Module 1: What Is AI Really?

2. Definitions of AI: AI vs ML vs DL vs GenAI

Description

In this lesson you'll untangle the most common confusion: what exactly "artificial intelligence", "machine learning", "deep learning" and "generative AI" are. In the media, on social networks and in conversation they're used as if they were the same thing; they aren't. They're nested concepts: one contains the other. When you finish, you'll be able to explain the difference precisely and you won't mix them up again.

This matters because throughout the rest of the guide (and in the bootcamp) you'll hear "ML model", "neural network", "LLM", "generative model". If the map of definitions isn't clear to you, it all turns into noise. Here you build that map.

The pyramid: from general to specific

Think of four levels, from the bottom up:

AI (Artificial Intelligence) — The broadest concept.
ML (Machine Learning) — A subset of AI: systems that learn from data.
DL (Deep Learning) — A subset of ML: learning with deep neural networks.
GenAI (Generative AI) — A type of application/use of AI (often based on DL) that generates new content.

Not everything called "AI" is ML; not all ML is Deep Learning; not all Deep Learning is generative. But almost everything sold today as "AI" in products (recommendations, chatbots, generated images) uses ML and, in many cases, DL.

Quick analogy: Think of AI as "everything a machine does that we associate with intelligence". ML is "one specific way of doing AI: having the machine learn from data instead of us programming all the rules". DL is "one specific way of doing ML: using networks with many layers". GenAI is "one specific use of AI (and often of DL): having the machine generate new content (text, images, etc.)". So when you hear "ML model", "neural network" or "LLM", you know they're inside the AI umbrella; when you hear "generative model" or "GenAI", you know it's a type of use (generating), not a different level in the pyramid.

What is Artificial Intelligence (AI)?

Practical definition: Systems or machines that display behavior we would associate with "intelligence": perceiving, reasoning, learning, deciding or acting in an environment to achieve a goal.

It includes:

Systems that follow fixed rules (e.g. an automated reply bot with "if X then Y" rules).
Systems that learn from data (ML).
Systems that combine symbolic logic and learning.
Systems that generate text, images or audio (GenAI).

It does not include (by industry convention):

Purely sequential programming with no learning or adaptation (e.g. a calculator).
Automation that doesn't involve "decision" or "learning" in the sense used in AI (e.g. a stopwatch).

The boundary isn't mathematically sharp: there are gray cases. But in practice, when someone says "this has AI", they usually mean something that learns from data, that reasons about inputs not seen exactly before, or that generates content. That brings us to ML.

Analogy: AI is like "sports". There are many sports (soccer, swimming, chess); they're all "sports" but they aren't the same thing. AI is the umbrella; under it there are systems that follow rules, systems that learn from data (ML), systems that generate content (GenAI), etc. Not everything under the AI umbrella is ML; but everything that is ML is under the AI umbrella.

What is Machine Learning (ML)?

Practical definition: An approach to AI in which the system improves its behavior from data (examples, experiences) instead of relying only on rules hand-written by humans.

Key characteristics:

Data: The system needs data (labeled or not, depending on the type of ML).
Learning: Behavior is adjusted through a "training" process.
Generalization: It's expected to work well on cases not seen exactly during training.

Typical examples:

Classifying email as spam or not spam.
Recommending movies (Netflix, Spotify).
Recognizing voice or images.
Predicting sales or machine failures.

Relationship with AI: ML is a subset of AI. All ML is AI; not all AI is ML (for example, a system with only "if-then" rules can be considered AI but not ML).

Analogy: In traditional programming, you write the rules ("if the user is over 18, show X"). In ML, you provide data (examples of users and what was shown) and the system learns the rules. It's like the difference between handing someone an instruction manual (classic programming) and letting someone learn by seeing many examples (ML). That's why when a product "improves with use" or "personalizes itself", there's usually ML behind it.

What is Deep Learning (DL)?

Practical definition: A type of Machine Learning that uses neural networks with many layers ("deep") to learn increasingly abstract representations of the data.

Key characteristics:

Architecture: Networks with many layers (hence "deep").
Representations: Each layer tends to capture more complex patterns (e.g. in images: edges → shapes → objects).
Scale: It usually requires a lot of data and a lot of compute; that's why it took off once there were enough data and GPUs.

Typical examples:

Computer vision (object recognition, segmentation).
Natural language processing (translation, language models).
Speech recognition.
Games (AlphaGo) and many of today's LLMs (GPT, Claude, etc.).

