UNIT 1 – INTRODUCTION TO ARTIFICAL INTELLIGENCE

1)- Understanding AI
What is AI? – Basic Meaning
• Artificial Intelligence (AI) is the ability of machines or software to perform tasks that
usually require human intelligence.
• It includes learning, reasoning, decision-making, perception, and problem-solving.
• Simple Meaning: AI is making machines think and act smart, like humans.

Example:
• Google Maps suggesting the fastest route.
• Face unlocks in smartphones
Definitions of AI (From Different Fields)
• Computer Science: AI is the study of intelligent agents that perceive their
environment and act to achieve goals.
• Psychology / Cognitive Science: AI simulates human thought processes and behavior.
• Mathematics: AI uses logic, probability, and algorithms to solve problems.
• Engineering: AI is building machines and systems that can function automatically.
• General: AI is “the science and engineering of making intelligent machines.”

Example:
• Computer science: Chess-playing program.
• Psychology: Chatbots imitating human conversation.
• Mathematics: Probability-based spam detection.
Goals of AI (What AI Tries to Do)
1. Automation – Perform tasks without human help.
Ex: Assembly line robots in factories.
2. Reasoning – Logical decision-making.
Ex: Fraud detection systems in banks.
3. Learning – Adapt from experience.
Ex: Netflix recommending shows based on past viewing.
4. Perception – Understand sensory data.
Ex: Self-driving cars detecting pedestrians.
5. Natural Interaction – Communicate like humans.
Ex: Alexa, Siri, Google Assistant.
 6. Problem-Solving – Solve complex issues.
Ex: AI in medical diagnosis.
7. Efficiency – Work faster and more accurately.
Ex: AI in logistics for delivery optimization.
Common Myths vs Reality in AI

Myth Reality

AI will replace humans AI supports humans; creativity and ethics still need
completely. people.

AI is 100% accurate. AI depends on training data, can be biased.

AI has emotions. AI only simulates emotions, does not feel them.

AI = Robots only. AI exists in software too (apps, websites, etc.).

AI is dangerous like in movies. Current AI is “narrow AI,” focused on specific tasks.

Example:
• Myth: Self-driving cars mean no accidents.
• Reality: They reduce accidents but still need human supervision.
Real-World Examples of AI
1. Voice Assistants – Alexa, Siri, Google Assistant.
2. Recommendation Systems – YouTube, Netflix, Amazon.
3. Self-Driving Cars – Tesla Autopilot, Waymo.
4. Healthcare – AI in MRI/CT scan analysis.
5. Finance – Fraud detection, robo-advisors.
6. Education – AI tutors, adaptive learning apps.
7. Security – Face recognition in airports.

