DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

I/11 Semester
Course: Introduction to AI and Applications

Course Code: 25ET15/25ET25

Credits: 03 & 2025 Scheme

Module 4: Trends in AI

Course Instructor
Ms. Divyashree S
Assistant Professor, Dept. of CSE,
EPCET
[Link]@[Link]

Department of Computer Science & Engineering, EPCET

 MODULE 4: Trends in AI
AI and Ethical Concerns, AI as a Service (AIaaS), Recent trends in AI,
Expert System, Internet of Things, Artificial Intelligence of Things (AIoT).
AI and Ethical Concerns
Ethics is the discipline that deals with the moral obligations and duties of humans.
Similarly, the ethics of Artificial Intelligence (AI) focuses on ethical issues related to robots
and artificially intelligent beings.
AI ethics can be broadly classified into two categories:
1. Roboethics
• Concerned with the moral behaviour of humans
• Focuses on how humans design, construct, use, and treat AI systems and robots
• Emphasizes responsibility, safety, and ethical use of AI
2. Machine Ethics
• Deals with the moral behaviour of machines
• Focuses on Artificial Moral Agents (AMAs)
• Ensures ethical decision-making by machines in Artificial General Intelligence (AGI)
• Operates within existing legal and social frameworks

Ethical Use of Artificial Intelligence

AI systems provide extensive functionality for businesses, but their use raises serious ethical
concerns.
Key Ethical Issues in AI

[Link] CSE, EPCET pg. 1

 • Bias in AI:
An ideal AI system must be unbiased. Bias is a major concern, especially in systems
using deep learning and Generative Adversarial Networks (GANs), which are often
difficult to interpret.
• Explainability (Black Box AI):
Many AI systems cannot explain how decisions are made. Such systems are known as
black box AI.
This is problematic in regulated industries like finance, where decisions (e.g., credit
approval or rejection) must be explained.
• Regulatory Compliance:
Financial institutions and other regulated sectors require AI systems to provide
transparent and justifiable decisions, which black box models often fail to do.
Features of a Responsible / Ethical AI System
A responsible AI system should be:
• Explainable
• Monitorable
• Reproducible
• Secure
• Human-centered
• Unbiased
• Justifiable
Legal and Privacy Challenges
• Formulating AI-specific laws is difficult due to the use of multiple and rapidly evolving
technologies.
• Strict laws may hinder innovation, while outdated laws may fail to address new
challenges.
• Existing privacy laws do not fully cover voice assistants such as Amazon Alexa and
Apple Siri, which collect data to improve learning models.
• Such data collection can be misused by malicious actors, posing security and privacy
risks.
Is AI Dangerous? Will Robots Take Over the World?
• Rapid advancement of Artificial Intelligence (AI) has increased concerns about its risks.
• Common fear: AI may replace humans or take control of the world.
• Researchers believe super-intelligent AI is unlikely to develop human emotions or
evil intentions on its own.
Conditions Under Which AI Becomes Dangerous
AI poses risks mainly in the following situations:
a) AI Programmed for Destructive Purposes
• Some AI systems are intentionally designed to cause harm.
• Autonomous weapons are AI systems programmed to kill. If misused or controlled by
malicious people, they may cause: AI-driven wars, Mass casualties, Loss of human
control
• Such systems are difficult to shut down once activated.

[Link] CSE, EPCET pg. 2

Examples:
• Weaponized drones with facial recognition
• Hacked self-driving vehicles
• Internet-connected medical equipment being misused
b) Incorrect or Incomplete Goal Definition
• AI may be assigned beneficial tasks without clear constraints.
• Example: A self-driving car instructed to reach a destination “as fast as possible” may
ignore safety rules.
Highlights the need for: Clear goal definitions, Safety constraints, Continuous
monitoring and control
Legal and Ethical Challenges
• No clear legal responsibility for AI actions.
• Key unanswered questions:
1. Should AI be judged like humans?
2. Who is responsible for autonomous, self-learning AI?
3. Is any error acceptable if AI causes fatal consequences?

1. The immediate risk posed by job automation( Job Automation and Employment Risks)
• AI and Robotic Process Automation (RPA) replace repetitive and predictable jobs.
• About 30% of low-skilled service and BPO jobs may be lost.
• Studies indicate:
o Millions of workers face job displacement
o White-collar jobs are also at risk
o Vulnerable communities may be affected the most
2. Biased Algorithms
• AI follows GIGO (Garbage In, Garbage Out) principle.
• Biased or poor-quality data leads to biased AI decisions.
• Can cause discrimination in:
o Hiring
o Law enforcement
o Advertising and public services
3. Too little Privacy Concerns
• AI and IoT systems collect vast amounts of personal data.
• Continuous data collection reduces individual privacy.
• Facial recognition systems can identify:
o Age, gender, ethnicity
o Gestures and emotions
• Users often lack control over how their data is used.
4. Everything becomes unreliable (Misinformation and Deepfakes)
• AI can generate fake text, images, videos, and voices.
• Deepfake technology enables:
o Fake celebrity videos
o Blackmail and harassment

[Link] CSE, EPCET pg. 3

 o Political manipulation
• Example: Cambridge Analytica misused Facebook data to influence elections.

Ethics in AI

• Ethics in AI refers to the application of moral principles in the design, development,

and use of artificial intelligence systems.
• Ethical AI ensures that AI systems are fair, just, transparent, and beneficial to
society.
• AI systems can become unethical if they are trained on biased data, leading to
discriminatory or unfair outcomes.
• Ethical AI aims to minimize bias, protect privacy, and ensure accountability in AI
decisions.
• Example: Microsoft’s Tay chatbot (2016) became offensive after learning biased
content from users, showing the risks of unethical AI.
• An ethical AI system uses unbiased data, appropriate models, and continuous
monitoring to ensure correct and fair results.
• Ethics in AI is essential to build trust, prevent misuse, and ensure responsible AI
adoption.

