AI Notes

MODULE 4

8.1 AI and Ethical Concerns

• Ethics define the discipline that deals with moral obligations and duties of humans.

• Ethics of AI deals with moral issues related to robots and intelligent systems

• It is classified into two main categories:

1. Roboethics

• Focuses on human moral behaviour

• Concerns how humans design, build, use, and treat AI systems

2. Machine Ethics

• Focuses on moral behaviour of AI systems

• Governs Artificial Moral Agents (AMAs)

• Supports ethical decision-making in Artificial General Intelligence (AGI)

• Operates within legal and social frameworks

8.1.1 Ethical Use of Artificial Intelligence

• AI systems offer a wide range of functionality for businesses, but their use raises ethical concerns.

• An ideal AI system must be unbiased, especially when using unexplainable models like deep learning and
GAN algorithms.

• AI systems must explain their decisions in regulated industries

• Financial institutions need explanations for credit decisions

• AI decisions are based on many complex variables. Such decisions are difficult to explain.

• AI systems that cannot explain decisions are called Black Box AI

• Regulating AI is difficult because the technology changes very fast.

1
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

Features of AI

8.1.2 Is AI Dangerous? Will Robots Take Over the World?

• Rapid growth of AI has raised concerns about its risks.

• Most researchers believe AI will not develop human emotions or become evil on its own.

When AI Can Be Dangerous

• When AI is programmed to cause harm (e.g., autonomous weapons).

• When AI is given a goal without clear limits, leading to unsafe actions.

Autonomous Systems Risks

• Autonomous weapons are hard to control and difficult to switch off.

• Hacked AI systems (drones, self-driving cars, hospital equipment) can cause serious damage.

Lack of Control and Responsibility

• Continuous monitoring of AI systems is necessary.

• There is no clear legal responsibility for AI actions.

• Questions arise about accountability, errors, and legal judgment of AI.

Additional Risks of AI

1. Job Automation

• AI and robots can replace repetitive and low-skilled jobs.

• Many service, BPO, and even white-collar jobs are at risk due to automation.

2
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

2. Biased Algorithms

• AI follows the Garbage-In-Garbage-Out (GIGO) principle.

• Biased data leads to biased and unfair decisions.

3. Loss of Privacy

• AI and IoT systems collect huge amounts of personal data.

• Facial recognition and surveillance threaten individual privacy.

4. Unreliable Information

• AI can generate fake news, deepfakes, and manipulated media.

• Deepfake videos are used for blackmail, misinformation, and political influence

• Even normal citizens are blackmailed with morphed images and/or videos. This is often known as
Faceswap videoblackmailing

8.1.3 Ethics in AI

• AI makes life easier, but its benefits matter only when it is ethical.

• Ethical AI follows moral principles and focuses on what is right, fair, and just.

• AI outputs and decisions must be fair and unbiased.

• AI trained on biased data gives biased results and is considered unethical.

Example: Microsoft Tay Chatbot

Characteristics of Ethical AI

• Uses unbiased data

• Uses the correct learning model

• Is continuously monitored

• Produces fair and just results

8.1.4 AI and Bias

Examples of Bias in AI

• Gender Bias in Images

• Searching for “doctor” often shows white men.

• “Nurse” usually shows women, “doctor” shows men.

• Voice Assistants

• Alexa, Siri, Google Assistant, etc., usually have female voices.

3
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

• Computer Vision

• Systems may misrecognize people of different skin colors.

• Speech Recognition

• AI often works better with American accent, struggles with Indian accents.

Definition of AI Bias

• AI bias occurs when an algorithm systematically favors or prejudices a particular group, producing
skewed results.

• Bias in AI usually comes from humans—either consciously or unconsciously—during development.

Three Basic Sources of Bias in AI Systems

• Data

• AI is only as good as the data it is trained on.

• Skewed or biased data produces biased outputs.

• Example: Voice assistants trained on white, upper-middle-class Americans struggle with Indian
accents.

• Algorithm

• Algorithms can amplify existing bias from the data.

• Example: Image search for “teacher” shows mostly women, even though men also teach—
algorithm prioritizes frequent patterns in biased data.

• People

• AI developers, engineers, and scientists can unintentionally introduce bias.

• Humans design algorithms for accuracy, but human biases can sneak into AI systems.

• Ethics and bias are ultimately a human issue, not a machine issue.

8.1.5 Towards Ethical and Trustworthy AI

• AI systems like emotional AI are used in hiring, but biased data can lead to unfair rejection, especially of
women.

• To make AI ethical and trustworthy, several steps are taken.

1. Laws and Regulations

• In 2021, the European Union introduced AI laws to protect people’s rights.

• Dangerous AI systems that harm freedom or manipulate people are banned.

