AI vs ML vs DL vs generative AI: study notes
0.02 AI vs ML vs DL vs generative AI: detailed study notes
Learn AI by Story · Module 0: What is AI · Level: Beginner
The idea in one sentence
AI, machine learning, deep learning and generative AI are not four names for the same thing: they nest inside each other like dolls, with AI the biggest and generative AI the smallest.
The story from the video
Asha finds a set of nesting dolls and asks whether AI, machine learning, deep learning and generative AI are all the same. They fit inside each other: AI is the biggest doll, machine learning sits inside it, deep learning inside that, and generative AI is the smallest. Every doll is AI, but not every AI is generative.
Key definitions
Term | Definition | In plain English |
|---|---|---|
Artificial intelligence (AI) | Any technique that lets a machine perform tasks that normally need human intelligence, by any method. | The whole field. |
Machine learning (ML) | A subset of AI in which a system learns patterns from data instead of following rules written by hand. | AI that learns from examples. |
Neural network | A model made of layers of simple connected units (neurons) whose connection strengths are adjusted during training. | A web of tiny calculators that learns. |
Deep learning (DL) | Machine learning that uses neural networks with many layers. | ML with deep, many-layered networks. |
Generative AI (GenAI) | AI, usually built with deep learning, that creates new content such as text, images, audio, video or code. | AI that makes new things. |
Discriminative model | A model that tells categories apart or predicts a label (cat or dog). | A judge. |
Generative model | A model that learns the data well enough to create new examples that look like it. | A creator. |
Training | Adjusting a model using data so that it performs better. | The learning phase. |
Rules-based system | Software whose behaviour comes from rules written by a human. | If this, then that. |
Large language model (LLM) | A very large deep-learning model trained on huge amounts of text to predict and generate language. | The engine behind many chatbots. |
Detailed explanation
1. The nesting picture
Think of four dolls, biggest to smallest:
- AI is the biggest doll. It simply means machines acting smart, by any method. That includes rule-based systems, search methods that plan moves, and learning systems.
- Machine learning sits inside AI. Instead of a programmer writing the rules, the system is shown many examples and works out the rules itself.
- Deep learning sits inside machine learning. It uses neural networks with many layers, which made big jumps in vision, speech and language possible once large datasets and fast chips (GPUs) became available.
- Generative AI sits inside deep learning. These models do not only label or predict, they create new text, pictures, music or code.
So: every generative AI is deep learning, every deep learning is machine learning, and every machine learning is AI. The reverse is not true.
2. AI without learning
A chess program with hand-written rules and a search procedure is AI, but it is not machine learning because it did not learn its rules from data. Many early AI systems, such as expert systems built from lists of if-then rules, belong in the AI doll but not in the machine learning doll.
3. Machine learning without deep learning
A spam filter can learn from labelled emails using a simple method such as logistic regression or a decision tree. That is machine learning, but it is not deep learning, because it does not use a many-layered neural network. Classic methods like these are still widely used, especially on tables of data.
4. Deep learning without generation
Face unlock uses a deep neural network to recognise a face. It outputs a decision, match or no match. It is deep learning but not generative, because it does not create new faces.
5. Generative AI
A generative model learns the patterns in its training data well enough to produce new examples. A text model writes a paragraph one piece at a time, an image model produces a picture from a description, and an audio model can produce speech or music. Most of today's generative systems are deep learning models.
What the video did not explain
- Where "data" comes in. The big difference between the AI doll and the machine learning doll is where the rules come from: a human (AI in general) or the data (machine learning). Video 0.04 covers this in depth.
- Why deep learning took off around 2012. Three things came together: large datasets, powerful GPUs, and better training methods. Video 0.03 tells that history.
- Other kinds of generative models. Beyond language models there are diffusion models (used for many image generators), generative adversarial networks (GANs) and variational autoencoders. You will meet them later in the roadmap.
- Not all machine learning is neural. Linear regression, decision trees, random forests, k-nearest neighbours, support vector machines and k-means clustering are classic machine learning methods that are not deep learning.
- Other AI families that sit outside machine learning. Search and planning algorithms, logic and knowledge representation, and optimisation methods.
- The risks specific to generative AI. It can produce false statements that sound confident (often called hallucination), copy biases from training data, and raise questions about copyright and misuse.
- The word "AI" in marketing. Many products call themselves AI when they use simple rules or basic statistics. The nesting picture helps you ask: is it learning from data, and is it a deep network?
Worked example
Take one task: dealing with photos of cats. Which doll does each system belong in?
System | AI? | ML? | DL? | GenAI? |
|---|---|---|---|---|
A program that says "cat" if the filename contains the word cat | Yes (very simple rules) | No | No | No |
A model that learns from 10,000 labelled photos using a decision tree on hand-picked features | Yes | Yes | No | No |
A deep neural network that recognises cats in photos | Yes | Yes | Yes | No |
A model that draws a brand new picture of a cat from a text description | Yes | Yes | Yes | Yes |
Read each row from left to right: a Yes in a later column always comes with Yes in the earlier columns.
Common mistakes and myths
- "AI and machine learning are the same." Machine learning is one way to build AI. Some AI uses no learning.
- "Deep learning is the only machine learning." Many successful ML systems use simpler methods.
- "Generative AI is all of AI." It is the newest, smallest doll. Most deployed AI still does classification, prediction and recommendation.
- "Generative AI understands what it creates." It produces output that matches patterns in its data, which is not the same as human understanding.
- "Bigger always means better." Larger models can be more capable but cost more and are not the right tool for every job.
Where it is used
- AI in general: route planning, game-playing programs, scheduling and logistics.
- Machine learning: spam filters, credit scoring, demand forecasting, recommendation lists.
- Deep learning: face unlock, speech recognition, medical image analysis, self-driving perception.
- Generative AI: chat assistants, writing and coding helpers, image and video creation, translation and summarisation.
Self-check questions
- Which doll is the biggest? AI.
- What is the difference between AI and machine learning? Machine learning learns its rules from data; AI also includes systems with hand-written rules.
- What makes deep learning "deep"? Neural networks with many layers.
- What does generative AI do that face unlock does not? It creates new content instead of only classifying input.
- Is a chess program with hand-written rules machine learning? No. It is AI but not machine learning.
- Give a machine learning example that is not deep learning. A spam filter using logistic regression or a decision tree.
- Is every generative AI also deep learning? Almost all today are, so it sits inside the deep learning doll.
- Name two other types of generative models besides language models. Diffusion models and GANs, for example.
- Why can generative AI be wrong while sounding sure? It generates likely-looking output and does not check facts.
- A shop calls its rule-based discount calculator "AI-powered". Which questions would you ask? Does it learn from data? Is it a neural network? Does it generate anything?
Key takeaways
- The four terms nest: AI, then machine learning, then deep learning, then generative AI.
- Every doll is AI, but not every AI is generative.
- The key question for machine learning is whether the rules come from data.
- Deep learning means many-layered neural networks.
- Generative AI creates new content.
Go deeper and what next
- Draw the four dolls and write one product example inside each.
- Look up one non-neural machine learning method and one generative model type.
- Next in the roadmap: 0.03 A short history: symbolic AI, AI winters, the deep-learning boom.
Comments
Post a Comment