AI and Bias
• AI systems are increasingly used in daily life, but they often reflect social biases.
• Bias in AI can be observed in areas related to gender, race, colour, caste, religion, and
nationality.
Examples of AI Bias
• Image searches for “doctor” mostly show white men.
• “Doctor” is often portrayed as male, while “nurse” is portrayed as female.
• Virtual assistants like Alexa, Siri, and Google Assistant commonly use female voices.
• Computer vision systems often make errors in recognizing people with darker skin
tones.
• AI bias occurs when an algorithm produces results that are systematically prejudiced
toward a particular group.
• Bias leads to skewed and unfair outputs.
• Since AI is created by humans, human preferences and prejudices can unknowingly
become part of AI systems.

Sources of Bias in AI Systems

There are three main sources of bias in AI:
1. Data Bias
• AI systems are only as good as the data they are trained on.
• Biased or incomplete data results in biased outputs.
• Machines cannot distinguish between correct and incorrect training data.

[Link] CSE, EPCET pg. 4

 Example:
• Voice assistants often understand American accents better than Indian accents because
the training data mostly contains speech samples from white, upper-middle-class
Americans.

2. Algorithmic Bias
• Algorithms can amplify biases present in training data.
• Example: Image search results for “teacher” show more women due to biased training
data.
• To maximize accuracy, algorithms may ignore minority patterns, reinforcing
stereotypes.
• This leads to gender bias embedded in AI algorithms.

3. Human Bias
• Developers, engineers, and data feeders are a major source of bias.
• Human decisions during data collection, labeling, and model design introduce
conscious or unconscious bias.
• Hence, bias and ethical issues originate from humans, not machines.

Towards Ethical and Trustworthy AI

Emotional AI in Hiring
• Algorithm-based emotional AI platforms are used to reduce cost and time in
recruitment.
• These systems rely on historical data to make hiring decisions.
• In male-dominated sectors, biased historical data can lead to unfair outcomes.
• Women applicants may be disproportionately rejected due to learned bias.
• This highlights the risk of assuming AI systems are neutral or inherently fair.
Regulating a More Ethical AI
• Governments are introducing legal and ethical frameworks to govern AI use.
• In April 2021, the European Commission introduced its first legal framework for AI.
• The framework aims to protect safety and fundamental rights of people and businesses.
• A risk-based approach classifies AI systems based on potential harm.
• High-risk AI systems must follow strict requirements.
• Certain AI systems are outright banned if they threaten safety, livelihood, or rights.
• Banned uses include manipulation of human behaviour, undermining free will, and
government-led social scoring.
• In June 2021, Australia launched a similar AI ethics framework.
• The Australian framework guides responsible design, development, and use of AI
across sectors.
Company and Organizational Engagement
• Organizations are actively working to build trust in AI technologies.
• Cultural and educational programmes are used to promote ethical AI awareness.
• Risk assessments help identify potential ethical and social impacts.

[Link] CSE, EPCET pg. 5

 • Third-party audits improve transparency and accountability.
• These efforts support prevention, detection, and response mechanisms.
• Ethical AI frameworks are being treated similarly to anti-corruption and tax-evasion
controls.

Rights and Activist Groups

• The interaction between people and AI is continuously evolving.
• Ethical progress requires AI systems to remain human-centric.
• Civil rights and activist groups play a key role in questioning AI practices.
• These groups amplify the voices of communities negatively affected by technology.
• Society must question whether a technology is truly necessary.
• AI systems should only be adopted when benefits clearly outweigh potential harm.
• It is important to ask who benefits from AI and who bears the risks.
Ensuring Data Privacy
• Protecting data is a core requirement of ethical AI systems.
• Encryption is used to secure sensitive data.
• AI models are restricted from learning beyond their authorized datasets.
• Secure computation ensures even developers cannot access raw data.
• Measures are taken to prevent reverse engineering of AI models.
• These steps reduce the risk of data leakage and misuse.
• Data ethics and privacy are now integral to trustworthy AI design.
Diversifying AI Teams
• Bias often arises from limited perspectives during system development.
• AI outputs affect people from diverse social, gender, racial, and economic backgrounds.
• Building diverse teams helps identify and reduce hidden biases.
• Teams should include data scientists, business leaders, and domain experts.
• Professionals from law, sociology, ethics, and accounting add critical viewpoints.
• Continuous monitoring of data and algorithms is necessary to ensure fairness.
• Bias detection tools, such as Bias Analyzer, support automated evaluation.
• These tools help assess trade-offs, costs, and benefits of bias mitigation strategies.

Why is Ethical AI Important?

Ethical AI refers to the design, development, and deployment of Artificial Intelligence systems
that align with moral values, societal norms, and ethical standards. Since AI is increasingly
used in high-impact domains such as medicine, law enforcement, recruitment, data privacy,
military defense, and self-driving vehicles, ensuring ethical behavior in AI systems has become
mandatory.

Need for Ethical AI

AI systems must produce accurate, transparent, understandable, and fair outcomes. If AI
systems are biased or incorrect, they can lead to serious consequences, such as:
• Violation of privacy due to misuse of personal data
• Unfair or discriminatory decisions

[Link] CSE, EPCET pg. 6

 • Lack of accountability, making it difficult to assign responsibility when harm occurs
• Decisions that humans cannot understand or challenge
Thus, ethical AI is essential to maintain trust, safety, and fairness in society.

Risks of Unethical AI
• AI systems may reinforce existing social biases present in data
• They can be deliberately or unintentionally misused for harmful purposes
• There is no guarantee that all AI systems will be used solely for the benefit of humanity
The credibility of AI systems depends not only on accuracy and lack of bias, but also on how
responsibly they are developed and used.

Real-World Examples of Bias in AI

1. Healthcare Bias (2019, USA)
o An algorithm used in US hospitals to predict patients needing extra care was
found to favor white patients over black patients.
o The bias arose because white patients historically spent more on healthcare,
which the algorithm incorrectly treated as a sign of greater medical need.
2. Amazon Hiring Algorithm
o Amazon’s recruitment AI was biased against women.
o This happened because the training data had more male applicants, causing the
system to associate suitability with male candidates.
These examples highlight how biased data can lead to unethical and discriminatory AI
decisions.

Efforts Toward Ethical AI

Several organizations and expert groups are actively working to ensure AI systems do more
good than harm.
• IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems
o Released Ethically Aligned Design (2016)
o Provides guiding standards for developing and governing ethical AI systems
The primary aim of ethical AI is to ensure that AI contributes positively to humanity and
the planet.