• Australia also introduced AI ethics guidelines.

2. Role of Companies

4
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

• Companies build trust by:

• Training employees

• Checking risks

• Using third-party audits

3. Role of Activist Groups

• AI should be human-centric.

• Activist groups speak for people harmed by AI.

• AI should be used only if its benefits are greater than its harm.

4. Data Privacy

Data privacy is a core part of ethical AI.

• Ethical AI protects data by:

• Encrypting data

• Limiting data access

• Preventing misuse of training data

5. Diverse Teams

• Teams with people from different backgrounds help reduce bias.

• Continuous checking of AI systems ensures fairness.

• Different viewpoints help identify bias early

• Continuous monitoring of data and algorithms

Tools like Bias Analyzer helps to Detect bias, and Compare costs and benefits of mitigation strategies.

8.1.6 Why is Ethical AI Important?

• AI is used in medicine, law enforcement, recruitment, military, data privacy, and self-driving vehicles.

• AI must give accurate, clear, transparent, and ethical results.

• Wrong or biased AI decisions can:

• Violate people’s privacy

• Cause unfair decisions

• Make it difficult to find who is responsible when harm occurs

Examples of AI Bias

• US hospitals (2019): An AI system favored white patients over black patients due to biased billing data.

• Amazon hiring AI: Biased against women because more men’s resumes were used for training.
5
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

Why Ethical AI is Important

• AI credibility depends on how humans design and use it.

• AI can be used for good or harmful purposes.

• Ethical AI aims to benefit humanity and the planet.

Efforts Towards Ethical AI

• Organizations like IEEE published Ethically Aligned Design (2016).

• Governments and expert groups work to reduce harm and increase fairness.

Principles of Trustworthy AI

• AI should support human decision-making.

• AI should be safe and reliable, with backup plans.

• AI must protect data privacy and data quality.

• AI should be transparent and explainable.

• AI should use fair and unbiased data.

• AI should benefit society and the environment.

• AI must be accountable for its decisions.

• AI must ensure data security and confidentiality.

• Someone must be responsible for AI-related problems.

8.1.7 Impact of AI on Jobs

• AI is one of the most in-demand skills today.

• According to the World Economic Forum 75 million jobs may be lost and 133 million new jobs will be
created

• AI will change jobs, not completely remove them.

Fear of Job Loss

• Many people fear AI will replace humans.

• Studies say automation may replace 30% of jobs by 2030.

• Similar fear existed during the Industrial Revolution, but it created more jobs.

• Likewise, AI will remove some jobs but create new opportunities.

• How AI Affects Jobs

AI is transforming jobs by making work faster, smarter, and more efficient in many fields.
• Jobs That Will Grow
Creative jobs such as artists, doctors, and scientists will continue to grow because AI helps them work more
efficiently without replacing human creativity and thinking.

6
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

Management jobs cannot be fully replaced by AI because they require human emotions, ethical decision-
making, leadership, and strong communication skills.
Tech jobs, including programmers, data scientists, and AI engineers, will always be in high demand since AI
systems need skilled people to build, manage, and improve them.
• New Job Roles by 2030
By 2030, new job roles will emerge such as Chief Bias Officers who ensure AI systems are fair, and Data
Detectives who analyze complex data. AI Business Development Managers will help companies use AI
effectively, while AI-assisted Medical Professionals will combine medical skills with AI tools. AI Tutors
will support personalized learning, and Man–Machine Teaming Managers will manage collaboration
between humans and AI at work.

Why AI Cannot Replace Humans Completely

• AI can solve only one task at a [Link] cannot think creatively or [Link] needs huge amounts of
data, which may be Limited, Biased and Restricted due to privacy
• AI requires high computing power and energy

AI in Business

• 90% of leading companies already use AI

• Businesses using AI report higher productivity
• AI skills demand will grow by 71% in the next 5 years

Sectors Strongly Impacted by AI

• Medical
• Disease prediction
• Better diagnosis
• Automotive
• Self-driving vehicles
• Smart manufacturing
• Cybersecurity
• Detect cyber fraud and threats
• E-commerce
• Product recommendations
• Chatbots and warehouse automation
• HR
• Resume screening and hiring support
• Legal
• Document review and summarization

8.2 Artificial Intelligence as a Service (AIaaS)

• AIaaS (Artificial Intelligence as a Service) is a way of providing ready-made artificial intelligence tools
through the internet (cloud) so that companies can use AI without building their own AI systems.
• These services are provided through cloud computing, which makes AI cheap, scalable, and easy to use.
• AIaaS helps small and medium businesses use advanced AI technologies without spending a lot of money
on infrastructure or experts.

Role of Cloud Computing in AIaaS

• AI services are delivered using cloud platforms.

7
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

• Users can access AI tools anytime through the internet.