Relationship with ML: Deep Learning is a subset of ML. All DL is ML; not all ML is DL (there's "classic" ML with trees, regressions, etc. without deep neural networks).

Analogy: Classic ML (decision trees, regressions, SVM) is like having simple layers: inputs → learned rules → output. Deep Learning is like having many intermediate layers that extract increasingly abstract patterns: in images, for example, the first layer might detect edges, the next shapes, the next parts of objects, and the last one "face" or "car". That depth of layers is what lets language models (LLMs) and image models handle enormous complexity; that's why when you hear "large model" or "LLM", it's almost always Deep Learning.

What is Generative AI (GenAI)?

Practical definition: AI systems that generate new content (text, image, audio, video, code) that didn't exist as such before, from inputs (prompts, data, conditions).

Key characteristics:

Output: "New" content (not just classifying or predicting a number).
Models: Today they're usually large Deep Learning models (LLMs, image models, etc.).
Use: Conversational assistants, image generation, summaries, code, music, etc.

Typical examples:

ChatGPT, Claude, Gemini (text).
DALL·E, Midjourney, Stable Diffusion (image).
Tools that generate code from a description.
Models that generate audio or video.

Relationship with AI/ML/DL: GenAI is a type of application/use of AI. Many of today's GenAI systems are based on DL (e.g. Transformers, LLMs). It isn't a "level" in the pyramid the way ML or DL are; it's "what the system does" (generate) more than "how it learns".

Analogy: Classifying email as spam or not spam is "discriminative" (it chooses between categories). Generating a new email from a topic is "generative" (it creates new content). GenAI is everything that generates: text, image, audio, code. ChatGPT generates text; DALL·E generates images. Both are GenAI and both usually use Deep Learning models (LLMs, image networks).

Why this matters for an AI Engineer

An AI Engineer doesn't train models from scratch; they integrate models (for example, via OpenAI, Anthropic, etc. APIs) into applications. Those models are almost always ML (they learned from data), DL (deep networks) and often GenAI (they generate text or images). When you read documentation or job postings you'll see "ML model", "LLM", "generative model". If you're not clear that ML is a type of AI, that LLMs are a type of DL used to generate (GenAI), and that all of that is "narrow AI" (not AGI), documentation and technical conversations become confusing. This map of definitions lets you read a product announcement or a job description and know exactly which layer of the stack they're referring to.

Quick comparison

Term	What it is	Short example
AI	Systems with "intelligent" behavior (perceiving, reasoning, learning, acting)	Any system that decides or learns
ML	AI that learns from data (subset of AI)	Spam filter, recommendations, prediction
DL	ML with deep neural networks (subset of ML)	Image recognition, LLMs, voice
GenAI	AI that generates new content (often via DL)	ChatGPT, DALL·E, code generation

Common mistakes when using these terms

1. "AI" and "ML" as synonyms

Mistake: Saying "this uses AI" and "this uses ML" as if they were the same thing in every case.
Reality: ML is a type of AI. There can be AI without ML (fixed rules); if there's ML, there's AI.

2. "Deep Learning" and "Machine Learning" as opposites

Mistake: "Is this ML or Deep Learning?" as if they were mutually exclusive categories.
Reality: DL is a type of ML. The correct question is "is it classic ML or Deep Learning?".

3. "Generative AI" as something separate from ML/DL

Mistake: Treating GenAI as a branch parallel to ML.
Reality: GenAI is a use (generating content); the underlying technology is usually ML/DL (e.g. LLMs are DL models used to generate).

4. Believing that "AI" always implies "neural networks"

Mistake: Thinking that if there's no neural network, it isn't AI.
Reality: There's AI based on rules, on decision trees, on logic; neural networks are a very important family within ML, not the only form of AI.

5. Confusing "model" with "system"

Mistake: Saying "the model is the AI" as if model and system were the same thing.
Reality: A "model" (for example, GPT-4) is the component that learned from data; a "system" can include the model plus rules, databases, APIs, frontend, etc. The system can have parts that aren't ML (business rules, validations); the model is the part that is ML/DL.

6. Believing that if it uses an LLM API, the whole system is GenAI

Mistake: "My app calls OpenAI, therefore my app is a generative model" or "my system is GenAI because it uses ChatGPT".
Reality: Your system uses GenAI (the LLM via API), but the complete system can have rules, validations, databases, UI, business logic. The correct way to put it is "my system uses an LLM to generate responses" or "my app incorporates GenAI via API". The system isn't the model; the system uses the model.