Principles of Trustworthy AI
The European Commission’s High-Level Expert Group on AI proposed key principles for
building trustworthy AI systems:
o Human Autonomy: AI should support human decision-making, not replace or override
it.
o Technical Robustness and Safety: AI systems must be reliable, secure, and have fallback
mechanisms in case of failure.
o Privacy and Data Governance: User data must be protected, and data quality and
integrity must be maintained.

[Link] CSE, EPCET pg. 7

 o Transparency and Explainability: AI systems should be clear and understandable about
their data sources and models.
o Fairness and Non-Discrimination: AI should be trained on diverse, unbiased, and fair
data to avoid discrimination.
o Societal and Environmental Wellbeing:AI systems should benefit everyone and promote
sustainability and environmental responsibility.
o Accountability: AI systems must ensure accurate and unbiased outcomes, with clear
responsibility for decisions.
o Data Security and Confidentiality: Data must be collected, used, stored, and managed
safely and responsibly.
o Responsibility for Decisions: There must be identifiable individuals or organizations
accountable for AI-related issues.

Impact of AI on Jobs

Artificial Intelligence (AI) is one of the most talked-about and in-demand technologies today.
It is transforming industries, redefining job roles, and reshaping the global workforce.
According to the World Economic Forum, AI and Machine Learning are expected to displace
75 million jobs, while simultaneously creating 133 million new jobs in the near future.

Fear of Job Loss Due to AI

Popular science-fiction movies and media often portray AI as a threat that may replace human
workers entirely. As AI technologies advance, many tasks that were traditionally performed by
humans have now become automated, leading to fears of large-scale unemployment.
However, such fears are largely exaggerated. AI is expected to cause a paradigm shift, not
total job elimination.

Automation and Workforce Transformation

A two-year study by the McKinsey Global Institute predicts that by 2030:
• Intelligent agents and robots may replace up to 30% of current human labor
• Between 400 and 800 million jobs could be displaced
• Around 375 million workers may need to switch job categories
Despite this, AI-driven automation is expected to transform jobs rather than eliminate them
completely.

Historical Perspective: Industrial Revolution

The impact of AI on jobs is similar to the Industrial Revolution, during which machines
replaced many manual tasks. Although people initially feared unemployment, the revolution
eventually created more job opportunities and new industries.
Likewise, AI will make certain professions obsolete, but it will also lead to the emergence of
new job roles and industries.

Jobs That Will Grow with AI

Creative Jobs

[Link] CSE, EPCET pg. 8

 • Creative professions such as artists, doctors, and scientists will be enhanced by AI
tools
• AI will improve efficiency, reduce complexity, and assist professionals rather than
replace them
• AI is unlikely to make humans obsolete in the near future
Management Jobs
• Management jobs cannot be fully replaced by AI
• Managing requires emotional intelligence, ethics, communication, and decision-
making
• Human managers will supervise and control AI systems to enhance productivity
Tech Jobs
• High demand for programmers, data scientists, data analysts, big data engineers,
and AI developers
• Some current tech jobs may decline, while new and more advanced roles will emerge
• Learning AI skills provides a significant career advantage

Emerging Job Roles by 2030

Some new AI-related job titles expected to emerge include:
• Chief Bias Officer
• Data Detective
• Man–Machine Teaming Manager
• AI Business Development Manager
• AI-Assisted Medical Professional
• AI Tutor

Growth of AI in Industry
• AI adoption in businesses has increased by 270% over the last four years
• AI promotes innovation rather than human obsolescence
• It has the potential to create new industries and employment opportunities

Limitations of AI
Despite its capabilities, AI has several limitations:
• Can solve only one specific problem at a time
• Lacks general intelligence and abstract thinking
• Depends heavily on large volumes of quality data
• Biased or incomplete data leads to inaccurate results
• Data privacy and security issues limit data availability
• Requires enormous computational power and energy
o Example: Powering a large AI model may cost around $4.6 million in
electricity
Humans and AI: A Collaborative Future

[Link] CSE, EPCET pg. 9

The future is not about “humans vs. computers”, but about “humans and computers
working together”. AI will augment human capabilities and drive productivity across
industries.
• 90% of leading businesses have already invested in AI
• Over 50% of companies using AI report increased productivity
• Demand for AI and ML skills is expected to rise by 71% in the next five years
Sector-Wise Impact of AI
1. Medical Sector
• Use of AI for diagnosis and predictive healthcare models
• Helps physicians analyze large medical datasets
• Potential to eliminate diseases and reduce global poverty
2. Automotive Sector
• Autonomous vehicles and navigation systems
• AI-driven manufacturing and quality control
3. Cybersecurity and Finance
• Cyber fraud increased by 600% during the pandemic
• AI helps detect threats, predict attacks, and prevent fraud
• Widely used in financial fraud detection
4. E-Commerce
• Chatbots for customer support
• Product recommendations and targeted advertising
• Inventory, warehouse, and logistics automation
5. Human Resource (HR) Applications
• AI filters up to 75% of unsuitable resumes
• Reduces recruiter workload and hiring time
• Recruiters earlier spent up to 23 hours per successful hire reviewing resumes
6. Legal Profession
• AI assists in document classification, summarization, discovery, and knowledge
extraction
• Improves efficiency and accuracy in legal support services

AI as a Service (AIaaS)
Artificial Intelligence as a Service (AIaaS) refers to the delivery of AI capabilities through
cloud-based platforms. Instead of building complex and costly in-house AI systems,
organizations can use ready-made AI tools, APIs, and platforms on a pay-per-use basis.
AIaaS leverages cloud computing to provide scalable, cost-effective AI solutions, enabling
even small and medium enterprises (SMEs) to benefit from artificial intelligence.

