M1-A - The Intelligence Stack
Do not read this like a dictionary of AI terms. Read it like a story: how computers slowly moved from following fixed instructions to helping with goals.
Most people meet AI through ChatGPT.
So when they hear words like AI, Machine Learning, Deep Learning, Generative AI, and Agentic AI, everything feels like one big cloud.
But these are not just fancy names. They are layers.
Rules-based AI
-> Machine Learning
-> Deep Learning
-> Generative AI
-> Agentic AI
Each layer became important because the previous layer had a limit.
The story is not simply:
AI became smarter.
The better story is:
Humans got tired of writing every instruction by hand.
So we built systems that could learn patterns, create content, and now even take steps toward a goal.
1. AI: The Big Umbrella
Artificial Intelligence means making computers do tasks that normally need some kind of human-like thinking.
That can include:
- Understanding a question
- Recognizing a face
- Predicting what a customer may buy
- Writing a summary
- Planning a trip
- Helping debug code
- Answering from documents
AI is the big umbrella. Under it, many methods exist.
AI is the goal. Machine Learning, Deep Learning, Generative AI, and Agentic AI are different ways of reaching that goal.
Think of "transport" as the big idea. Walking, cycling, trains, and flights are different methods. In the same way, AI is the big idea, and the stack below it contains different methods.
2. Rules-Based AI: The Computer Follows a Rulebook
The earliest style of AI was very direct.
Humans wrote rules.
IF this happens
THEN do that
Example:
IF the user types "hello"
THEN reply "Hello, how can I help you?"
This works well when the world is small and clear.
For example:
| Good for rules | Why it works |
|---|---|
| Login permissions | The rule is clear: allowed or not allowed |
| Form validation | A phone number has a required format |
| Tax slabs | The calculation follows fixed conditions |
| Chess legal moves | The game has strict rules |
But conversation is not like a tax form.
A student may ask:
What is AI?
Explain AI simply.
Bro what does artificial intelligence mean?
Is ChatGPT AI or ML?
Tell me like I am in 10th.
All five questions are asking almost the same thing, but the words are different.
If we try to write a separate rule for every sentence, spelling mistake, language style, and mood, the rulebook becomes impossible to manage.
Rules-based AI breaks when real life has too many variations. Humans cannot write a rule for every possible situation.
The real insight: rules are useful when the world is fixed. Intelligence becomes difficult when the world is messy.
3. Machine Learning: The Computer Learns from Examples
Machine Learning changed the approach.
Instead of writing all rules, humans gave examples.
Old style:
Human writes the rule.
Machine Learning style:
Human gives examples. The machine finds patterns.
Example: spam email detection.
Rules-based version:
IF email contains "free money"
THEN mark as spam
But real spam is smarter than that. It may use different words. It may hide the same idea in a new sentence.
Machine Learning version:
Here are 100,000 emails.
Some are spam. Some are normal.
Learn the pattern.
The system studies examples and learns signals.
Maybe spam emails often have strange links, urgent language, fake prizes, or unusual sender patterns. The model learns these patterns from data.
| Traditional programming | Machine Learning |
|---|---|
| Humans write exact steps | Model learns from examples |
| Best when rules are clear | Best when patterns are hidden |
| Easy to predict | Gives probability-based answers |
| Debug by reading code | Improve by checking data and results |
Machine Learning is powerful, but it is not magic.
If the examples are poor, the learning is poor.
If the examples are biased, the model can learn bias.
If the examples are missing important cases, the model may fail on those cases.
If a teacher gives you only easy math problems during practice, you may feel confident. But in the exam, a tricky problem may expose the gap. ML models also depend heavily on the examples they train on.
The real insight: Machine Learning moved intelligence from hand-written rules into learned patterns.
4. Deep Learning: Bigger Pattern Learning
Deep Learning is a type of Machine Learning.
It uses systems called neural networks. You do not need to imagine a human brain. Think of a neural network as a large pattern-finding machine made of many connected layers.
Earlier Machine Learning often needed humans to decide what features mattered.
Example: if we wanted a system to recognize cats in images, humans might manually guide it:
Look for ears.
Look for eyes.
Look for whiskers.
Look for fur texture.
Deep Learning can learn many of these useful signals by itself from lots of examples.
That is why it became important for:
- Images
- Speech
- Language
- Translation
- Medical scans
- Recommendation systems
But here is the interesting part: neural networks are not a brand-new idea. People explored them decades ago. They became seriously useful when the world finally had three things:
| Needed ingredient | Why it mattered |
|---|---|
| More data | The internet created huge training material |
| More compute | GPUs made the math faster |
| Better training methods | Researchers learned how to train larger networks |
GPUs were originally famous for graphics and gaming. Later, they became important for AI because neural networks need a lot of repeated math, and GPUs are very good at that.
