote areas.
Customer Support Systems:
Various businesses, schools, banks and public services etc. deploy customer support systems in the form of
people, computers, equipment and response systems. Al revolutionizes the customer support and response
systems using natural language processing and smart search capabilities in following ways:
Chatbots: Al-enabled chat systems are called chatbots. Al chatbots are smarter than traditional chat
engines. Speech recognition is concerned with understanding "what" is spoken while voice recognition is
concerned with "who" is speaking. For voice recognition a machine needs to be trained to identify which
person in particular is speaking.
Al enabled support systems: Customer response systems and most Interactive Voice Response Systems
(IVRS) use speech recognition as part of Natural Language Processing (NLP) feature of Al. Ex. Google
Assistant, Amazon Alexa etc
Public addressing and alert systems: Publicly addressing and guiding people in case of emergency such as
fire break out or earthquake. Also useful in home and industrial security like identifying an unauthorized
face in any area and raising an alert.
Key applications of Al in Customer support are:
1. Better customer experience.
2. Faster and accurate guidance.
3. Improved public relation.
4. Multi-language interpretation.
5. Useful for physically challenged.

Service-oriented Businesses:
Various businesses do not manufacture products but provide services. Banking, Education,
public transport & hotels are such businesses. Al systems take in the data generated by these businesses to
produce results in following useful ways:
1. Checking patterns useful to offer new services.
2. Understanding customer behaviour — if customers are liking/disliking certain services.
3. Assessing customer loyalty — Al can alert if some customers may quit using the services.
4. Assessing service quality
5. Assessing service improvement areas
6. Predicting future customer behaviour - if customers may buy or quit new more services.
7. Al-driven recruitment industry use Al for automated assessments and psychometric evaluations to
reduce time-to-hire, costs and better quality.

Product-oriented Businesses:
Businesses that produce or manufacture tangible products use Al in various stages of product development
life cycle - product planning, design, manufacturing and delivering. Ex. Automobile industry. Some popular
Al applications in product-driven businesses are:
Autonomous vehicles: Driverless taxis, drones and smart missiles.
Smart home devices: Al-driven home appliances (smart refrigerator, smart TV), home security
systems (Smart intruder alert and home maintenance systems (Al solar power, waste disposal, water filter)
Smart cities: Smart cities equipped with AI- enabled traffic control systems minimize traffic congestion and
perform smart route search, disaster prevention and alert systems, smart public transport system
Robots: Al-driven autonomous, intelligent robots enhance public assistance and can minimize threat to
human lives such as underground constructions, mining, oil-extraction, heavy-machines operations etc.

Ecommerce and retail businesses

E-commerce industry records billions of transactions and user activities daily which can only be handled by
AI- enabled systems. Amazon, Walmart, Flipkart, Alibaba and all popular online E-Commerce platforms
using AI to improve businesses and to enhance customer experience
Key operations in which AI is used in E-Commerce are:
1. Digital marketing.
2. Customer relation.
3. Product delivery, service tracking, Logistics
As user browses E-commerce website, data generated from user's clicks on the items, the details of items
purchased, location of the user are used by AI to provide better shopping experience to the user such as
items of interest, suggesting related popular items.
Social Media Platforms, News and Entertainment:
Today, online social interactions, news search and entertainment through music, movies, games and stories
are merged into an abstract form online. Online platforms use this ocean of data to promote their services
and products and to enrich user experience with innovative and quality offerings.
Entertainment, gaming and media industry thrive on subscription and viewership. Increasing viewer-base is
their growth indicator. Al can help in improving and developing content in multiple languages
easily. Al based analysis of customer preferences and choice Of entertainment sources can help customize
offerings for the customers. Some of the major applications could be:
1. Movie production (screenwriting, storyboarding, scheduling, budgeting etc.).
2. Automatic multilingual subtitles.
3. Editing and recording.
4. Targeted advertisement insertions.
5. Content (news) compilation and organization.