Key Characteristics of AIaaS

• Cloud-based delivery of AI services
• Off-the-shelf tools and pre-built models
• Pay-as-you-use pricing model
• Minimal infrastructure and expertise required

[Link] CSE, EPCET pg. 10

 • Rapid deployment and scalability

Factors Triggering the Growth of AIaaS

1. Availability of affordable cloud platforms
o Cloud providers offer flexible and low-cost data storage and processing.
2. Advancements in data storage technologies
o Data storage has become cheaper, scalable, and more reliable.
3. Rise of IoT and streaming devices
o Massive data generation from sensors and devices enables data-driven insights.
4. Automated data management and BI tools
o Semi-automated analytics and Business Intelligence tools have transformed
business analysis.
Business Factors
1. Expensive infrastructure for building in-house AI systems
2. Lack of skilled AI professionals
3. High implementation cost charged by AI experts
4. Insufficient data availability in many organizations

Growth and Benefits of AIaaS

With cloud services becoming highly accessible, AIaaS has enabled organizations to use AI
without heavy investment in talent or hardware.
Major Benefits
1. Focus on core business activities
o Organizations need not become ML experts.
2. Reduced operational and infrastructure cost
o No need for expensive GPUs and parallel computing systems.
3. Lower investment risk
o Pay only for what is used.
4. Faster deployment time
o Ready-to-use services reduce development time.
5. Pre-built algorithms and models
o Makes AI accessible even to small businesses.
6. On-demand services
o Improves competitive advantage.
7. Better insights from data analytics
o Enhances decision-making.
8. Cost-effective, flexible, and transparent solutions
o Uses templates, drag-and-drop tools, and pre-trained models.
9. Strategic flexibility
o Ideal for organizations with sporadic computing needs.
10. High scalability
o Services can be scaled up or down as per project requirements.
Challenges of AIaaS

[Link] CSE, EPCET pg. 11

 1. Reduced security
o Data must be shared with third-party cloud vendors, increasing security risks.
2. Reliance on service providers
o Dependence on third parties may cause delays and reduced control.
3. Reduced transparency
o AIaaS often works as a black box; internal algorithms are not visible.
4. Data governance issues
o Regulatory and organizational data policies may restrict cloud usage.

Vendors of AIaaS
Organizations choose AIaaS providers based on:
• Business goals
• Budget
• Technical expertise
• Data processing requirements
Major AIaaS Vendors
• Amazon Web Services (AWS)
• Microsoft Azure
• Google Cloud Platform (GCP)
• IBM Watson
These platforms offer:
• AI APIs
• Bots and virtual assistants
• Machine learning frameworks
• GPU-based processing

Examples of AIaaS Platforms

MonkeyLearn
• No-code text analysis platform
• Sentiment analysis, keyword extraction, topic modeling
• API and app integrations
IBM Watson
• Watson Assistant (chatbots)
• Watson Natural Language Understanding
• Watson Studio for model development and deployment
Microsoft Azure
• Azure Cognitive Services (vision, speech, NLP APIs)
• Azure Bot Service
Google Cloud ML
• AI tools for big data analytics
• AutoML for custom ML models
• Visualization and performance analysis tools
AIPaaS (AI Platform as a Service)

[Link] CSE, EPCET pg. 12

 • End-to-end AI solution hosted on the cloud
• Offers complete intelligent data management
• Pay-per-use or pay-per-service model
• Integrates third-party AI APIs

Recent Trends in Artificial Intelligence

Recent years have seen rapid advances in Artificial Intelligence due to improvements in
computing power, data availability, cloud platforms, and algorithms. Extensive research is
ongoing in several emerging areas that aim to make AI more collaborative, intelligent,
adaptive, and human-like.

Collaborative Systems
Collaborative systems focus on effective human–machine partnership, where AI systems
and humans complement each other by overcoming individual limitations and enhancing
combined performance.
Key Idea
• Humans contribute creativity, intuition, leadership, and social skills
• Machines contribute speed, scalability, accuracy, and quantitative analysis
Examples of Human–Machine Collaboration
1. Foldit Game
o Humans and AI collaborate to fold proteins.
o AI performs algorithmic operations; humans apply intuition and imagination.
o Helps understand protein structures related to human diseases.
2. Chess Collaboration
o Two amateur players assisted by an AI system defeated grandmasters and
supercomputers.
3. AI on Corporate Board
o In 2014, a Japanese firm (Knowledge Ventures) appointed an AI system to its
board of directors.
4. Military Applications
o UAV drones operate on front lines with humans providing remote support.
o TALOS exoskeleton augments soldiers with strength, monitoring, and HUD
systems.
Impact on Jobs
• Research on 1,500 companies shows:
o Automation alone gives short-term productivity gains.
o Best results occur when humans and machines collaborate.

Humans Assisting Machines

Humans play three critical roles in collaborative AI systems:
1. Training AI systems
2. Explaining AI decisions

[Link] CSE, EPCET pg. 13

 o Especially important when outcomes are controversial or counterintuitive.
o Example: GDPR mandates explanation for algorithmic decisions (e.g., loan
approval).
3. Ensuring responsible and ethical AI use
o Prevent misuse, bias, and harm.
o GDPR compliance alone is expected to create ~75,000 new jobs.
Monitoring and Governance
• Continuous human monitoring ensures:
o Safety (e.g., self-driving cars avoiding obstacles)
o Privacy protection (e.g., user data handling by tech companies)
Human-Centered AI Interaction
• AI systems are trained to:
o Be confident, caring, and helpful
o Avoid being bossy or intrusive
• Example:
o Koko (MIT Media Lab) – an empathetic AI assistant that responds emotionally
to users.

Machines Assisting Humans

AI systems enhance human capabilities in three major ways:
1. Amplify
• Improves creativity, analytics, and decision-making.
• Provides the right information at the right time.
2. Interact
• Virtual assistants and chatbots engage with customers and employees.
• Example:
o Aida (SEB Bank, Sweden)
▪ Handles banking queries
▪ Detects customer emotions from voice tone
▪ Learns from unresolved cases escalated to humans
3. Embody (Cobots and Exoskeletons)
• AI embedded in physical machines to assist humans.
Examples
• Cobots
o Work alongside humans in factories.
o Handle repetitive, heavy, or dangerous tasks.
• Exoskeletons
o Used by Hyundai and the US military.
o Enhance endurance and strength.
• Mercedes-Benz
o Uses robotic arms to customize cars based on real-time customer choices.