2)- EVOLUTION OF AI
1. AI in the 1950s – Logic-Based Systems
• The 1950s marked the beginning of AI as a formal field of research.
• Inspired by mathematical logic and symbolic reasoning, early AI tried to mimic how
humans solve problems.
• The systems were based on rules and logic, not learning.
• Important Works & People:
 o Alan Turing (1950): Proposed the concept of machine intelligence and the
famous Turing Test.
o John McCarthy (1956): Called the “Father of AI,” organized the Dartmouth
Conference where the term “Artificial Intelligence” was coined.
o Allen Newell & Herbert Simon: Developed Logic Theorist (1955), which could
prove mathematical theorems.
• Example: Logic Theorist solved Principia Mathematica problems using formal logic.
• Limitations:
o Could not handle uncertainty (if one rule failed, the system collapsed).
o Too rigid, required complete knowledge upfront.
2. AI in the 1980s – Expert Systems
• By the 1980s, AI moved from pure research into practical applications.
• Expert Systems became popular. These were computer programs designed to mimic
the decision-making of human experts.
• They used IF–THEN rules (Rule-Based Reasoning).
• Working Principle: Knowledge (rules) + Inference Engine = Decision/Advice.
• Famous Examples:
o MYCIN (1970s–80s): Helped doctors diagnose blood infections and
recommend antibiotics.
o DENDRAL: Assisted chemists by analyzing chemical compounds.
• Applications: Medicine, business, engineering, and troubleshooting technical
problems.
• Advantages: Captured expert knowledge and applied it consistently.
• Limitations:
o Difficult to update or expand rule base.
o Could not learn from data → required manual entry of knowledge.
o Very expensive to maintain.
3. AI in the 2000s – Deep Learning Begins
• In the 1990s, AI slowed (called the AI Winter), but in the 2000s, AI made a comeback
with machine learning and neural networks.
• Researchers realized that computers could learn patterns from data instead of
following strict rules.
• Key Developments:
 o 2006 (Geoffrey Hinton): Introduced Deep Belief Networks, reviving neural
networks.
o 2012 (AlexNet): A deep CNN that won the ImageNet competition, reducing
error rates drastically. This proved deep learning was powerful for vision.
• Applications:
o Computer Vision (face recognition, medical imaging).
o Speech Recognition (Siri, Google Voice).
o Natural Language Processing (translation, chatbots).
• Limitations:
o Required huge labeled datasets.
o Needed powerful hardware (GPUs).
o “Black-box” → hard to interpret how models worked.
4. 2018 – Present: Rise of Generative AI
• Since 2018, AI shifted from recognition/prediction to generation (text, images, video,
music).
• The biggest breakthroughs came from:
o Large Language Models (LLMs): GPT, BERT, Claude. They can write essays,
answer questions, and code.
o Diffusion Models: Stable Diffusion, MidJourney. They generate images from
random noise based on prompts.
• Applications:
o Chatbots (ChatGPT).
o Image generation (art, design).
o Content creation (blogs, videos).
o Drug discovery, personalized learning.
• Challenges:
o Ethical concerns (plagiarism, bias).
o Hallucination (giving wrong answers confidently).
o Energy usage for training.
5. What Are LLMs and Diffusion Models?
• Large Language Models (LLMs):
o Trained on billions of words from books, websites, articles.
 o Based on Transformer architecture (introduced in 2017).
o Predicts the next word in a sequence but can generate full paragraphs.
o Examples: GPT (OpenAI), BERT (Google), LLaMA (Meta).
o Use Case Example: Customer support chatbot that answers like a human.
• Diffusion Models:
o Generate images/audio by adding and then removing noise step by step.
o Start with random static noise → refine into a detailed picture.
o Examples: DALL·E, Stable Diffusion, MidJourney.
o Use Case Example: Generating a realistic picture of "a cat wearing glasses
sitting on the moon."

3)-Real-World Applications of Artificial Intelligence

1. AI in Daily Life (Everywhere Around Us)
• AI is deeply integrated into everyday technology we use.
• Examples:
o Voice Assistants: Siri, Alexa, Google Assistant (NLP + speech recognition).
o Navigation & Maps: Google Maps suggests routes, predicts traffic using real-
time AI.
o E-commerce: Personalized shopping suggestions (Amazon, Flipkart).
o Banking & Finance: Fraud detection, chatbots for customer service.
o Healthcare apps: AI watches (Fitbit, Apple Watch) track heart rate, oxygen
levels.
o Smart Homes: AI in appliances (fridge suggesting grocery lists, smart AC
adjusting temperature).
• Impact: Saves time, increases convenience, improves personalization.

Example: Netflix recommending the perfect movie for you.

2. AI in Robotics
• AI powers intelligent robots that can sense, think, and act.
• Applications:
o Industrial robots: Used in car manufacturing (welding, assembling).
o Service robots: Hotels, restaurants using robots for delivery.
o Healthcare robots: Robotic surgery (Da Vinci system), elderly care robots.
 o Military robots: Drones for surveillance, bomb disposal bots.
o Humanoid robots: Sophia (Hanson Robotics) can talk and mimic expressions.
• Impact: Efficiency, safety, and automation of risky or repetitive jobs.

Example: Amazon warehouse robots organizing and shipping items.

3. AI in Natural Language Processing (NLP)
• NLP enables machines to understand and generate human language.
• Applications:
o Chatbots: Customer support bots answering queries.
o Machine Translation: Google Translate, DeepL.
o Speech Recognition: Virtual assistants (voice to text).
o Sentiment Analysis: Companies analyzing social media for customer opinion.
o Text Summarization & Search Engines: Quick info extraction.
• Impact: Breaks language barriers, improves communication, saves time.

Example: Gmail’s “Smart Reply” suggesting automatic responses.

4. AI in Computer Vision
• Computer vision helps machines “see and interpret images/videos.”
• Applications:
o Face Recognition: Unlocking smartphones, airport security.
o Medical Imaging: AI detecting tumors in X-rays or MRIs.
o Autonomous Vehicles: Self-driving cars use vision to detect pedestrians,
signals.
o Retail & Security: CCTV with AI for theft detection.
o Agriculture: AI drones analyzing crops for pests and diseases.
• Impact: Enhances safety, accuracy, and automation in real-world environments.