Gray cases: when to hesitate

Sometimes it isn't clear whether something is "just AI" or "ML", or whether it's "classic ML" or "DL". Some typical cases:

A system that was trained once and no longer learns: If data was used in the past to train a model and then the model was left fixed in production, it's still ML (it learned from data); what's missing is "continuous learning". If there was never any training and there are only fixed rules, it's AI without ML.
Recommendations with rules + ML: Many products mix business rules (e.g. "never recommend X to minors") with an ML model (e.g. personalized recommendations). The complete system is AI; the part that learns from data is ML. You don't need to label the whole product; it's enough to know there's an ML layer.
"Rule-based AI": Some systems are called "AI" but only have "if-then" rules. By convention, they're considered AI (behavior we associate with intelligence) but not ML (they don't learn from data). In practice, when someone says "AI" about modern products, they usually mean ML; if you want to be precise, ask "does it learn from data or are they fixed rules?".
Tools that use LLMs behind the scenes: If a tool (e.g. a writing assistant) uses ChatGPT or similar via API, the tool incorporates AI (and ML, DL, GenAI) even though you don't see the model. Classify the tool as "uses GenAI" or "uses an LLM" when applicable.
AutoML and no-code platforms: Platforms that train models for you (e.g. Google AutoML, Azure ML Studio) without you writing training code. Is it ML? Yes, because there's training with data. The fact that you don't write the code doesn't change that the system learns from data; the platform just automates the ML process.
Pre-trained models (transfer learning): If you use an already-trained model (for example, an ImageNet vision model) and adjust it with your data (fine-tuning), that's ML and DL. If you use the model as-is without training or adjusting it, you're still using ML (the model learned from data in the past); what you don't do is train, but the artifact is still a product of ML.
Hybrid systems (rules + model): A product that is 70% business rules and 30% ML model. How do you classify it? The complete system is AI; the part that learns is ML. In practice, describe the system as "combines rules with ML" or "uses business rules plus ML recommendations". You don't need to force a single label onto the whole product; what matters is identifying which parts are which.
Agents with LLM + tools: An agent that uses an LLM to decide what to do and then calls APIs or runs code. Is it GenAI? It depends on what it does: if it generates text to decide, yes (the LLM is GenAI); if it only executes actions based on rules after parsing, the "executing" isn't generative. In practice, if the decision core is an LLM, the agent "uses GenAI for orchestration".

Questions to reflect on

Use these questions to check that you've internalized the definitions. There's no single correct answer; what matters is that you use the concepts consistently.

If a system "learned" from data just once (training) and was then left fixed in production without updates, is it ML? Why or why not? Does it matter if it's never updated again after going to production?
A product is 80% business rules and 20% ML model. How do you classify it? Is it AI? Is it ML? Both? Justify in one sentence.
If I use a tool that internally calls the OpenAI API (GPT-4), is the tool "GenAI" or does it "use GenAI"? Is there a difference between "being GenAI" and "using GenAI"? How would you describe the tool in a technical conversation?
A vision model was trained on ImageNet and I use it as-is without adjusting or training it. Is it ML? Does it matter that I didn't train the model? Is the model still an ML artifact even though I only use it?
A system generates images with Stable Diffusion but also applies fixed rules (e.g. "never generate violent content"). Is the system GenAI or "GenAI + rules"? How would you describe the architecture in one sentence?
If someone says "this uses AI", can you assume it uses ML? In what cases does "AI" not imply "ML"? Give an example of AI without ML.

Goal: If you can answer 4-5 of these questions consistently (using the pyramid definitions and the gray cases), you're ready to classify products in exercise 06.

How to explain it in one sentence (for interviews or conversations)

AI: "Systems that display intelligent behavior: perceiving, reasoning, learning or acting toward a goal."
ML: "AI that improves with data; not just fixed rules."
DL: "ML that uses neural networks with many layers to learn complex representations."
GenAI: "AI that generates new content (text, image, audio, etc.), often with large Deep Learning models."

How to explain it in an interview (examples): If they ask you "What's the difference between AI and ML?" you can say: "AI is the broad concept: systems that display intelligent behavior. ML is a type of AI where the system learns from data instead of relying only on programmed rules. All ML is AI, but not all AI is ML; for example, a chatbot with fixed rules can be AI but not ML." If they ask you "What is an LLM?" you can say: "An LLM is a large language model: it's Deep Learning (deep neural networks) trained on a lot of text to generate or complete text. It's a type of GenAI because it generates new content, and it's narrow AI because it does one family of tasks (language) very well but not any intellectual task."