Algorithmic Game Theory and Computational Social Choice

[Link] CSE, EPCET pg. 14

With increased internet connectivity and online gaming, especially during the pandemic,
games such as chess, checkers, poker, and solitaire became extremely popular. These games
follow clearly defined rules, which created the need to embed algorithms in machines so that
they can play ethically and rationally. This requirement led to the development of Game
Theory.
In multi-agent environments, where multiple players or agents interact, game theory helps in
selecting optimal strategies while considering that one player’s decision affects others and
vice versa.
John von Neumann is credited as the founder of Game Theory.

Types of Games in Game Theory

Game theory can be understood using the following five classifications:
1. Cooperative vs Non-Cooperative Games
• Cooperative games: Players can form alliances to improve outcomes.
• Non-cooperative games: No alliances are allowed; players act independently.
2. Symmetric vs Asymmetric Games
• Symmetric games: All players have the same goals and payoff structure.
• Asymmetric games: Players have different or conflicting goals.
3. Perfect vs Imperfect Information Games
• Perfect information games: All players can observe all moves (e.g., chess).
• Imperfect information games: Some information is hidden (e.g., card games).
4. Simultaneous vs Sequential Games
• Simultaneous games: Players act at the same time.
• Sequential games: Players act in turns, knowing previous actions.
5. Zero-Sum vs Non-Zero-Sum Games
• Zero-sum games: One player’s gain equals another’s loss.
• Non-zero-sum games: Multiple players can benefit simultaneously.

Nash Equilibrium
A Nash Equilibrium is a state in which no player can improve their outcome by unilaterally
changing their strategy, assuming other players keep their strategies unchanged.
• Once reached, no player has an incentive to deviate.
• Represents a stable solution.
Example: Prisoner’s Dilemma
• Two prisoners are isolated and cannot communicate.
• Outcomes:
o Both confess → 5 years each
o Neither confess → 1 year each
o One confesses → goes free, other gets 10 years
• Nash equilibrium occurs when both confess, even
though mutual silence is collectively better.

Inverse Game Theory

[Link] CSE, EPCET pg. 15

 • Traditional game theory analyzes outcomes based on game rules.
• Inverse game theory works in reverse:
o Designs games based on players’ strategies, goals, and behaviors.
• Plays a crucial role in designing AI agent environments.

Practical Example: Generative Adversarial Networks (GANs)

A Generative Adversarial Network (GAN) is inspired by game theory.
Structure of GAN
• Generator
o Creates fake data samples
resembling real data.
• Discriminator
o Classifies samples as real or
fake.
Game-Theoretic Perspective
• Generator and discriminator compete with each other.
• Each model tries to improve its strategy.
• Training continues until a Nash equilibrium is reached.
Applications
• Image generation
• Game playing
• Identifying real vs fake data
Nash equilibrium is easier to achieve in symmetric games, but real-world problems are usually
asymmetric.

Multi-Agent Reinforcement Learning (MARL)

Reinforcement Learning (RL) is a learning paradigm where an agent learns by interacting
with an environment using a reward and punishment mechanism. Traditional RL is based
on Markov Decision Processes (MDPs), assuming a stationary and stochastic environment,
where a single agent eventually converges to an optimal or satisfactory policy.

However, when multiple agents operate in the same environment, the assumptions of
standard RL break down. The environment becomes non-stationary, since each agent’s action
affects the learning experience of other agents.
This leads to Multi-Agent Reinforcement Learning (MARL).

Key Characteristics of MARL

• Learning depends on:
o Interaction between agent and environment
o Interaction among multiple agents
• The number of possible interactions grows exponentially with the number of agents.
• Designing optimal strategies becomes computationally complex.

[Link] CSE, EPCET pg. 16

Challenges in MARL
1. Non-stationarity
o The environment keeps changing as agents learn simultaneously.
2. Scalability
o Adding more agents increases the state–action space exponentially.
3. Coordination and Conflict
o Agents may compete or collaborate, leading to conflicts in decision-making.

Mean Field Scenarios (MFS)

To reduce complexity, MARL models often use Mean Field Scenarios (MFS).
• Assumption: All agents have similar reward functions
• Each agent reacts to the average behaviour of other agents rather than modelling each
individually.
• This approximation reduces computational overhead and improves scalability.

Example: Traffic Management with Self-Driving Cars

• AI-powered autonomous vehicles interact efficiently with the environment.
• Complications arise when:
o Multiple cars select routes independently.
o Conflicts occur when several cars choose the same “optimal” path.
Game Theory Perspective
• Each car is treated as a player.
• Their routing decisions form a game.
• Nash Equilibrium represents a stable traffic flow where:
o No car can improve its travel time by changing routes unilaterally.
• This balances competition and cooperation among agents.

Applications of MARL
• Autonomous vehicles
• Traffic signal optimization
• Robotics swarms
• Multi-agent games
• Distributed control systems

Neuromorphic Computing
Neuromorphic computing is a computing paradigm where hardware and software are
modelled after the human brain and nervous system. It integrates ideas from:
• Computer science
• Biology
• Mathematics
• Electronics
• Physics
The goal is to build cognition machines that can:

[Link] CSE, EPCET pg. 17

 • Learn
• Retain information
• Adapt
• Make logical decisions like humans

Motivation
• The human brain consumes only ~20 watts of power
• Traditional supercomputers require megawatts of energy
• Neuromorphic systems aim to deliver high performance with low power
consumption

Key Features of Neuromorphic Systems

1. Massively Parallel
o Handle many tasks simultaneously.
2. Event-Driven
o Respond dynamically to changes in the environment.
3. Low Power Consumption
o Much more energy-efficient than traditional systems.
4. Adaptive and Flexible
o Capable of learning and generalization.
5. Fault-Tolerant
o Continue functioning even if some components fail.

Spiking Neural Networks (SNN)

• Neuromorphic systems use Spiking Neural Networks (SNNs).
• Information is transmitted as electrical spikes, similar to neurons.
• Unlike traditional neural networks:
o They use temporal and spatial signals
o They are not limited to binary (0/1) outputs.
Neurons and Synapses
• Neurons process information.
• Synapses control signal strength.
• Synaptic strength varies, mimicking learning in the human brain.