Example: Tesla’s self-driving car detecting stop signs.

5. AI in Recommender Systems
• Recommendation systems help personalize content, products, and services.
• Applications:
o E-commerce: Amazon suggests products based on browsing history.
o Streaming Platforms: Netflix, YouTube, Spotify recommend shows, videos,
songs.
 o Social Media: Instagram & TikTok suggest posts based on user activity.
o Education: AI suggests study material based on progress.
• Techniques: Collaborative filtering, content-based filtering, hybrid systems.
• Impact: Improves engagement, sales, and user satisfaction.

Example: Spotify auto-generating personalized “Discover Weekly” playlists.

6. New and Creative Ideas of AI
• AI is now used in creative and surprising ways.
• Applications:
o AI in Art & Music: DALL·E, Stable Diffusion generate artworks; AIVA composes
music.
o Fashion Industry: AI suggests outfit combinations, designs clothes.
o Education: AI tutors for personalized learning, VR-based classrooms.
o Healthcare Innovation: AI predicting new drug molecules.
o Environment: AI predicting climate patterns, detecting illegal deforestation.
o Gaming: AI generates dynamic storylines and smart NPCs.
• Impact: Expands creativity, innovation, and problem-solving beyond human limits.

Example: AI designing new drug compounds for cancer research.

7. LLMs and Generative AI in Today’s World
• LLMs (Large Language Models):
o GPT, LLaMA, Claude trained on billions of words.
o Uses Today:
▪ Writing essays, blogs, stories.
▪ Coding assistants (GitHub Copilot).
▪ Virtual tutors for students.
▪ Legal & business document drafting.
• Generative AI (Images, Videos, Music):
o Stable Diffusion, MidJourney, DALL·E create images from text prompts.
o Runway & Synthesia create AI-generated videos and avatars.
o ChatGPT + Canva → design + content automation.
• Everyday Use Cases:
o Companies use LLMs for customer care bots.
 o Content creators use AI for marketing posts.
o Students use it for summaries and brainstorming ideas.
o Doctors use AI for patient report analysis.
• Impact: Boosts productivity, creativity, and accessibility but raises concerns about
misuse (fake content, deepfakes).

Example: ChatGPT writing emails, DALL·E generating marketing posters.

4)-Turing Test and Its Limitations

1. What is the Turing Test?
• The Turing Test was proposed by Alan Turing in 1950 in his paper “Computing
Machinery and Intelligence”.
• Aim: To answer the question, “Can machines think?”
• Instead of defining “thinking,” Turing suggested a practical experiment: If a machine
can engage in a conversation that is indistinguishable from a human, it can be
considered intelligent.
• Setup:
o A human judge interacts via text with two unseen entities: a human and a
machine.
o If the judge cannot reliably distinguish the machine from the human, the
machine is said to have passed the test.

Example: A chatbot convincing a judge that it is human during a conversation.

2. Original Idea of the Turing Test
• Turing’s original concept was called the “Imitation Game.”
• In the game:
o A judge communicates with a human and a machine through text.
o The judge’s task: decide who is human and who is machine.
• Turing predicted that by the year 2000, machines would be able to fool humans in
30% of conversations lasting 5 minutes.
• This shifted the question of intelligence from “Can machines think?” → to “Can
machines imitate human conversation?”

Example: Early chatbot ELIZA (1966) sometimes tricked users by repeating or rephrasing
their input.
3. Modern Criticisms of the Turing Test
Although historically important, the Turing Test has several limitations:
1. Focuses on Imitation, Not Understanding
o A machine may fool humans with tricks, without actual reasoning or
knowledge.
2. Easy to Mislead Humans
o Even simple chatbots can sometimes fool judges (e.g., Cleverbot).
3. Ignores Other Aspects of Intelligence
o Intelligence is more than conversation → includes perception, learning,
creativity, reasoning, emotional understanding.
4. Human Bias
o Judges may be too easily impressed or biased by clever responses.