Public Services
Public transport: Al systems can control city transport systems such as Metro rails, taxis and buses.
Some common uses are routing of vehicles, smart traffic control minimizing congestion, smart parking,
crowd control and time management etc.
Healthcare: Decision systems for heart stroke prevention, patient risk alert systems, expert
diagnosis systems, prevention of disease outbreak and Al-assisted surgery
Demographic Trends: Population related data is immense & used in various ways. AI is used to find patterns
in population data such as poverty, hunger, homelessness, unemployment, nutrition, childbirth, education
etc. By predicting trends in demographic data, Al can help in addressing many social issues efficiently.
Environmental Data Analytics: Data related to land, agriculture, forest, rivers, water quality, roads,
mountains, air quality, weather, ocean, various ecosystems make an enormous lot of data. Al can analyse
trends in it. Enhancing living conditions for rural areas, preventing damage to environment, managing
damage due to natural calamities, improving agricultural practices

Electronics Industry:
The core of entire electronics industry is the silicon chip. Al-chip based intelligent devices, equipment,
appliances and vehicles which will function in a much better way like self-fault-diagnostic, self-alert
systems. Internet of Things (IoT). Robotics is the primary emerging field in this industry.

Research and Development:

Research is majorly concerned with browsing and compiling data then generating information in many
useful ways. Al systems are capable to process bulk data very fast and generate useful patterns and
predictions. This can help greatly in research.
1. NLP capability can help voice-based search and process audio and sound inputs.
2. Computer Vision (CV) Al helps in identifying and processing visual data from images. This can also help
in processing image-based research.
3. Machine learning algorithms can enable the self-learning systems to help in innovative design and
intelligent predictions.
4. Deep Learning can revolutionize visual search, photograph recognition etc.

Keywords:
Immersive training: Interactive learning environment simulating real-life setup to teach skills & techniques.
Virtual reality: A simulated 3D environment that seems real and user can interact with it using special
equipment such as gloves, helmet and visors fitted with sensors. VR helps in immersive learning also.
Adaptive content: Content that is delivered according to the choice and capacity of the user or learner.
Cloud: A term used for Internet-based ecosystem which allows access to software and services and data
storage online instead of having them installed on one's computer.
Data visualization: Graphical presentation (Charts/maps) of data in the form of trends and patterns.
Chatbot: Al-based interactive online chat system mostly used in customer support and enquiry.

4. DOMAINS OF ARTIFICIAL INTELLIGENCE

In every technology involving computers, the underlying concept is input-processing-output. In Al,
this can be rewritten as data-analysis-prediction.
Data makes the foundation for Al. Data is the key of the three domains of AI. 3 domains of Al
1. Data
2. Computer Vision.
3. Natural Language Processing (NLP)

Data:
Huge and constant in-flow of data for an Al system is a prerequisite. More the data, better an Al
system would be put to use. Al system does not process the data like a traditional computer does.
An Al system, more than processing, "analyses" the data and tries to identify some sort of trend or pattern
in it. So, with enormous data feed to the Al systems, businesses can harness the power of Al for business
projections, identifying problems, threats and opportunities, performance analyses, addressing
social issues etc.

Computer Vision:
One rich form of data is graphics, images and video. Processing graphics and video to recognize some
pattern in them falls under computer vision. Intelligent machines analyse enormous graphical data,
understand them for one or more patterns. One simple example could be a face captured by a CCTV
camera and the Al machine is able to match that face with a picture of a known criminal in the crime
database
So, Computer Vision simply refers to the artificial intelligence of a machine to analyse constant feed of huge
visual data, understanding various patterns in it and finally making decisions on it
For example, if a machine is fed with the images of known criminals in police records, then it must store the
distinct features on the basis of which a face is recognized. The algorithms help Al machine create patterns
of facial features. Later these patterns would help the Al map these patterns and the face descriptions in
the crime database to the faces they capture through constant feed of several CCTV cameras in a city. As
the time passes, Al machine becomes more and more "intelligent" in matching the faces accurately.

Natural Language Processing

NLP is the technology applied to enable computers to understand the languages spoken by humans. This
includes speech recognition, audio to text and text to audio conversion and processing audio data.
Google Translate, Google Assistant, Interactive Voice Response Systems (IVRS), Cortana, Siri, Alexa use NLP.
NLP works using special algorithms which convert the unstructured voice data or speech into a form that is
understood by a computer system. Using language rules, semantics and pronunciation variation rules,
algorithms try to interpret what has been spoken and then it is digitized into a form for the computer as a
detailed command to perform a task.
NLP works on two aspects— syntax and semantics of the language.
Syntax refers to the grammar of the language and sentence formations.
Semantics refers to the interpretation of the meaning carried by the words which depends on the context.

How are the three domains related to each other?

The three domains of Al are closely related. Data, in the form of visuals, is helpful in Computer Vision and
data in the form of speech, voice, audio and music, can help in natural language processing. When data is
available in the form of audio-visuals then both, Computer Vision and NLP come into play.
So, very efficient and useful applications can be designed by integrating Data, Computer Vision and NLP.