Exercises

Exercise 1: Classify systems

Classify each one as AI only, ML, DL or GenAI (there can be more than one label when applicable):

A chatbot that only responds with predefined phrases based on keywords.
A system that recommends movies based on what you watched before.
ChatGPT.
A program that solves sudokus with logical rules, without learning from data.
A model that generates images from text (e.g. DALL·E).

See solution

AI only (rules). "Intelligent" behavior in the broad sense, but it doesn't learn from data → it isn't ML.
ML (and therefore AI). It learns from data (your history) to recommend; it can be classic ML or use networks, not necessarily DL.
DL and GenAI (and therefore ML and AI). A large language model (deep neural network) that generates text.
AI only (logic/rules). Automated reasoning, but no learning from data → it isn't ML.
GenAI (and typically DL, ML, AI). It generates new content (an image); image models are usually Deep Learning.

Exercise 2: The odd sentence out

One of these sentences is incorrect or confusing. Which one and why?

A) "Machine Learning is a type of Artificial Intelligence."
B) "Deep Learning is the opposite of Machine Learning."
C) "Generative AI generates new content; it often uses Deep Learning models."
D) "Not everything called AI uses Machine Learning."

See solution

B) is the incorrect one. Deep Learning is not the opposite of Machine Learning; it's a subset of ML. The opposite would be something like "systems that don't learn from data" (e.g. fixed rules), not "Deep Learning".

Exercise 3: Define without jargon

Explain to someone without a technical background (in 2–3 sentences) the difference between "artificial intelligence" and "machine learning".

See answer guide

Possible guide: "Artificial intelligence is the broad concept: machines or programs that do things we associate with intelligence, like deciding or reasoning. Machine learning is one specific way of doing AI: instead of programming all the rules by hand, the system learns patterns from lots of data and improves with experience. So all machine learning is artificial intelligence, but not all artificial intelligence uses machine learning."

Exercise 4: ML or not?

A customer service system that, based on the words you type, chooses among 5 predefined responses (without learning from new data after going to production). Is it Machine Learning? Justify in one sentence.

See solution

It isn't ML (as described). The behavior is fixed by rules or by a model that no longer updates with new interaction data. There's no continuous "learning" from data. If at some point it was trained with data and then left fixed, there was ML at that moment; but if it only assigns responses by fixed rules, it's AI without ML.

Exercise 5: The pyramid in reverse

Order from most general to most specific: GenAI, DL, AI, ML. Justify in one sentence why that order.

See solution

Order: AI → ML → DL → GenAI (from more general to more specific in terms of "how it's built"; GenAI is a "use" not a level, but if you think of it as "what it does", GenAI is more specific than "AI" in general).
Justification: AI is the umbrella; ML is a subset (it learns from data); DL is a subset of ML (deep networks); GenAI is a type of use of AI (generating content), which today is usually implemented with DL. So from "more general" to "more specific" in architecture it would be AI → ML → DL; GenAI is "what it does" (generate), not "how it's built", but if you include it in the conceptual pyramid, it goes after DL because many of today's GenAI systems are DL.

Exercise 6: One sentence for each

Write a sentence explaining the difference between (a) ML and DL, (b) GenAI and DL.

See answer guide

(a) ML and DL: "ML is AI that learns from data; DL is a type of ML that uses neural networks with many layers to learn complex representations. All DL is ML, but not all ML is DL (there's classic ML with trees, regressions, etc.)."
(b) GenAI and DL: "GenAI is a use of AI: generating new content (text, image, etc.). DL is a type of ML (deep networks). Many of today's GenAI systems are implemented with DL models (for example, LLMs), but GenAI describes what the system does (generate) and DL describes how it learns (deep networks)."

Summary

One sentence per concept: AI = systems with intelligent behavior; ML = AI that learns from data (subset of AI); DL = ML with deep neural networks (subset of ML); GenAI = AI that generates new content (often via DL). All ML is AI; all DL is ML; not all AI is ML; not all ML is DL.