Sometimes an idea is not wrong. It is just waiting for the right ecosystem: data, hardware, tools, and timing.
The real insight: Deep Learning became powerful when big data and GPUs made large-scale pattern learning practical.
5. Generative AI: The AI That Creates
Generative AI creates new content.
It can create:
- Text
- Code
- Images
- Audio
- Video
- Summaries
- Plans
- Explanations
ChatGPT is the most familiar example for many students.
A simple mental model for language-based Generative AI is:
Very powerful autocomplete trained on huge amounts of text.
This does not mean it is weak. Autocomplete at small scale is a phone keyboard guessing your next word. Autocomplete at massive scale can learn grammar, writing style, coding patterns, question-answer patterns, and explanation patterns.
When you type:
The capital of India is
the model predicts that the next likely words are:
New Delhi
It does not "know" like a human sitting in geography class. It predicts likely continuation based on patterns it learned and the context you gave.
This is why Generative AI feels impressive and risky at the same time.
| Strength | Risk |
|---|---|
| Explains difficult topics quickly | Can sound correct while being wrong |
| Creates first drafts fast | May invent facts |
| Changes tone and level | May miss the real context |
| Helps with code | May create subtle bugs |
| Summarizes long content | May skip important details |
A hallucination is when AI gives an answer that sounds confident but is false or unsupported. The danger is not only that it is wrong. The danger is that it is written like it is sure.
So how should we use it?
Use it like a fast junior assistant:
- Good for drafts
- Good for options
- Good for explanations
- Good for structure
- Good for starting
But do not treat it like a final authority.
The real insight: Generative AI creates useful outputs, but its confidence is not proof of truth.
6. Agentic AI: From Chat Box to Goal System
Most people use AI like this:
Ask question -> Get answer
Agentic AI changes the shape:
Give goal -> AI plans steps -> uses tools -> checks progress -> continues
This is a big shift.
Generative AI can answer:
Explain how to make study notes.
Agentic AI can be designed to do something closer to:
Read these three PDFs.
Create study notes for each chapter.
Save them as markdown files.
Tell me which parts need human review.
An AI agent is not just a smarter chatbot. It is usually a system that combines a model with extra parts.
| Part | Simple meaning |
|---|---|
| Goal | What the system is trying to finish |
| Planning | Breaking the goal into steps |
| Tools | Access to files, web, apps, code, or databases |
| Memory | Keeping useful context |
| Checking | Reviewing whether the result is good |
| Human control | Stopping risky or wrong actions |
A normal chatbot can suggest a workout plan. An agentic system could ask your schedule, create a plan, add it to a calendar, remind you, track progress, and adjust next week. The model is only one part. The workflow is the bigger system.
Agentic AI matters because many real jobs are not one-question jobs. They are workflows.
Examples:
- Prepare a report
- Follow up with students
- Compare documents
- Update a CRM
- Create content from meeting notes
- Build a small app
- Search files and answer from them
The real insight: Generative AI creates content. Agentic AI tries to create progress toward a goal.
7. The Stack in One Table
| Layer | Easy meaning | Example |
|---|---|---|
| Rules-based AI | Follows human-written rules | If password is wrong, deny login |
| Machine Learning | Learns patterns from examples | Detect spam email |
| Deep Learning | Learns deeper patterns from huge data | Recognize speech or images |
| Generative AI | Creates new content | Write an explanation or image prompt |
| Agentic AI | Plans and acts toward a goal | Read files, create notes, save outputs |
Here is another way to remember it:
Rules: "Tell me exactly what to do."
ML: "Show me examples and I will learn the pattern."
Deep Learning: "Give me huge data and compute, I will learn deeper patterns."
Generative AI: "Give me a prompt, I will create an output."
Agentic AI: "Give me a goal, I will work through steps."
8. Why This Matters for This Course
The brochure says this program is not only about using AI. It is about becoming an AI builder.
That means you should not stop at:
I know how to ask ChatGPT questions.
You are moving toward:
I know how to design useful AI workflows.
That requires understanding the stack.
If you do not understand the stack, every AI tool looks like magic.
If you understand the stack, you can ask better questions:
- Is this a rules problem?
- Is this a pattern problem?
- Is this a content generation problem?
- Is this a workflow problem?
- Where should a human review the result?
- What can go wrong?
AI is not one tool. It is a shift from humans writing every instruction to humans designing systems that learn, generate, and act under control.
Final Recap
Rules-based AI followed fixed instructions.
Machine Learning learned from examples.
Deep Learning learned bigger patterns using huge data and compute.
Generative AI created new text, code, images, and other content.
Agentic AI connected models to goals, tools, memory, and workflows.
The important mindset:
Do not ask only, "What can AI answer?"
Ask, "What kind of system am I dealing with, and how much control does it need?"
That mindset will make the rest of the course much easier.