• Businesses pay only for what they use, reducing cost.

• Cloud-based AI improves business performance efficiently.

8.2.1 Factors Triggering Growth of AIaaS

AIaaS has grown rapidly due to major IT advancements. Many startups and companies are adopting AIaaS to stay
competitive.

Technological Factors

1. Availability of cloud platforms

Affordable cloud services help manage enterprise data efficiently.

2. Improved data storage technologies

Data storage has become cheaper, faster, and more reliable.

3. Growth of IoT and streaming devices

These devices generate huge amounts of data that can be analyzed for valuable insights.

4. Automated data management and analytics tools

Semi-automated and fully automated BI and analytics tools have improved business decision-making.

Business Reasons for Adopting AIaaS

1. Expensive infrastructure
Building in-house AI systems requires costly hardware and software.

2. Lack of trained AI professionals

Many organizations do not have skilled AI programmers.

3. High cost of AI implementation

Hiring AI experts is very expensive.

4. Insufficient data
Some companies do not have enough data to develop their own AI models.

8.2.2 The Growth of AIaaS

• With cloud services, AI is now accessible to many companies, allowing easy data storage and analysis.
• AIaaS lets businesses use AI and gain insights without heavy investment in resources or skilled staff.
• AIaaS is beneficial because it allows businesses to achieve key goals:
➢ Focus on its core area
➢ Reduce operational and infrastructural cost
➢ Minimize the risk of investment.
➢ Time required to deploy a solution is short.
➢ Pre-built algorithms make complex data management accessible
➢ Availability of on-demand services empower business
➢ Increase the benefits obtained by analysing data trends.
➢ Develop cost effective, flexible and transparent solution.
➢ Improve its strategic flexibility: Companies can only pay for what they actually use.
➢ Scalability is another big advantage of AIaaS: Now companies can start using AI with small projects.
If it works well, they can try more services and easily increase or decrease their usage based on their
needs.

8
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

8.2.3 Challenges of AIaaS

➢ Reduced security: Companies must share their data with cloud providers to use cloud services. Good
security is needed to keep the data safe and stop misuse.
➢ Reliance: When an organization depends on third-party services, it has to rely on them for information,
which can cause delays in getting responses.
➢ Reduced transparency: In AIaaS, companies only see the input and output, not how the system works. Since
it works like a black box, this can cause confusion about how reliable the results are.
➢ Data governance: Each company has its own data rules. These rules may stop them from using third-party
cloud services fully, so they cannot use AIaaS completely.

8.2.4 Vendors of AIaaS

➢ Many AI provider platforms offer different AI and machine learning services.

➢ Organizations can compare services and costs to choose the best option.
➢ Cloud AI providers offer GPU-based processing for heavy workloads and Providers handle hardware,
maintenance, and technical support.
➢ Major AIaaS providers are Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform
(GCP). These vendors provide bots, APIs, and machine learning frameworks.
➢ Many startups also focus on providing AIaaS [Link] companies often acquire startups to add new
AI services.
➢ Organizations select AI services based on business goals and also includes facts like company size, budget,
technical skills, and data volume.
1. MonkeyLearn is an AI platform that makes text analysis easy using simple, no-code tools. Users
can use ready-made models (like a survey analyzer) or create their own models to find sentiment,
keywords, and topics, and then connect them to apps using easy integrations or APIs.
2. IBM Watson provides a set of AI tools and ready-made applications like Watson Assistant and
Watson Natural Language Understanding for building virtual assistants and performing text
analysis. Watson Studio helps users build, train, and deploy machine learning models on any cloud,
even without deep knowledge of AI or data science.
3. Microsoft Azure is a cloud platform that provides AI and machine learning solutions. Azure
Cognitive Services adds features like computer vision and text extraction to apps, while Azure Bot
Service helps create and integrate bots into applications.
4. Google Cloud ML is Google’s AI platform that helps developers and data scientists create and
deploy machine learning projects. AutoML allows training custom models for tasks like text
analysis, image classification, and translation, with tools to visualize data and evaluate model
performance.
5. Google Cloud ML is Google’s AI platform that helps developers and data scientists create and
deploy machine learning projects.

9
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

6. AutoML allows training custom models for tasks like text analysis, image classification, and
translation, with tools to visualize data and evaluate model performance.
7. AIPaaS (AI Platform-as-a-Service) is a cloud-based solution that lets businesses use AI services
on a pay-per-use basis. It provides access to third-party APIs for complete and intelligent data
management.

8.4.1 Collaborative Systems

• Collaborative systems combine humans and AI to overcome each other’s limits and enhance abilities.

• Examples of human-AI collaboration:

1. Foldit game: Players and AI fold simulated proteins to study diseases; AI handles complex
calculations, humans use intuition.

