(2nd Sem) AIML Unit I: Complete Concept Guide
Unit – I: History of AI -> Generated and Prepared By Thiruselvan (ThiruXD)
Level 1: Absolute Basics – What is AI?
1. What is Artificial Intelligence (AI)?
Artificial Intelligence is a branch of computer science that makes machines able to do things that normally need human intelligence.
Examples of human intelligence: learning, reasoning, solving problems, understanding language, seeing, making decisions.
Simple definition you can write in exam:
AI is the science of making computers perform tasks that require human intelligence such as learning, reasoning, problem-solving, perception and decision-making.
Real-life examples:
- Gemini / Grok / ChatGPT / Claude / DeepSeek / KiMi / Qwen
- Sarvam Ai (Text to speech, Speech to Text, ai audio dubbing etc…)
- Siri / Google Assistant (talks and answers)
- Netflix recommending movies
- Self-driving cars
- Chatbots on websites
- Face unlock on mobile
Key point: AI is a broad field. It includes many techniques.
Level 2: The Three Main Layers (Most Important Concept)
AI is like a big circle. Inside it there are smaller circles.
Artificial Intelligence (AI) ← Biggest circle
└── Machine Learning (ML) ← Medium circle
└── Deep Learning (DL) ← Smallest circle1. Artificial Intelligence (AI)
- Goal: Make machines intelligent like humans
- Can use rules or learning from data
- Examples: Expert systems, robots, Siri
2. Machine Learning (ML) – Subset of AI
- Machine learns from data automatically (without writing every rule)
- Improves with experience
- Examples: Spam filter, Netflix recommendations, fraud detection
3. Deep Learning (DL) – Subset of Machine Learning
- Uses artificial neural networks with many layers (inspired by human brain)
- Needs huge data and powerful computers
- Examples: Face recognition, self-driving cars, ChatGPT, AlphaGo
Memory Trick:
- AI Examples → Rule-based or assistants
- ML Examples → Recommendation or prediction
- DL Examples → Image, speech, or self-driving
Level 3: Difference Between AI, ML and DL (Exam Favourite)
| Aspect | Artificial Intelligence (AI) | Machine Learning (ML) | Deep Learning (DL) |
|---|---|---|---|
| Definition | Broad field of intelligent machines | Learns from data | Uses multi-layer neural networks |
| Goal | Simulate human intelligence | Learn automatically from data | Mimic human brain |
| Scope | Widest | Narrower | Narrowest |
| Approach | Rules + Learning | Statistical algorithms | Neural networks |
| Human Intervention | High | Moderate | Low |
| Data Needed | Low to Medium | High | Very High |
| Complexity | Less | Moderate | High |
| Performance | Limited in complex tasks | Good | Excellent in complex tasks |
| Examples | Siri, MYCIN, Roomba | Spam filter, Netflix | Image recognition, AlphaGo, Tesla |
Level 4: History & Evolution of AI (Very Important)
The history of AI has ups and downs called AI Winters (periods when funding and interest reduced).
Stage 1: Symbolic AI (1950s – 1980s)
Also called Good Old-Fashioned AI (GOFAI)
- Approach: Top-Down
- Humans write every rule (if-then rules)
- Based on Physical Symbol System Hypothesis
- Intelligence = Logic
- Technology: Expert Systems (Knowledge Base + Inference Engine)
Example: MYCIN (medical diagnosis system)
Problem (Brittle Problem):
If the situation is not exactly in the rules, the system fails. Cannot handle uncertainty or common sense.