Example: In 2014, a chatbot called Eugene Goostman (pretending to be a 13-year-old

Ukrainian boy) convinced 33% of judges it was human, but it relied on tricks and evasive
answers.
4. Can ChatGPT Pass the Turing Test?
• ChatGPT and similar LLMs generate human-like responses, often fluent and context-
aware.
• In casual conversation, ChatGPT can often fool humans into thinking it’s another
person.
• However:
o ChatGPT sometimes hallucinates facts (confidently gives wrong answers).
o It lacks true understanding, emotions, and consciousness.
• Verdict: ChatGPT could pass short conversations of the Turing Test, but longer and
deeper discussions reveal its limitations.

Example: ChatGPT can write essays or jokes like a human, but may fail when asked for
deep reasoning or common sense about the real world.
5. Turing Test & Large Language Models (LLMs)
• LLMs (like GPT, Claude, LLaMA):
o Trained on massive text datasets, allowing them to generate fluent, human-
like language.
o Often capable of passing informal Turing-like tests (chatting on social media,
customer service).
• Modern Perspective:
 o Passing the Turing Test is no longer a strong indicator of intelligence since
LLMs can mimic human text without real understanding.
o Researchers now focus on other benchmarks:
▪ Reasoning ability (e.g., solving math problems).
▪ Knowledge grounding (accuracy of facts).
▪ Ethics and safety (avoiding harmful content).

Example:
• An LLM can simulate a doctor’s advice in chat, but it may provide unsafe medical
recommendations → proving Turing Test success doesn’t guarantee real intelligence
or reliability.

5)-Cognitive Modelling vs Rule-Based AI

1. What is Rule-Based AI?
• Definition:
o Rule-Based AI systems work using a set of predefined rules (IF–THEN
statements).
o They do not “learn” from data but instead apply logic programmed by
humans.
• How it Works:
o Knowledge base (rules) + Inference engine (reasoning mechanism).
o Example Rule:
▪ IF temperature > 100°F THEN display “High Fever”.
• Strengths:
o Transparent and explainable (easy to understand why the system made a
decision).
o Useful in domains with well-defined, structured knowledge.
• Weaknesses:
o Cannot handle uncertainty or incomplete data.
o Hard to maintain when rule base grows very large.
o No adaptability – must be manually updated.
• Real-Life Examples:
o MYCIN (1970s): Diagnosed infections using medical rules.
o Expert Systems in Banking: Loan eligibility systems based on financial rules.
 Think of Rule-Based AI as “decision trees” coded by humans.
2. What is Cognitive Modelling?
• Definition:
o Cognitive Modelling is an approach where AI systems are designed to mimic
how humans think, learn, and solve problems.
o Inspired by cognitive science, psychology, and neuroscience.
• How it Works:
o Uses mathematical models or neural networks to replicate processes like
memory, reasoning, and learning.
o Instead of relying on fixed rules, it learns from data and experience.
• Strengths:
o Can handle uncertainty and incomplete information.
o More flexible and adaptive compared to rule-based AI.
o Better for real-world, dynamic environments.
• Weaknesses:
o Often works like a “black box” → hard to explain decisions.
o Requires large amounts of training data.
• Real-Life Examples:
o Self-Driving Cars: Learn from road situations like humans.
o IBM Watson: Mimics human reasoning to analyze huge datasets.
o Chatbots with ML: Learn conversational patterns instead of using fixed
responses.

Think of Cognitive Modelling as “AI that thinks like a human.”

3. Where Do LLMs Fit?
• Large Language Models (LLMs) like GPT, LLaMA, Claude are closer to cognitive
modelling than rule-based AI.
• Why?
o They don’t follow rigid IF–THEN rules.
o Instead, they are trained on massive text data and learn patterns of human
language.
o They simulate how humans use memory, context, and reasoning in
conversations.
• LLMs Characteristics:
 o Can generate text, answer questions, and adapt to context.
o But they don’t “understand” language like humans → they are statistical
learners.
• Example:
o Rule-Based AI → “IF user says ‘hello’ THEN reply ‘hi’.” (Fixed rule).
o LLM → Learns from millions of conversations and generates a natural “Hi
there! How are you doing today?” (Context-aware).

LLMs = Cognitive-style AI, not rigid rule-based AI.