2. Chess collaboration: Two amateur players teamed with AI to beat supercomputers and
grandmasters.

3. AI in business: In 2014, a Japanese company elected an AI system to its board of directors.

4. Military UAV drones: AI takes front-line actions while humans provide virtual support.

5. Talos exoskeleton (US military): Augments soldiers with power suit features like heating, cooling,
vital monitoring, and HUD, working alongside humans.

• Concerns: Overreliance on AI may lead to job loss and reduced human skills.

• Research insight: 1,500 companies found short-term productivity gains from AI automation; long-term
benefits occur only when humans and machines complement each other.

• Human strengths: Leadership, teamwork, creativity, social skills.

• Machine strengths: Speed, scalability, quantitative analysis.

• Outcome: AI enhances analytics, decision-making, and creativity when humans and machines collaborate
effectively.

Human Assisting machine

• Humans play a key role in human–machine collaboration by training AI, explaining AI decisions, and
ensuring responsible use.

• Laws like GDPR(General Data protection Rights) require explanations for AI-based decisions and
promote safe, ethical AI use.

• Employees continuously monitor AI systems to ensure safety, correctness, and privacy protection. (This
creates the jobs)

• Humans train AI to interact politely and empathetically with users, often with help from creative experts.

• Modern AI assistants (e.g., Koko) are designed to show empathy and support users emotionally.

8.4.2 Machines Assisting Humans

10
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

• Smart machines enhance human abilities in three ways: amplify, interact, and embody.

• AI improves human creativity, analysis, and decision-making by giving the right information at the right
time.

• AI systems interact effectively with employees, customers, and users, including home digital assistants.

• SEB bank (Sweden) uses a virtual assistant called Aida to help customers with banking tasks like opening
accounts and payments.

• Aida also understands customer emotions from voice tone and transfers complex issues to humans, learning
from those interactions.

Embodying

• Embodying AI means using AI in physical machines like robots to support human workers.

• These machines use sensors, motors, and actuators to safely recognize people and objects. Such robots are
called cobots.

• Cobots work with humans and handle repetitive, heavy, or dangerous tasks.

• This allows humans to focus on tasks that need skill and dexterity.

• Examples: Hyundai uses AI-powered exoskeletons to give workers extra strength and endurance, and
Mercedes-Benz uses cobot arms to customize cars based on customer choices.

8.4.3 Algorithmic Game Theory and Computational Social Choice

Game Theory (in AI)

• With faster internet, many people play online games like chess, poker, and solitaire.
• These games follow fixed rules, so machines need algorithms to play fairly and ethically.
• This requirement led to the development of Game Theory.
• Game theory studies decision-making in situations where one player’s choice affects others.
• It is mainly used in multi-agent situations (more than one player).
• John von Neumann is credited with inventing game theory.

Categories of Games in Game Theory

1. Cooperative vs Non-cooperative Games
o Cooperative: Players can form alliances.
o Non-cooperative: Players cannot form alliances.
2. Symmetric vs Asymmetric Games
o Symmetric: Players have the same goals (strategies may differ).
o Asymmetric: Players have different or conflicting goals.
3. Perfect vs Imperfect Information Games
o Perfect information: All players can see each other’s moves (e.g., chess).
o Imperfect information: Some moves are hidden (e.g., card games).
4. Simultaneous vs Sequential Games
o Simultaneous: Players act at the same time.
o Sequential: Players act one after another, knowing previous moves.
5. Zero-sum vs Non-zero-sum Games
o Zero-sum: One player’s gain is another’s loss.
o Non-zero-sum: Multiple players can benefit at the same time.
Nash Equilibrium
• Nash Equilibrium is a situation in a game where no player can benefit by changing their strategy alone.
11
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

• Every player is already choosing the best possible option, given what the others are doing.
• Any change in decision will not improve the outcome.
Example: Prisoner’s Dilemma
• Two criminals are arrested and cannot communicate.
• Each prisoner has two choices: Confess or Remain silent
Possible outcomes:
1. One confesses, other stays silent
o Confessor: Free
o Silent prisoner: 10 years in prison
2. Both remain silent
o Both get 1 year in prison
3. Both confess
o Both get 5 years in prison
Nash Equilibrium in this Example
• Both prisoners choose to confess.
• Even though staying silent together gives a better result, confessing is safer individually.
• No prisoner can reduce their punishment by changing their decision alone.
👉 Therefore, “both confess” is the Nash equilibrium.