Overcoming the Von Neumann Bottleneck

Traditional computers follow Von Neumann architecture, with:
• Separate memory and processing units
• Constant data movement → energy and time inefficiency
Neuromorphic Approach
• Memory and processing are collocated at each neuron
• Eliminates frequent data transfer
• Improves speed and energy efficiency

[Link] CSE, EPCET pg. 18

Neuromorphic Computing and Artificial General Intelligence (AGI)
AGI refers to machines with human-level intelligence, capable of:
• Reasoning under uncertainty
• Planning
• Learning
• Natural language communication
• Knowledge representation
• Goal-oriented behaviour
Neuromorphic computing is seen as a promising path toward AGI.
Example
• Human Brain Project (HBP)
• Neuromorphic supercomputer SpiNNaker
• Simulates large-scale neural activity

Comparison: Human Brain vs Computers

Aspect Human Brain Traditional Computers
Power consumption ~20 watts Megawatts
Architecture Massively parallel Mostly serial
Adaptability Very high Limited
Fault tolerance High Low

Real-World Neuromorphic Systems

• Intel Loihi
o Used in Pohoiki Beach system
o Millions of neurons and synapses
o Applications: artificial skin, prosthetics
• IBM TrueNorth
o 64 million neurons, 16 billion synapses
o Used in energy-efficient AI and drones
• SpiNNaker & BrainScaleS
o Outcomes of EU Human Brain Project
o Large-scale neuromorphic supercomputers

Challenges of Neuromorphic Computing

1. Shift from Von Neumann architecture is complex
2. New programming languages required
3. Specialized memory and sensors needed
4. Rethinking hardware–software integration

Expert System
• An Expert System is a computer program that solves complex problems in a specific
domain with human-like intelligence and expertise.

[Link] CSE, EPCET pg. 19

 • It is designed to make decisions similar to a human expert.
• Expert systems are domain-specific, such as medicine, science, engineering, etc.
• The system analyses user queries and uses reasoning and inference rules to extract
knowledge from the knowledge base.
• The first expert system was developed in the 1970s.
• Expert systems use both facts and heuristics (rules of thumb) for decision-making.
• The performance of an expert system depends on the amount and quality of
knowledge stored in its knowledge base.
• Google spelling suggestions are a common real-world example of an expert system.
• Expert systems are not meant to replace human experts but to assist humans in
making complex decisions.
• Main components:
1. User Interface
2. Knowledge Base
3. Inference Engine

Working of an Expert System

1. The non-expert user enters a query or problem through the User Interface.
2. The Inference Engine (Rules Engine) receives this input and applies logical IF–
THEN rules.
3. These rules are matched with facts stored in the Knowledge Base, which contains
expert knowledge.
4. The inference engine reasons using this knowledge to derive a solution or decision.
5. The final advice or solution is sent back to the user through the User Interface.

Popular Examples of Expert Systems

DENDRAL
• Chemical analysis expert system
• Identifies unknown organic molecules
• Uses mass spectrometry data and chemistry knowledge base
MYCIN
• Medical expert system using backward chaining
• Diagnoses blood infections and clotting disorders
• Recommends antibiotics and drug dosages
PXDES
• Medical imaging expert system

[Link] CSE, EPCET pg. 20

 • Determines type and level of lung cancer
• Analyses upper-body images
CaDeT
• Diagnostic support expert system
• Helps in early detection of cancer
R1 / XCON
• Computer configuration expert system
• Selects hardware/software based on user requirements
• Commercially successful system
DXplain
• Clinical decision support system
• Suggests possible diseases based on symptoms and findings

Characteristics of an Expert System

1. High performance: Solves complex domain problems accurately and efficiently
2. Understandable: Uses human-readable input and output
3. Reliable: Produces consistent and accurate results
4. Highly responsive: Gives quick solutions, faster than human experts
5. No memory limitations: Stores and processes large amounts of data
6. Domain expertise: Combines knowledge of multiple human experts
7. Emotion-free: Not affected by fatigue or stress
8. Unbiased: Decisions based purely on facts and rules
9. Cost-effective: Reduces cost of expert consultation
10. Non-perishable: Knowledge remains available permanently
11. Intelligent: Uses inference engine and IF–THEN rules for reasoning

Components of an Expert System

An expert system mainly consists of three components:
User Interface, Inference Engine, and Knowledge Base.

1. User Interface
• Acts as a link between the user and the expert
system.
• Allows users to enter queries in a readable format.
• Displays the final solution or advice to the user.
• Designed to be simple and user-friendly, even for
non-experts.
• Provides explanations of how conclusions are derived.
Features of a good user interface:
• Helps users achieve goals quickly
• Easy to customize
• Efficient use of user input

[Link] CSE, EPCET pg. 21

2. Inference Engine (Rules Engine)
• Known as the brain of the expert system.
• Applies IF–THEN rules to the knowledge base to derive conclusions.
• Resolves conflicts when multiple rules apply.
• Can include explanation and debugging capabilities.
Types of Inference Engines
• Deterministic:
Uses facts and rules; conclusions are assumed to be true.
• Probabilistic:
Uses probability; conclusions may involve uncertainty.
Inference Strategies
• Forward Chaining:
❖ A data-driven reasoning approach.
❖ Starts with known facts and rules.
❖ Applies inference rules to derive new conclusions, which are added to the fact base.
❖ Continues until no new facts can be derived or a conclusion is reached.
❖ Answers the question: “What can happen next?”
❖ Example: Predicting stock prices after changes in interest rates.

• Backward Chaining:
❖ A goal-driven reasoning approach.
❖ Starts with a goal or hypothesis and works backward to verify it.
❖ Checks whether known facts satisfy the conditions of rules leading to the goal.
❖ Used to identify causes or reasons for an event.
❖ Answers the question: “Why did this happen?”
❖ Example: Diagnosis of blood cancer in humans.

3. Knowledge Base
• Stores domain-specific knowledge obtained from human experts.
• Performance improves with more accurate and rich knowledge.
• Consists of:
o Factual Knowledge: Based on verified facts
o Heuristic Knowledge: Based on experience and judgment
• Knowledge is represented using IF–THEN rules.

[Link] CSE, EPCET pg. 22

 • Knowledge is extracted, organized, and structured with expert guidance.