4. Real-Life Examples
1. Rule-Based AI:
o Airport security system with fixed rules:
▪ IF baggage weight > 25 kg → Extra charge.
o Expert medical system suggesting treatment only based on entered
symptoms.
2. Cognitive Modelling AI:
o Self-driving cars (Tesla): Learn from millions of driving scenarios, not fixed
rules.
o Virtual Assistants (Siri, Alexa): Use NLP models that mimic human language
understanding.
3. LLMs in Practice (Cognitive-like):
o ChatGPT: Generates essays, answers, or jokes based on patterns of language.
o Google Bard / Gemini: Uses reasoning-like abilities to summarize or analyze
text.
o GitHub Copilot: Writes code by predicting programmer’s intent, not by
following strict rules.

Comparison: Rule-Based AI vs Cognitive Modelling

Aspect Rule-Based AI Cognitive Modelling

Mimics human thought
Uses predefined IF–THEN rules to
Definition processes (reasoning, learning,
make decisions.
memory).
Learns patterns from data,
Knowledge
Human experts define rules manually. experience, psychology,
Source
neuroscience.
Very rigid – can only handle situations Flexible – can adapt to new,
Flexibility
covered by rules. unseen situations.
 Aspect Rule-Based AI Cognitive Modelling
Learns from examples, improves
Learning Ability No learning (static knowledge base).
with experience.
Easy to explain (decisions traceable to Black-box nature – harder to
Transparency
rules). explain how decisions are made.
Handling Poor – struggles with incomplete or Good – can estimate probabilities
Uncertainty uncertain data. and reason under uncertainty.
Difficult – rule base becomes huge and Scales well with more data and
Scalability
unmanageable. computational power.
Expert systems like MYCIN, Self-driving cars, IBM Watson,
Examples
DENDRAL, loan approval systems. ChatGPT (LLMs).
Structured, well-defined domains
Dynamic, real-world problems
Best Use Cases (medical diagnosis rules, banking
(language understanding
eligibility).

6)-Laws of Thought vs Rational Agent Approach

1. What Are the Laws of Thought?
• The Laws of Thought come from classical logic and philosophy.
• They describe fundamental principles of reasoning that humans are expected to
follow.
• Main classical laws (from Aristotle & symbolic logic):
1. Law of Identity → A is A (things are what they are).
2. Law of Non-Contradiction → A statement cannot be both true and false at
the same time.
3. Law of Excluded Middle → A statement is either true or false; no middle
option.
• In early AI, researchers tried to design systems that reason like humans using logic
rules.
• Example: Logic Theorist (1955) → proved theorems using formal logic.
2. Problems with Laws of Thought in AI
• While logical reasoning is powerful, it had serious limitations in AI:
o Too rigid: Real-world situations are messy, uncertain, and incomplete → logic
cannot capture everything.
o Knowledge bottleneck: Encoding every rule of reasoning is impossible at
large scale.
o Computational limits: Even simple problems can explode into thousands of
logical steps (combinatorial explosion).
o No learning: Systems based on strict laws cannot improve from experience.
 Example: A logic-based system cannot handle “maybe,” “probably,” or “uncertain”
situations (e.g., weather prediction).
3. What is the Rational Agent Approach?
• Shift from thinking to acting.
• Instead of focusing only on reasoning, AI researchers moved to the Rational Agent
Approach.
• Definition:
o An agent is anything that perceives its environment (via sensors) and acts
upon it (via actuators).
o A rational agent is one that chooses the best possible action to achieve its
goals, given what it knows.
• Focus: Not just on reasoning, but on taking actions that maximize success.
• Introduced in modern AI textbooks (Russell & Norvig) as the foundation of AI.
4. Modern Rational Agents
• Modern rational agents are built with machine learning, probability, and
optimization.
• They are capable of:
o Acting under uncertainty.
o Learning from data and experience.
o Making decisions that maximize utility (expected outcome).
• Technologies used: Reinforcement Learning, Decision Theory, Deep Learning,
Planning algorithms.
• Unlike Laws of Thought, rational agents don’t just “reason” → they sense, think, and
act in the real world.
5. Real-Life Examples of Rational Agents
1. Self-Driving Cars (Tesla, Waymo):
o Perceive environment (cameras, LIDAR).
o Decide best action (brake, turn, accelerate) to reach destination safely.
2. Recommendation Systems (Netflix, Amazon):
o Sense: Collect user behavior.
o Act: Suggest movies/products that maximize user satisfaction.
3. Robotics (Warehouse Robots, Drones):
o Perceive warehouse or aerial environment.
 o Act: Pick and deliver items efficiently.
4. Virtual Assistants (Siri, Alexa, Google Assistant):
o Sense: Understand user commands.
o Act: Play music, set reminders, answer queries.
5. Healthcare AI (IBM Watson Health):
o Perceive patient records.
o Act: Suggest treatments that maximize chances of recovery.