Inverse Game Theory

• Game Theory: Studies a game to find the best strategy for players.
• Inverse Game Theory: Works in the opposite way.
• It designs the game based on players’ strategies and goals.
• Widely used in designing AI agent environments.
Practical Example: Generative Adversarial Networks (GANs)
• A GAN is a machine learning model where two neural networks compete with each other.
• This competition helps both models improve accuracy.
• Components of a GAN
• Generative Model (Generator)
• Learns the data distribution.
• Creates new fake samples similar to real data.
• Discriminative Model (Discriminator)
• Classifies samples as real or fake.

12
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

8.4.4 Multi-Agents Reinforcement Learning (MARL)

Reinforcement Learning (RL)
• Reinforcement Learning is a type of machine learning where agents learn by interacting with an environment.
• Agents receive rewards or punishments based on their actions.
• RL is based on the concept of Markov Decision Processes (MDPs).
Multi-Agent Scenario
• In a multi-agent system, learning depends on interactions between the agent and the environment.
• Learning also depends on interactions among multiple agents.
Challenges
• As the number of agents increases, the number of interactions grows exponentially.
• This increase in interactions makes learning more complex and difficult.
Solution – Mean Field Scenarios (MFS)
• Mean Field Scenarios assume that all agents have similar reward functions.
• Each agent responds to the average behavior of all other agents instead of individual actions.
• This approach reduces complexity and improves scalability.
Example – Traffic Management
• In traffic management, self-driving cars act as agents.
• Route conflicts occur when cars behave selfishly.
• Game theory is used to model interactions between cars.
• Nash Equilibrium represents a state where cars cooperate in a balanced way, leading to efficient traffic flow.
8.4.5 Neuromorphic Computing
Neuromorphic computing is a computing approach in which both hardware and software are designed based on the
structure and functioning of the human brain and nervous system. It combines ideas from computer science, biology,
mathematics, electronics, and physics.
Aim
• To create artificial neural systems inspired by biological brains
• To enable machines to learn, store information, and make logical decisions like humans
• To transform computers into cognition machines
Key Fact
• The human brain consumes only about 20 watts of power, which is much less than traditional computers.
How Neuromorphic Computing Works
• Traditional computing systems face a trade-off between speed and power efficiency.
• Neuromorphic systems provide fast computation with low power consumption.
• They resemble the human brain due to the following features:
1. Massively parallel processing
2. Event-driven response to environmental changes
3. Low power usage
4. High adaptability and generalization
5. Fault tolerance
Neuromorphic Architecture
• Uses artificial neurons and synapses to transmit electrical signals
• Implements Spiking Neural Networks (SNNs) using analog or digital circuitry
13
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

• Unlike traditional neural networks, SNNs process information temporally and spatially
• Memory and processing are colocated in each neuron, avoiding the Von Neumann bottleneck
• This improves speed, power efficiency, and performance
Neuromorphic Computing and AGI
• Artificial General Intelligence (AGI) refers to machines with human-level intelligence
• Traditional computers may not achieve AGI, but neuromorphic systems show promise
• Example: Human Brain Project and SpiNNaker neuromorphic supercomputer
AGI Capabilities
• Reason and make decisions in uncertain situations
• Plan, learn, and communicate using natural language
• Represent knowledge and integrate skills to achieve goals
• May include imagination, learning from experience, and self-awareness
AGI Tests
• Turing Test
• Robot College Student Test
Ethical Debate
• Whether intelligent machines should have legal or ethical rights is still debated
Human Brain vs Neuromorphic Systems
• Brain uses ~20 watts; Fugaku supercomputer uses ~28 megawatts
• Brain requires no cooling system; computers require extensive cooling
• Brain is adaptable and multifunctional
• Uses massively parallel processing and is highly fault-tolerant
Brain Function and Neuromorphic Design
• Neurons transmit signals through synapses using action potentials
• Signals can be spatial or temporal
• One neuron can connect to thousands of others
• Neuromorphic systems mimic this using SNNs and memristors
• Memristors store variable values, enabling learning like the brain
Uses of Neuromorphic Systems
• Enables on-device processing instead of cloud-based computation
• Handles noisy and uncertain data effectively
• Suitable for complex problems like constraint satisfaction
• Supports next-generation AI beyond rule-based systems
Examples of Neuromorphic Systems
• Intel Loihi: Used in artificial skin and prosthetic limbs
• IBM TrueNorth: Used in energy-efficient drones (Blue Raven project)
• Human Brain Project:
o SpiNNaker (million-core system)
o BrainScaleS and BrainScaleS-2
Challenges
• Difficult transition from Von Neumann architecture
• Requires new programming languages
• Needs new memory, storage, and sensor technologies
• Changes the way hardware and software are designed and integrated
9.1 Expert System
An expert system is a computer program that solves complex problems in a specific domain at a level comparable to
a human expert. It provides decisions using stored knowledge and reasoning rules.
• Designed for domains like medicine, science, finance, etc.
• Assists human experts rather than replacing them
• First expert system developed in the 1970s
• Performance depends on the quality and quantity of the knowledge base
• Example: Google search spelling suggestions