Example: MYCIN Expert System (in brief)

• Knowledge base contains medical knowledge about bacterial infections.
• Patient details and symptoms are input.
• Inference engine applies rules to diagnose disease.
• Result is displayed through the user interface.
Participants in the Development of Expert System

Expert
• An expert is a human specialist with deep knowledge and experience in a specific
domain.
• The performance of an expert system depends on the quality of knowledge provided
by experts.
• Experts supply facts, rules, and heuristics to build the knowledge base.

Knowledge Engineer
• Knowledge engineering is the process of building an expert system.
• A knowledge engineer collects knowledge from domain experts.
• Converts expert knowledge into a form suitable for the knowledge base.
• Ensures the system has sufficient knowledge to solve problems effectively.

End User
• The end user is the person who uses the expert system.
• May or may not have expertise in the problem domain.
• Uses the system to obtain solutions or advice for complex queries.

Capabilities of Expert Systems

An expert system can:
1. Advise users on simple or complex domain-specific problems
2. Support decision-making in fields like medicine, finance, and engineering
3. Demonstrate devices by explaining features and usage
4. Solve complex problems efficiently
5. Explain problems in detail
6. Interpret user input correctly
7. Predict results using past data
8. Assist humans in decision-making
9. Diagnose diseases, especially in medical science

[Link] CSE, EPCET pg. 23

 10. Justify conclusions derived from facts
11. Suggest alternative solutions

Expert systems cannot:

• Replace human decision-makers
• Possess human intelligence or emotions
• Work correctly with an inadequate knowledge base
• Refine their knowledge automatically

Advantages of Expert Systems

1. Provide reproducible and consistent results
2. Can work in risky or unsafe environments
3. Give accurate results with updated knowledge
4. Not affected by fatigue or emotions
5. Easily available due to software replication
6. Cost-effective compared to human experts
7. Work at high speed, reducing human effort
8. Can explain reasoning behind solutions
9. Improve decision quality
10. Preserve and utilize scarce expert knowledge

Limitations of Expert Systems

1. Incorrect output if knowledge base is wrong or outdated
2. Lack creativity and common sense
3. High development and maintenance cost
4. Knowledge acquisition is difficult
5. Depend completely on human experts
6. Cannot learn automatically
7. Require large amounts of accurate knowledge
8. Domain-specific only
9. Fail in extraordinary situations
10. Follow GIGO principle (Garbage In, Garbage Out)
11. Do not have emotions

Applications of Expert Systems

1. Designing and manufacturing (VLSI, automobiles, lenses)
2. Advisory systems (tax advisor, financial advisor)
3. Fraud detection and stock market analysis
4. Medical diagnosis and treatment
5. Planning and scheduling tasks
6. Project tracking and monitoring
7. Pipeline leakage detection
8. Process control systems

[Link] CSE, EPCET pg. 24

 9. Fault diagnosis in vehicles and computers
10. Chemical compound identification
11. Crop damage and result prediction
12. Virus detection in computer systems
13. Repair and maintenance
14. Warehouse optimization
15. Configuration of manufactured products

Expert System Technology

1. Development Environment
• Hardware: Workstations, minicomputers, mainframes
• Languages: LISP, PROLOG
• Large databases
2. Tools
• Powerful editors
• Debugging support
• Rapid prototyping
• Built-in models for knowledge representation and inference
3. Shells
• An expert system without a knowledge base
• Provides inference engine, UI, explanation facility
• Examples:
o JESS – Java-based expert system shell
o Vidwan – Uses IF–THEN rule-based knowledge encoding

Development of Expert Systems

The development of an expert system is an iterative process, meaning the steps are repeated
and refined until the desired performance is achieved.

Steps in Developing an Expert System

1. Identify the Problem Domain
• Identify the problem to be solved and check whether it is suitable for an expert system.
• Ensure the problem is knowledge-intensive and solvable using rules.
• Identify human experts in the domain.
• Analyse cost-effectiveness of developing the system.
• Knowledge engineers and domain experts work together to define system requirements.

2. Design the System

• Select appropriate hardware, software, tools, and expert system shells.
• Decide how the system will integrate with existing databases and systems.
• Design a framework for knowledge representation and reasoning.

3. Develop the Prototype

[Link] CSE, EPCET pg. 25

 • Knowledge engineer acquires knowledge from domain experts.
• Represent knowledge using IF–THEN–ELSE rules.
• Incorporate heuristic knowledge into reasoning.
• Decide on the type of explanation facility required.

4. Test and Refine the Prototype

• Test the system using sample and real-life cases.
• Perform testing with end users to detect errors and inconsistencies.
• Analyse performance and refine rules, knowledge, and reasoning strategies.

5. Develop and Complete the System

• Integrate the expert system with users, databases, and other information systems.
• Ensure smooth interaction and correct functionality.
• Prepare documentation and train users for effective use.

6. Maintain the System

• Regularly update and review the knowledge base.
• Modify the system based on new user requirements.
• Ensure compatibility with evolving technologies and interfaces.

[Link] CSE, EPCET pg. 26

Internet of Things (IoT)
The Internet of Things (IoT) refers to a system of interconnected physical objects
(“things”) such as devices, machines, animals, or people that are embedded with sensors,
software, and unique identifiers, enabling them to collect and exchange data over the
Internet without human intervention.

Examples of IoT Applications

1. Smart transportation: Cars suggest best routes and send delay notifications.
2. Smart home: Alarm triggers coffee machine; AC controlled via smartphone.
3. Wearables: Smartwatches track activity and productivity.
4. Healthcare: Sensors and implants monitor health parameters.
5. Self-driving cars: Sensors detect objects and obstacles.
6. Smart appliances: Refrigerators notify expiry dates or shortages.
7. Sports analytics: Smart football tracks speed and distance.
8. Smart cities: Traffic control, pollution monitoring, smart parking.

Benefits of IoT
• Improves process efficiency and productivity
• Enables real-time monitoring and tracking
• Reduces cost, waste, and losses
• Supports data-driven decision making
• Helps in predictive maintenance
• Enhances resource utilization in smart cities

IoT Products and Platforms

Products
• Amazon Echo (Alexa): Smart home voice assistant
• Fitbit: Wearable fitness tracker
• Connected cars (AT&T): Vehicle connectivity
IoT Platforms
• Amazon Web Services (AWS)
• Microsoft Azure
• IBM Watson

[Link] CSE, EPCET pg. 27

 • Cisco IoT Cloud
• Oracle Integrated Cloud
• GE Predix

Challenges of IoT

• Scalability
Billions of connected devices generate huge volumes of data. Systems must support big
data analytics and cloud storage while scaling efficiently as data and devices keep
increasing.