7)-Three Big Ideas in AI: Symbolic, Connectionist, and Emergent

Approaches
1. Symbolic Approach (Classical AI)
• Also called Good Old-Fashioned AI (GOFAI).
• Developed in the 1950s–1980s.
• Core Idea: Intelligence can be built by manipulating symbols and rules (logic, if-then
statements).
• Inspired by human reasoning and formal logic.
Features:
• Knowledge-based: Requires explicit representation of facts and rules.
• Deterministic reasoning: Works step-by-step, like solving equations or proofs.
• Top-down approach: Start with rules → apply them to problems.
Examples:
• Expert Systems (1980s): MYCIN (medical diagnosis), DENDRAL (chemical analysis).
• Chess-playing systems: Deep Blue used symbolic search with heuristics.
• Logic Theorist (1955): First AI program to prove theorems.
Limitations:
• Hard to handle uncertainty, learning, and perception (vision, speech).
• Knowledge bottleneck: You can’t manually program all rules of the real world.
2. Connectionist Approach (Neural Networks)
• Developed as an alternative to symbolic AI (1950s → became powerful in 2000s).
• Core Idea: Intelligence emerges from networks of simple units (neurons), inspired
by the brain.
Features:
 • Data-driven: Learns patterns from examples, not rules.
• Bottom-up approach: Start with data → system learns structure.
• Can approximate complex functions and recognize patterns humans struggle to
formalize.
• Powered by Deep Learning (2000s onwards).
Examples:
• Image recognition: Convolutional Neural Networks (CNNs) → detect cats, faces,
traffic signs.
• Speech recognition: Siri, Alexa.
• Natural Language Processing (NLP): Transformers (GPT, BERT).
• AlphaGo: Learned to play Go better than humans.
Strengths:
• Handles uncertainty, noise, perception tasks.
• Learns from big data instead of requiring hand-coded rules.
Weakness:
• Often seen as a “black box” (hard to explain decisions).
• Requires huge datasets and computing power.
3. Emergent Approach (Complex from Simple)
• Newer idea: Intelligence can emerge from simple rules interacting together at scale.
• Inspired by nature (ants, birds, ecosystems).
Features:
• Self-organization: No central control; intelligence emerges from interactions.
• Adaptive & flexible: Works well in dynamic environments.
• Often combined with agent-based modeling and evolutionary algorithms.
Examples:
• Flocking behavior: Birds flying in groups → complex coordination from simple rules.
• Swarm robotics: Many small robots cooperating to explore environments.
• Internet/Web: Collective intelligence emerges from billions of users.
• Emergent properties in AI: GPT-like behavior (coherent answers) emerging from
scaling up neural networks.
4. Generative AI = Connectionist + Statistical Power
• Generative AI (GenAI), like ChatGPT, Stable Diffusion, MidJourney, combines:
 o Connectionist foundations: Neural networks (Transformers, Diffusion
Models).
o Statistical power: Trained on massive data, uses probability to generate
outputs.
• It is an emergent intelligence: behaviors (creativity, reasoning, coding ability) emerge
from scaling neural networks, not from explicit programming.
Real Examples:
• ChatGPT (LLM): Generates human-like text.
• DALL·E / Stable Diffusion: Generates images from text prompts.
• Music generation models: AI composing original songs.

✅ Comparison Table

Approach Core Idea Method Examples Strengths Weaknesses

Expert
Explainable, Can’t handle
Reasoning Top-down Systems
Symbolic good for perception,
with symbols logic, if-then (MYCIN),
(Classical AI) structured uncertainty,
& rules rules Logic
problems learning
Theorist

Great at Black-box,
Connectionist Intelligence Bottom-up
CNNs, GPT, vision, speech, needs huge
(Neural from neuron- learning from
AlphaGo NLP, adapts data &
Networks) like units data
from data compute

Swarm
Intelligence Self- Less
Emergent robots, Adaptive,
emerges from organization, predictable,
(Complex from flocking scalable,
simple agents swarm harder to
Simple) models, LLM resilient
interacting behavior design directly
behaviors