Key points:

AI = broad concept: systems with intelligent behavior (perceiving, reasoning, learning, acting).
ML = subset of AI: systems that learn and improve from data.
DL = subset of ML: learning with deep neural networks.
GenAI = use of AI (often DL) to generate new content (text, image, audio, etc.).
Not all AI is ML; all ML is AI; all DL is ML; many of today's GenAI systems are based on DL.
Avoid using "AI" and "ML" as synonyms and "DL" and "ML" as opposites; use the subset map to be precise.

Examples of systems by category

To lock in the definitions, here are concrete examples by category:

AI only (rules or logic, no ML): A chatbot that responds with predefined phrases based on keywords; a program that solves sudokus with logical rules; an automated reply system "if X then Y" with no learning. They're AI in the broad sense (behavior we associate with intelligence) but they don't learn from data.
ML (and therefore AI): A spam filter that learned from labeled emails; a movie recommendation system that learns from your history; a model that predicts sales from historical data. They can be classic ML (trees, regressions) or Deep Learning; either way, they learn from data.
DL (and therefore ML and AI): Object recognition in images; speech recognition; machine translation with neural networks; language models (LLMs like GPT, Claude). They use networks with many layers; they usually require a lot of data and a lot of compute.
GenAI (and often DL, ML, AI): ChatGPT, Claude, Gemini (they generate text); DALL·E, Midjourney, Stable Diffusion (they generate images); tools that generate code from a description. They generate new content; today they're usually implemented with Deep Learning models (LLMs, image networks).

Use this list as a reference when you classify products in lesson 06: if the system generates new content, it's GenAI (and probably DL); if it only classifies or recommends, it's ML (and it may or may not be DL); if it only follows fixed rules, it's AI without ML.

Usage patterns in the industry

When you hear these terms in documentation, job postings or conversations, they usually mean the following:

"ML model" or "machine learning model": It almost always refers to a system that learned from data (supervised, unsupervised or reinforcement). It can be classic ML (trees, regressions) or Deep Learning (networks); if they don't specify, assume it could be either.
"Language model" or "LLM": It refers to a Deep Learning model trained on a lot of text to generate or complete text. It's a type of GenAI (it generates content). Examples: GPT-4, Claude, Llama. When they say "LLM" or "large language model", they're talking about DL + GenAI (text).
"Generative model" or "GenAI": It refers to systems that generate new content (text, image, audio, code). Today they're usually large Deep Learning models (LLMs, image models). Don't confuse "generative" with "general"; generative models are narrow AI (they're very good at generating in their domain), not AGI.
"AI" on its own: In products and marketing it usually means "it uses ML or DL somewhere". In more technical contexts it can also include systems with only rules; if you want precision, ask "does it learn from data or are they fixed rules?".
"Fine-tuning": Taking a pre-trained model (for example, an LLM or a vision model) and adjusting it with data specific to your domain. It's ML and DL (you're still training, even if less than training from scratch). Don't confuse it with "using a model via API without training"; fine-tuning implies additional training.
"Transfer learning": Using knowledge learned on one task (e.g. ImageNet) for another task (e.g. classifying your own images). It's ML and DL; you reuse learned representations. Very common in vision and language; that's why many current models aren't trained from scratch but fine-tuned from a pre-trained checkpoint.
"Foundation model" or "base model": A very large model pre-trained on lots of data (e.g. GPT, BERT, CLIP) that serves as a base for many tasks. It's DL; it's usually GenAI (if it generates) or a representation model (if it encodes). Example: GPT-4 is a foundation model for language; you can use it as-is or fine-tune it for your domain.
"Multimodal": A model that handles more than one type of data (e.g. text + image). It's DL (deep networks trained with data from multiple modalities). Example: GPT-4V (text + image), CLIP (text + image). It's still narrow AI (limited to the modalities it was trained on), not AGI.

Connection with the rest of the module

What you learned in this lesson is used like this in the rest of Module 1: in lesson 03 (history) you'll see how the field went from rules to ML and then to DL; without the definitions from this lesson, that history would be confusing. In lesson 04 (types) you'll see narrow vs general; "narrow" includes everything that is ML/DL/GenAI today. In lesson 05 (current state) you'll see what's possible today; almost everything mentioned (LLMs, image generation, recommendations) is ML, often DL, and often GenAI. In lesson 06 (exercise) you'll classify products; you'll use this lesson to decide whether each product is AI, ML, DL or GenAI. Don't skip this lesson; it's the foundation of the module.