14
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

Working of an Expert System

• User submits a query through the User Interface
• The Inference Engine applies reasoning rules
• Knowledge is extracted from the Knowledge Base
• Output and explanation are shown to the user

Popular Examples of Expert Systems

• DENDRAL – Chemical analysis of organic molecules
• MYCIN – Diagnosis of bacterial infections and drug recommendation
• PXDES – Lung cancer detection using images
• CaDeT – Early cancer detection
• R1/XCON – Configures customized computer systems
• DXplain – Clinical decision support system
Characteristics of Expert Systems
• High performance and accuracy
• Easy to understand (human-readable input/output)
• Reliable and fast response
• Large memory capacity
• Domain-specific expertise
• Not affected by emotions
• Unbiased decision-making
• Cost-effective
• Non-perishable (permanent knowledge)
• Intelligent (uses if-then rules)
Components of an Expert System
1. User Interface
• Allows interaction between user and system
• Accepts queries and displays results
• Explains how conclusions were reached
• Easy to use and customizable

2. Inference Engine (Brain of the System)

• Applies rules to derive conclusions
• Resolves conflicts between rules
• Provides explanation and debugging
Types of Inference Engines:
15
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

• Deterministic – Uses facts and rules (certain results)

• Probabilistic – Uses probability (uncertain results)
Inference Strategies:
• Forward Chaining – Starts from facts → predicts outcomes

• Backward Chaining – Starts from goal → finds causes

3. Knowledge Base
• Stores domain knowledge from human experts
• More knowledge → better performance
Types of Knowledge:
• Factual Knowledge – Verified facts
• Heuristic Knowledge – Experience-based rules
• Knowledge represented using if-then rules
Case Study: MYCIN
1. Knowledge base created with medical expertise
2. Regular updates made
3. Patient data collected
4. Inference engine applies rules
5. Diagnosis and recommendations displayed
Participants in Expert System Development

• Expert – Provides domain knowledge

• Knowledge Engineer – Builds and structures knowledge base
• End User – Uses the system for advice or solutions
Capabilities of Expert Systems
• Advising and decision-making
• Problem solving and diagnosis
• Demonstrating devices

16
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

• Interpreting inputs and predicting results

• Explaining and justifying conclusions
• Suggesting alternative solutions
Cannot:
• Replace human decision-makers
• Use emotions or common sense
• Work with insufficient knowledge
Advantages of Expert Systems
• Consistent and reproducible results
• Fast and accurate
• Useful in risky environments
• Affordable and widely available
• Reduces human effort
• Improves decision quality
• Can explain solutions
Limitations of Expert Systems
• Dependent on correct knowledge base
• High development and maintenance cost
• Lack creativity and common sense
• Cannot learn automatically
• Domain-specific only
• Cannot handle extraordinary situations
• Follow GIGO (Garbage In, Garbage Out) principle
9.1.8 Applications of Expert Systems
• Design & Manufacturing: VLSI systems, camera lenses, automobiles.
• Knowledge Publishing: Tax advisors and other advisory systems.
• Finance: Detect fraud, suspicious activities, stock trading, loan approvals.
• Medical Diagnosis: Used extensively for diagnosing diseases and treatment.
• Planning & Scheduling: Project goals, airline scheduling, cargo, manufacturing.
• Software Development: Tracking project progress.
• Monitoring Systems: Detect leaks in pipelines, continuous data comparison.
• Process Control: Control physical processes through monitoring.
• Automobile & Electronics: Fault detection in vehicles and computers.
• Complex System Diagnosis: Electronic, electromechanical, diesel-electric systems
9.1.9 Expert System Technology
Expert System Development Environment
The development environment includes the necessary hardware and software resources required to build expert
systems.
Components:
• Hardware platforms:
o Workstations
o Minicomputers
o Mainframe computers
• Programming languages:
o LISP (LISt Processing)
o PROLOG (PROgrammation en LOGique)
• Large databases for storing and managing domain knowledge
Tools
Tools help reduce the cost and complexity of expert system development by assisting developers.
Features of Tools:
• Powerful editors for knowledge and rule creation
• Debugging capabilities with multi-window support