Stage 2: Machine Learning (1990s – 2010s)
- Approach: Bottom-Up
- Machine finds patterns from data itself
- Humans still do Feature Engineering (tell the computer which features are important)
- Intelligence = Pattern Recognition
- Algorithms: Linear Regression, Random Forest, SVM
Example: Email spam filter, recommendation systems
Stage 3: Deep Learning (2012 – Present)
- Special type of Machine Learning using Neural Networks with many layers
- Automatically extracts features (no manual feature engineering)
- Intelligence = Representation Learning
- Catalysts: Big Data + GPUs + Backpropagation algorithm
Key Architectures:
- CNN (Convolutional Neural Networks) → for images
- Transformers → for language (ChatGPT uses this)
Paradigm Shift Summary (Must Remember):
Symbolic AI → Intelligence = Logic
Machine Learning → Intelligence = Pattern Recognition
Deep Learning → Intelligence = Representation Learning
Level 5: Branches of Artificial Intelligence
-
Natural Language Processing (NLP)
Computers understand and generate human language (text + speech).
Tasks: Translation (Google Translate), Chatbots, Sentiment analysis
-
Computer Vision
Computers “see” and understand images/videos.
Uses CNNs.
Tasks: Face recognition, detecting tumors in X-rays, self-driving cars
-
Robotics
AI + Mechanical + Electrical engineering.
AI acts as the “brain” of the robot.
Examples: Roomba, Surgical robots, Mars rovers
-
Expert Systems
Oldest branch. Mimics a human expert.
Structure:
- Knowledge Base (facts + rules)
- Inference Engine (applies the rules)
-
Speech Recognition
Converts spoken words into text.
Used in Alexa, Siri, voice typing.
Level 6: Applications of AI in Different Fields
1. Healthcare
- Detecting diseases from X-rays/MRI
- Drug discovery
- Predicting patient emergencies (sepsis, heart attack)
2. Finance
- Fraud detection
- Algorithmic trading (high-speed buying/selling)
- Credit scoring
3. Education
- Personalized learning (adapts to student’s speed)
- Automatic grading
- Intelligent tutoring bots (24×7 help)
4. Transportation
- Self-driving cars
- Smart traffic lights
- Best delivery routes
5. Entertainment
- Netflix / Spotify recommendations
- Creating movie visual effects
- Virtual influencers and deepfakes
Level 7: Ethical and Societal Implications (Advanced but Scoring)
What is AI Ethics?
Moral rules for designing and using AI systems so that they help society and do not harm people.
Key Principles of AI Ethics:
- Transparency (Explainable AI)
- Accountability
- Fairness
- Privacy
- Safety & Security
- Human Control (Human-in-the-Loop)
Major Issues:
1. Bias and Fairness (The “Mirror” Problem)
AI learns from past data. If data is unfair, AI becomes unfair.
- Data Bias: Medical AI trained only on men → fails for women
- Historical Bias: Hiring AI learns that most past managers were men → prefers men
2. Privacy Concerns
AI needs huge data. Issues:
- Data collection without permission
- Facial recognition surveillance
- AI guessing private information from your behaviour
3. Impact on Jobs and Economy
- AI automates boring/repetitive jobs
- New jobs are created (AI engineers, data scientists)
- Solution: Reskilling (learning to work with AI)
4. AI Governance and Regulations
Rules and laws that control how AI is built and used.
Risk Categories (Modern Approach):
- Unacceptable Risk → Banned (e.g., social scoring)
- High Risk → Strictly regulated (hospitals, hiring, schools)
- Limited Risk → Must be transparent (chatbots must say “I am AI”)
Key Global Rules:
- AI-generated images/videos must have labels (watermark)
- If AI rejects loan/job → must explain why
- Strong privacy protection
- Human always remains responsible
Quick Revision Summary (Last Minute)
- AI → ML → DL (nested circles)
- Evolution: Symbolic (Logic) → ML (Patterns) → DL (Representation)
- Symbolic AI failed because of brittle problem
- Deep Learning succeeded because of Big Data + GPUs + Backpropagation
- Five Branches: NLP, Computer Vision, Robotics, Expert Systems, Speech Recognition
- Ethics pillars: Transparency, Accountability, Fairness, Privacy, Human Control
- Biggest problems today: Bias, Privacy, Job impact, Lack of explainability