Visual summary: the pyramid in a table

To review at a glance:

Level	Name	What it is	Contains	Short example
1 (broadest)	AI	Systems with intelligent behavior	Everything else	Any system that decides or learns
2	ML	AI that learns from data	DL, and often GenAI (when implemented with DL)	Spam filter, recommendations
3	DL	ML with deep neural networks	Many of today's GenAI systems (LLMs, image)	Image recognition, LLMs
4 (use)	GenAI	AI that generates new content	—	ChatGPT, DALL·E

Remember: GenAI isn't a "level" in the pyramid of "how it's built"; it's "what it does" (generate). But many of today's GenAI systems are implemented with DL, which is why it appears in the table as a "use" that usually relies on DL.

Frequently asked questions about definitions

Is everything called "AI" in products ML?
Almost everything. When a product says it "has AI" or "uses AI", in practice it usually means it uses a model that learned from data (ML), often Deep Learning. Systems with only fixed rules (no learning) are less common in new products; if in doubt, ask "does it learn from data?".

Are Deep Learning and Machine Learning the same thing?
No. Deep Learning is a type of Machine Learning (the one that uses deep neural networks). All Deep Learning is Machine Learning; not all Machine Learning is Deep Learning (there's classic ML with trees, regressions, SVM, etc.).

Is Generative AI the same as an LLM?
No. GenAI is the use (generating content: text, image, audio, etc.). An LLM is a type of model (large language model) that's usually used for GenAI (generating text). There are other types of GenAI that aren't LLMs (for example, models that only generate images). So "LLM" is a type of model; "GenAI" is what the system does (generate).

Can I say "AI" when I mean an ML model?
Yes. All ML is AI, so saying "this product uses AI" when it uses an ML model is correct. What you shouldn't do is say "this isn't AI, it's ML" as if they were mutually exclusive; ML is a subset of AI.

Quick decision table: is it ML?

When you're unsure whether a system is Machine Learning, ask yourself:

Question	If the answer is yes	If the answer is no
Did the system improve or adjust using data (examples, history)?	Probably ML	Probably not ML (fixed rules)
Was the behavior "trained" at some point with data?	Yes, ML	Not ML (unless there was training and then it was left fixed)
Does it only follow hand-written rules (if-then) without learning?	It isn't ML	It's AI without ML

If the system never used data to adjust its behavior, it isn't ML. If at some point it was trained with data (even if it was later left fixed), it is ML.

Connection with the rest of the guide (Modules 2-8)

What you learned here is used throughout the rest of the guide: in Module 2 (Machine Learning) you'll see types of ML (supervised, unsupervised, reinforcement) and the training vs inference distinction; without this lesson, "ML" would be a black box. In Modules 4 and 5 (Transformers, LLMs) you'll see that LLMs are a type of DL used for GenAI; without the definitions from this lesson, "Transformer" or "LLM" would be acronyms with no context. In Module 6 (APIs) you'll see that when you "call a model" you're using a type of AI (often ML, DL, GenAI); without this lesson, API documentation would be more confusing. In Module 7 (AI Engineering) you'll see that the role consists of integrating models (narrow AI, often ML/DL/GenAI) into products; without this lesson, the "AI Engineer vs ML Engineer" comparison would be less clear. In short: this lesson is the foundation of the map; don't skip it or skim it if it's your first time.

Additional resources

Google ML Crash Course: Introduction to ML — Official introduction to ML: what ML is, types (supervised, unsupervised, etc.) and examples. In English. Useful for going deeper after this lesson.
IBM: What is Machine Learning? — Definitions and types of ML (supervised, unsupervised, reinforcement) with usage examples. In English. Complements the "What is ML?" section of this lesson.
DeepLearning.AI: What is Deep Learning? — Definition and context of DL: why "deep", relationship with ML, examples. In English. Useful for reviewing the AI → ML → DL pyramid.
IBM: What is Generative AI? — What GenAI is, examples (text, image, code), and how it differs from other types of AI. In English. Although it's on IBM Topics, it's neutral and clear.
Russell & Norvig, Ch. 1 — Formal definitions of AI in the field's reference book. Chapter 1 expands on "what is AI" and the different definitions (acting humanly, thinking humanly, etc.). In English; requires access to the book.
Elements of AI – Chapter 1 — Free course in several languages (including Spanish) on the fundamentals of AI; the first chapter covers definitions similar to this lesson. Useful for practicing with additional classification exercises.

← Previous1. Module Introduction: What Is AI Really?Next →3. A Brief History of AI: From Turing to ChatGPT