17
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

• Support for rapid prototyping

• In-built models for:
o Knowledge representation
o Inference design
o System modeling
Shells
• A shell is an expert system without a knowledge base.
• It provides a ready-made structure where domain knowledge can be added.
Functions Provided by Shells:
• Knowledge acquisition modules
• Inference engine
• User interface
• Explanation facility
Popular Expert System Shells:
• JESS (Java Expert System Shell)
o Provides Java API for developing expert systems
• Vidwan
o Allows knowledge encoding using IF–THEN rules
9.1.10 Development of Expert Systems
• Identify Problem Domain:
o Identify the problem to solve. Ensure it is solvable by an expert system. Collaborate with human
experts and assess cost effectiveness
• Design the System:
o Choose hardware, tools, and shells. Plan system integration with other systems and databases.
Design a framework for representing knowledge
• Develop the Prototype:
o Acquire domain knowledge from experts. Represent knowledge in If-THEN-ELSE rules. Integrate
heuristic knowledge for reasoning
• Test and Refine the Prototype:
o Test the prototype with sample cases. Work with end-users to find errors and discrepancies. Refine
the system based on test results.
• Develop and Complete the System:
o Test the system with end-users and databases. Integrate the system with other information systems.
Document the system and train users
• Maintain the System:
o Regularly review and update the knowledge base. Integrate new requirements and interfaces as
systems evolve.

18
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

9.2 Internet of Things (IoT)

• The Internet of Things (IoT) is a network of interconnected devices, objects, animals, or people with unique
identifiers that can exchange data over the Internet without human interaction.
• The concept was named in 1999, though its development began earlier.
• IoT connects people–people, people–things, and things–things.
• Devices such as heart monitor implants, biochip transponders in animals, vehicle sensors, and everyday
objects are assigned IP addresses to transmit data. The large address space of IPv6 has enabled the massive
growth of IoT.
• The first Internet appliance was a Coke machine at Carnegie Mellon University in the 1980s. Today, IoT is
widely used in healthcare, agriculture, transportation, energy, and building management. By 2025, IoT
devices are expected to reach 27 billion, generating massive data (73.1 zettabytes).
Applications of IoT
• Smart cars suggest routes and send delay notifications
19
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

• Alarm clocks trigger coffee machines

• Smart watches track activity and productivity
• Self-driving cars detect obstacles
• Smart sports equipment tracks performance
• Smart refrigerators monitor stock and expiry
• Smart ACs allow remote temperature control
Benefits and Impact
IoT reduces waste and cost, improves efficiency and productivity, enables predictive maintenance, supports
smart decision-making, and plays a key role in smart cities by reducing traffic, pollution, crime, and resource
wastage.

9.2.2 IoT Products

Many leading companies such as Honeywell, Hitachi, GE, Cisco, AT&T, Apple, Google, IBM, Microsoft, and
Oracle are actively developing IoT solutions.
Examples of IoT Products and Platforms:
• Amazon Echo (Smart Home): Uses voice assistant Alexa to perform tasks like playing music, giving
weather updates, ordering services, and more.
• Fitbit One (Wearables): Tracks steps, calories burned, floors climbed, and sleep quality; syncs with
smartphones and computers for data analysis.
• Barcelona Smart City: Implements IoT for smart parking, environmental monitoring, and improved urban
management.
• IoT Platforms:
o Amazon Web Services (AWS)
o Microsoft Azure
o IBM Watson
o Cisco IoT Cloud Connect
o Salesforce IoT Cloud
o Oracle Integrated Cloud
o GE Predix
• AT&T Connected Car: Added 1.3 million cars in Q2 2016, reaching 9.5 million connected vehicles.
9.2.3 Challenges of IoT
• Security is the biggest challenge due to billions of connected devices.
• Privacy and data protection are major concerns for individuals and organizations.
• According to the 2016 Vormetric Data Threat Report, protection of sensitive data is the top enterprise
concern.
• Many IoT devices do not encrypt data over local Wi-Fi networks, making them vulnerable.
• Hackers can target connected cars, critical infrastructure, homes, and medical devices.
• Faulty or unattended wearable medical IoT devices can be life-threatening.
Key IoT Challenges
1. Scalability: Massive data generation requires big data analytics, cloud storage, and scalable systems.
2. Interoperability: Fragmented standards prevent seamless communication among devices.
3. Lack of standardization: Poor compatibility with legacy systems limits full IoT integration.
4. Lack of government support: Insufficient regulations for device safety and data security.
5. IoT malware and ransomware: Devices can be hijacked, locked, or misused for ransom or spying.
6. Connectivity: Poor or unreliable Internet affects real-time data transmission.
7. Weak regulations: Legislators often lack deep understanding of IoT technology.
8. No clear privacy policies: Personal data may be collected and shared without consent.
9. Bandwidth strain: High volumes of text, audio, and video data overload networks and servers.
10. IoT skills gap: Shortage of trained professionals and lack of in-house expertise.
Case Study Examples
• Jeep Cherokee (2015): Hackers remotely controlled the car and disabled brakes.
• Interoperability issue: Insulin pump worked only with a specific operating system.
• Bandwidth overload: High-resolution data in factories and hospitals caused server strain.