In short:
• Symbolic AI = logic & rules (what intelligence should be).
• Connectionist AI = neural networks (how intelligence can be learned).
• Emergent AI = complexity from simple rules (how intelligence appears at scale).
• Generative AI = Connectionist + Emergent properties, powered by huge data and
compute.
8)-Core Skills of Artificial Intelligence
Artificial Intelligence systems are generally designed to exhibit four fundamental
capabilities: Search, Reasoning, Learning, and Planning.
1. Search in AI
• Definition:
o Search refers to the ability of an AI system to explore possible options or
solutions to find the best one for a given problem.
• Classic AI Techniques:
1. Uninformed Search: No information about goal location. Examples:
▪ Breadth-First Search (BFS) → explores level by level.
▪ Depth-First Search (DFS) → explores paths deeply before
backtracking.
2. Informed Search (Heuristic Search): Uses problem-specific knowledge to improve
efficiency. Examples:
▪ A* algorithm → finds shortest path efficiently.
▪ Greedy Best-First Search → picks path closest to goal.
• Applications:
o Game AI: Chess, Tic-Tac-Toe, Go.
o Route Planning: GPS and map services (shortest route search).
o Puzzle Solving: Rubik’s cube solvers, Sudoku solvers.
2. Reasoning in AI
• Definition:
o Reasoning is the AI’s ability to derive conclusions or make decisions based on
knowledge or facts.
• Classic AI Methods:
1. Logic-Based Reasoning: Using formal logic rules (propositional & predicate logic).
▪ Example: “All humans are mortal. Socrates is human. Therefore,
Socrates is mortal.”
2. Rule-Based Systems: Expert systems that use IF–THEN rules.
▪ Example: MYCIN diagnosing diseases using medical rules.
3. Probabilistic Reasoning: Handles uncertainty using probabilities (Bayesian networks).
▪ Example: Weather prediction, spam email detection.
• Applications:
 o Medical diagnosis, legal reasoning, decision support systems, AI in games.
3. Learning in AI
• Definition:
o Learning allows AI to improve its performance or knowledge from
experience or data.
• Types of Learning:
1. Machine Learning (ML):
▪ Learns patterns from data without being explicitly programmed.
▪ Types:
▪ Supervised Learning: Learn from labeled data (e.g., spam
email detection).
▪ Unsupervised Learning: Find patterns in unlabeled data (e.g.,
customer segmentation).
▪ Reinforcement Learning: Learns by trial and error using
rewards (e.g., AlphaGo).
2. Deep Learning (DL):
▪ A subset of ML using multi-layered neural networks to handle
complex tasks.
▪ Capable of automatic feature extraction from raw data.
▪ Applications: Image recognition, NLP, autonomous vehicles, speech
recognition.
• Real-Life Examples:
o Google Translate (language translation).
o Netflix recommendations (predicting user preferences).
o Self-driving cars (learning from sensor data).
4. Planning in AI
• Definition:
o Planning is the AI’s ability to create a sequence of actions to achieve specific
goals.
• Types of Planning:
1. Classical Planning: Works with fully known environments and deterministic
outcomes.
▪ Example: Robot moves from point A to B in a grid world.
2. Probabilistic/Modern Planning: Works in uncertain or dynamic environments.
▪ Example: Autonomous drones navigating changing weather or
obstacles.
• Applications:
o Robotics: Robot task planning in factories.
o Games: AI decides optimal moves ahead of time.
o Logistics: Supply chain optimization, route planning.
• Relation to Search: Planning often uses search algorithms to explore possible action
sequences.

Summary Table: Four Core Skills of AI

Skill Definition Methods/Techniques Real-Life Examples

Exploring options
Search BFS, DFS, A*, heuristic search GPS route planning, Chess AI
to find solutions

Drawing
Logic, Rule-based systems, MYCIN (medical), legal AI,
Reasoning conclusions from
Probabilistic reasoning game AI
knowledge

Improving from ML: supervised, Netflix recommendations, self-

Learning
data or experience unsupervised, RL; DL driving cars, Google Translate

Sequencing
Classical planning, Robotics, logistics,
Planning actions to achieve
probabilistic planning autonomous drones
goals

9)-Classic AI vs Generative AI & Modern AI Concepts

Feature Classic AI Generative AI (GenAI)