• Interoperability
Different vendors use different technologies and protocols. Lack of a common framework
makes it difficult for IoT devices to communicate with each other smoothly.

• Lack of Standardization
Absence of universal standards limits compatibility between new IoT systems and legacy
devices, preventing fully connected and interoperable smart environments.

• Lack of Government Support

Insufficient regulations and standard committees for IoT safety, security, and data
protection slow down adoption and create legal uncertainties.

• IoT Malware and Ransomware

IoT devices are vulnerable to cyberattacks. Hackers can lock devices, steal data, or
demand ransom, posing serious security and privacy risks.

• Connectivity Issues
Reliable internet connectivity is essential. Poor or unstable networks affect real-time data
transmission and system performance.

• Ineffective Government Regulations

Many policymakers lack deep understanding of IoT, resulting in weak, outdated, or
unenforced regulations.

• No Clear Privacy Policies

There are unclear rules on what data is private or public. Devices may collect and share
personal data without proper user consent.

• Bandwidth Strain
Large numbers of IoT devices transmitting data (text, audio, video) increase bandwidth
demand and overload servers and networks.

• IoT Professional Skills Gap

Rapid IoT growth has created a shortage of skilled professionals. Many organizations
lack expertise in IoT deployment, management, and security.

[Link] CSE, EPCET pg. 28

Sensors in IoT
A sensor detects changes in the physical environment and converts them into signals for
processing.
Common Types of Sensors
• Temperature sensors: Measure heat
• Motion sensors: Detect movement
• Pressure sensors: Measure force per unit area
• Proximity sensors: Detect nearby objects
• Humidity sensors: Measure moisture in air
• Gas & chemical sensors: Detect gases
• Vision & imaging sensors: Detect objects and colors
• Light / photoelectric sensors: Detect light or object presence
• Level sensors: Measure liquid or solid levels
• Flow sensors: Detect movement of fluids
• Speed sensors: Measure object speed
• Ultrasonic sensors: Measure distance using sound waves
• Contact sensors: Detect physical touch
• Non-contact sensors: Detect objects without physical contact

Artificial Intelligence of Things (AIoT)

1. Introduction to AIoT
• AIoT is the integration of Artificial Intelligence (AI) with the Internet of Things
(IoT).
• It enhances IoT systems by enabling intelligent data analysis, automation, and
decision-making.
• While IoT focuses on collecting and transmitting data using connected devices, AI
processes this data to simulate human intelligence.
• AIoT improves efficiency, productivity, human–machine interaction, and data
management.

2. How Does AIoT Work?

• AI is embedded into IoT hardware (chips, sensors) and software components.
• IoT devices collect and generate data, which is analysed using AI/ML algorithms.
• Devices communicate using APIs without human intervention.
• Edge AI processes data near the source, reducing:
o Bandwidth usage
o Latency
o Dependency on cloud processing
• Example: Autonomous vehicles using sensors + AI chips connected via IoT networks.

3. Where Does AI Unlock IoT?

[Link] CSE, EPCET pg. 29

IoT operations follow five steps:
1. Create – Sensors generate data
2. Communicate – Data is transmitted
3. Aggregate – Data is collected
4. Analyse – Data is interpreted
5. Act – Decisions are executed
➡️ AI significantly improves the Analyse step, which directly enhances the Act phase.
AIoT enables:
• Meaningful insights from data
• Fast and accurate analysis
• Balance between centralized and edge intelligence
• Improved security and privacy
• Protection against cyberattacks
4. Applications and Examples of AIoT
1. Smart Cities – Traffic control, energy optimization, resource management
2. Smart Retail – Facial recognition, customer movement analysis, dynamic staffing
3. Smart Homes – Personalized appliance control based on user habits
4. Manufacturing – Predictive maintenance and fault detection
5. Hiring – AI-based talent identification using IoT and social data
6. Autonomous Vehicles – Real-time sensing and instant decision-making
7. Robots – Environment-aware intelligent movement
8. Healthcare – Wearables, diagnostics, telemedicine, remote patient monitoring
9. Smart Thermostats – AI-based temperature optimization (e.g., Nest)
10. Drone Traffic Monitoring – Traffic flow optimization using AI analysis
11. Smart Office Buildings – Energy-efficient lighting, HVAC, facial recognition access
12. Fleet Management – Vehicle tracking, fuel optimization, driver behavior analysis

5. Benefits of AIoT
1. Improved Operational Efficiency – Automated optimization using AI insights
2. Real-time Decision Making – Edge AI enables instant responses
3. Reduced Human Workload – Less manual monitoring and analysis
4. Scalability – Supports billions of devices by summarizing data intelligently
5. Enhanced Productivity – Identifies redundant and inefficient processes
6. Better Risk Management – Predicts failures, ensures safety, prevents cyber threats
7. New & Enhanced Products/Services – NLP-based interaction and smart services
8. Predictive Maintenance – Reduces downtime and maintenance costs
Industry Example:
AIoT-enabled predictive maintenance can:
• Reduce maintenance planning time by 20–50%
• Increase equipment uptime by 10–20%
• Reduce maintenance costs by 5–10%

6. Challenges and Risks of AIoT

[Link] CSE, EPCET pg. 30

 • Failure of AIoT systems can lead to:
o Delivery delays (autonomous robots)
o Accidental theft (smart retail)
o Accidents (autonomous vehicles misreading signals)
• Data privacy, security vulnerabilities, and system reliability remain critical concerns.

7. Future of AIoT
• AIoT combined with 5G will enable:
o High-speed data transfer
o Low latency
o Massive device connectivity
• AIoT will revolutionize:
o Supply chains
o Urban infrastructure
o Industrial automation
• By 2025:
o ~42 billion IoT devices globally
o Massive exponential data growth
• AI enables IoT systems to:
o Learn from past decisions
o Predict future events
o Continuously improve performance
• Devices can “think and act independently” without constant cloud dependency.

[Link] CSE, EPCET pg. 31