20
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

9.2.4 Sensors
A sensor is a device that detects and responds to input from the physical environment.
Inputs include heat, light, motion, pressure, moisture, and other environmental phenomena.
The output is a signal converted into human-readable form or transmitted over a network for further processing.

Sensor Type What it Detects / Measures Common Applications

Oxygen Sensor Gasoline–oxygen ratio Vehicle emission control systems
Motion Sensor Movement or interruption of waves Security lights, automatic doors
Vision & Imaging Sensor Objects, colours, images Surveillance, machine vision
Temperature Sensor Heat / thermal parameters Industrial processes, HVAC
Radiation Sensor Alpha, beta, gamma radiation Nuclear surveys, sample counting
Proximity Sensor Nearby objects without contact Manufacturing, robotics
Pressure Sensor Force per unit area Hydraulic systems, gas pipelines
Position Sensor Position of components Valves, doors, throttles
Photoelectric Sensor Object, colour, cleanliness Automation, material handling
Particle Sensor Dust and airborne particles Pollution monitoring, baghouses
Metal Detector Presence of metal Security screening, packaging
Level Sensor Level of liquid, gas, solid Tanks, bins, reservoirs
Leak Sensor Liquid or gas leakage Vacuum packaging, pipelines
Humidity Sensor Moisture in air Weather systems, HVAC
Gas / Chemical Sensor Gas or chemical presence Safety systems, labs
Force Sensor Weight, load, torque Industrial testing, robotics
Flow Sensor Flow of liquids or gases Processing industries
Flaw Sensor Surface/material defects Welding, quality inspection
Flame Detector Fire presence and quality Burners, combustion systems
Electrical Sensor Current, voltage Power systems, electronics
Contact Sensor Physical contact Alarm systems, switches
Non-contact Sensor Object detection without touch Radar guns, speed monitoring
Speed Sensor Speed of object or vehicle Vehicles, industrial machines
Ultrasonic Sensor Distance using sound waves Parking systems, object detection

9.3 Artificial Intelligence of Things (AIoT)

• AIoT combines Artificial Intelligence (AI) with the Internet of Things (IoT) to enhance efficiency, improve
decision-making, and enable better data management.
• AI simulates human intelligence, while IoT connects devices to transfer data over networks without human
intervention.
• AIoT improves IoT by adding machine learning algorithms to enhance decision-making.
• How Does AIoT Work?AI is embedded in IoT devices and connected through IoT networks. Devices collect
data, which is analyzed using AI techniques to improve productivity and efficiency. Edge AI reduces
bandwidth needs and minimizes delays in data processing.
9.3.1 How Does AIoT Work?

• AI enhances the Analysis step in IoT, which directly impacts efficiency.

AIoT helps:

21
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

• Manage, analyze, and obtain meaningful insights from data.

• Provide fast, accurate data analysis.
• Balance localized and centralized intelligence.
• Ensure data privacy and security.
9.3.3 Applications and Examples of AIoT
• Smart Cities: Enhances operational efficiency, improves traffic management, and optimizes resource use.
• Smart Retail: Uses smart cameras to track customers and optimize store operations.
• Smart Homes: Devices analyze user habits to provide customized support.
• Manufacturing: Monitors equipment and detects malfunctions.
• Autonomous Vehicles: Uses real-time data to make driving decisions.
• Healthcare: Devices monitor patient data, enabling remote care and accurate diagnoses.
• Smart Thermostats: AI-powered devices adjust temperature based on user preferences and schedules.
• Fleet Management: Monitors vehicles, tracks fuel usage, and identifies driver behavior.
9.3.4 Benefits and Challenges of AIoT
Benefits:
• Increases operational efficiency by analyzing and adjusting system operations in real-time.
• Enables on-the-fly decision-making with real-time data analysis.
• Reduces workload and saves time by automating data analysis.
• Scalable systems that connect more devices as needed.
• Better risk management by predicting potential problems in advance.
• Improves products and services through real-time data analysis.
Challenges:
• Security: Protection of sensitive data is a significant concern.
• Scalability: Managing large volumes of data from multiple devices.
• Interoperability: Lack of standardization among different IoT devices.
• Privacy: Handling user data and ensuring privacy compliance.
9.3.5 Future of AIoT
• AIoTGrowth with 5G: 5G will boost AIoT by providing faster data transfer and lower latency.
• Revolutionizing Industries: AIoT will improve supply chains, operational efficiency, and decision-making.
• Exponential Data Growth: With more devices connected, AI will help manage the massive amount of data
generated.
By 2025, the number of IoT devices is expected to rise to 42 billion, and AI will help process and analyze the increasing
data efficiently.

22
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga
 AI Notes

23
BY,
Anusha prima Fernandes,Dept. of ISE,
JNNCE,Shimoga