Creating new content: text, images,

Focus Reasoning, rules, search, logic
audio, video

Symbolic rules, logic, planning, Neural networks, deep learning,

Method
search probabilistic models

Data Limited; works with pre-coded Huge datasets; learns patterns

Dependence knowledge statistically

Flexibility Rigid, domain-specific Adaptive, creative, multimodal

Feature Classic AI Generative AI (GenAI)

MYCIN (medical expert system), ChatGPT (text), DALL·E/Stable Diffusion

Examples
Deep Blue (chess AI) (images), MidJourney

10)-The Role of Generative AI in the Future of AI

• GenAI will drive AI from prediction & recognition → creation & innovation.
• Applications across education, content creation, healthcare, research, art, and
robotics.
• Enables personalization at scale, creativity augmentation, and autonomous decision-
making.

1. What is Generative AI?

• AI that creates new content (text, images, audio, video, code) from learned patterns.
• Uses deep learning models trained on massive datasets.
• Key property: can generate outputs not explicitly seen in training data.

2. Types of Generative AI
1. Text Generation: ChatGPT, GPT-4, Claude
2. Image Generation: DALL·E, Stable Diffusion, MidJourney
3. Video/Animation Generation: Runway, Synthesia
4. Audio/Music Generation: Jukebox (OpenAI), MusicLM
5. Code Generation: GitHub Copilot, CodeT5

3. Key Generative AI Models to Know

• LLMs (Large Language Models): GPT-4, LLaMA, Claude
• Diffusion Models: Stable Diffusion, MidJourney
• Transformers: Foundation for text & multimodal AI
• Multimodal Models: Combine text, image, and sometimes audio (e.g., GPT-4
multimodal)
4. Multimodal AI and Autonomous Agents
• Multimodal AI: Can process multiple types of data (text + image + audio).
• Autonomous Agents: AI systems that perceive, plan, and act without human
intervention.
• Examples: Autonomous vehicles, AI-powered drones, AI chatbots with planning
ability.

5. Risks of Generative AI
• Misinformation & Deepfakes: AI-generated fake news, videos, images.
• Bias & Discrimination: AI reproduces societal biases from training data.
• Job Displacement: Automation of content creation and coding tasks.
• Security Threats: Malicious use for cyberattacks or phishing.

11)- AI Agents and Agentic AI

1 What is an AI Agent?
• An AI Agent is an entity that:
1. Perceives its environment through sensors.
2. Acts via actuators to achieve goals.
3. Operates autonomously to some extent.
Features of an AI Agent:
• Autonomy (operates without human input)
• Reactivity (responds to changes)
• Proactivity (plans ahead to achieve goals)
• Social ability (interacts with other agents or humans)
2 What is Agentic AI?
• Agentic AI: AI that behaves like a rational agent capable of goal-directed
autonomous action.
• Can plan, make decisions, and execute tasks with minimal human guidance.
Real-Life Examples:
• Autonomous Vehicles: Tesla, Waymo
• Robotics: Boston Dynamics robots performing tasks autonomously
• Virtual Assistants: ChatGPT as a planning/decision-making assistant
 • Autonomous Trading Bots: Financial AI agents executing trades automatically

12). Ethics, Explainability, and Responsible AI

1 What is Responsible AI?
• AI developed and deployed ethically, safely, transparently, and accountably.
• Focuses on avoiding harm, bias, and misuse.
2 Why Ethics in AI is Important
• Prevents harmful societal impacts (bias, discrimination).
• Ensures trust and accountability in AI systems.
• Critical for adoption in healthcare, law, finance, and governance.
3 What is Explainable AI (XAI)?
• AI that provides clear, understandable reasoning behind its decisions.
• Helps humans trust, validate, and debug AI systems.
4 Importance of Explainability
• Regulatory compliance (laws like GDPR require transparency).
• Safety-critical applications (medical AI, autonomous vehicles).
• Understanding AI mistakes and biases.
5 Why Generative AI is Hard to Explain
• LLMs and diffusion models are black-box neural networks.
• Outputs emerge from millions/billions of parameters → difficult to trace specific
reasoning.

Social and Ethical Issues in Modern AI

• Bias & Fairness: AI may reproduce societal inequalities.
• Privacy: Data used to train AI may expose sensitive information.
• Misinformation: Generative AI can produce false or misleading content.
• Job & Economic Impact: Automation may displace jobs.
• Autonomy & Accountability: Who is responsible when an AI agent makes a wrong
decision?