Module 1: What Is AI Really?

6. Exercise: AI all around you

Description

In this lesson you'll apply what you learned in Module 1: identifying and classifying artificial intelligence in products and services you use every day. There's no code or new theory; it's an exercise in observation and reflection that reinforces the definitions (AI vs ML vs GenAI), the history (what kinds of systems exist today) and the current state (what's possible in real products).

The goal is for you to move from "this must be AI" to "this is narrow AI, probably ML, and it does X". That prepares you to speak precisely and to connect the module's concepts with the real world. If you're going to do the AI Engineering Bootcamp or work with AI APIs, this exercise helps you be clear about what kind of systems you're integrating and how to describe them.

Goal of the exercise

By the end you'll have:

Identified at least 5 products or services you use that incorporate AI.
Classified each one (AI in general, ML, Deep Learning, GenAI when applicable).
Described in one sentence what the AI does in each product (what task it solves).
Reflected on what AI can do today that it couldn't a few years ago (optional but recommended).

There's no single correct answer; what matters is that you use the module's concepts consistently and briefly justify your classifications. The self-assessment guide at the end lets you check that you've covered everything needed.

Quick variant (if you're short on time): If you only have 15-20 minutes, list 3 products instead of 5, and answer the 3 questions for just 1 of them instead of 3. That gives you the minimum practice; later on you can expand to 5 products when you have time.

Step 1: List of products

Make a list of at least 5 products or services you use (or have used) that you believe incorporate artificial intelligence. They can be:

Apps (mobile or web): social networks, email, maps, music, video, banking, shopping, etc.
Voice assistants (Alexa, Siri, Google Assistant, etc.).
Work tools (search, proofreading, translation, meetings, code, etc.).
Consumer products (recommendations from Netflix, Spotify, Amazon, etc.).
Generation tools (ChatGPT, Claude, Copilot, DALL·E, etc.).

Example list (reference only):

Netflix — series and movie recommendations
Gmail — spam filter and reply suggestions
ChatGPT — text generation and conversation
Google Maps — routes and arrival times
Spotify — discovery and playlists (Discover Weekly, etc.)

More ideas if you're short on products:

Search (Google, Bing, DuckDuckGo): result ranking, direct answers, sometimes an LLM.
Proofreading and writing (Grammarly, Word, Google Docs): grammar and style suggestions.
Translation (Google Translate, DeepL): machine translation.
Meetings (Zoom, Meet, Teams): transcription, summaries, captions.
Code (GitHub Copilot, Cursor, Codeium): autocomplete and code generation.
Social networks (Instagram, TikTok, YouTube): content recommendation, moderation, filters.
Banking and finance: fraud detection, customer service chatbots.
Shopping (Amazon, Mercado Libre): recommendations, image search, dynamic pricing.

Your list can be different; what matters is that they're real products you know and that you can describe what the AI does in each one.

Step 2: Classify each product

For each product on your list, fill in a table (or list) with:

Product	What does the AI do? (1 sentence)	AI? (yes/no)	ML? (yes/no)	DL/GenAI? (if applicable)	Brief justification

Quick guide:

AI? If the system perceives, decides, learns or acts in a way we'd associate with "intelligence" (even in a bounded domain), it's usually yes.
ML? If the system improves or personalizes itself with data (e.g. your history, clicks, preferences), it's usually ML.
DL/GenAI? If you know or suspect it uses deep neural networks (speech, image, language recognition) or generates new content (text, image), note it down.

Example row (reference only):

Product	What does the AI do?	AI?	ML?	DL/GenAI?	Justification
Netflix	Recommends what to watch based on what you watched and rated	Yes	Yes	Probably DL in recommendation	It learns from data (ML); recommendation at scale usually uses networks (DL).

You don't need to know the internal technology with certainty; it's enough to apply the module's definitions and explain your reasoning in one sentence. If you're not sure whether it uses DL or GenAI, you can write "probably" or "possible"; what matters is that you justify it with the criteria (does it generate new content? does it recommend based on data? does it recognize speech or images?).

Quick reminder of the definitions (Lesson 02):

AI: A system that perceives, decides, learns or acts in a way we'd associate with "intelligence" in a domain.
ML: A system that improves or personalizes itself with data (history, clicks, preferences).
DL/GenAI: Deep neural networks (speech, image, language) or generation of new content (text, image, audio).

Reminder about types (Lesson 04): Everything that exists today in products is narrow AI (one or a few tasks) and weak AI (no consciousness attributed to it); if you classify something as "general" or "strong", review lesson 04.

Step 3: Three cases to go deeper

Choose three of your products and, for each one, write 2–3 sentences answering:

What would happen if it didn't have AI? (What would the product be like without that capability?)
Is it narrow AI, or could you argue it comes close to something "general"? (Remember: today everything is narrow; the question is whether you see any nuance.)
What limit or error have you noticed (if applicable)? (E.g. recommendations that don't fit, incorrect answers, hallucinations.)

Example (reference only) — Gmail:

Without AI: there'd be no automatic spam filter or reply suggestions; you'd have to mark spam by hand and write every reply.
It's narrow AI: it only does specific tasks (classifying email, suggesting phrases); it doesn't do any intellectual task.
Limit: sometimes the spam filter gets it wrong (false positives/negatives) or the suggested replies don't fit the tone you want.

Example (reference only) — ChatGPT:

Without AI: it would just be a form with no generated responses; there'd be no conversation or text generation.
It's narrow AI: very good at language (generating, summarizing, translating) but not at any intellectual task; it fails at facts, long reasoning and domains outside its training.
Limit: hallucinations (making up facts or references), sensitivity to the prompt, limited context window.

Example (reference only) — Netflix:

Without AI: there'd be no personalized recommendations; you'd see a catalog ordered by fixed criteria (date, global popularity).
It's narrow AI: it only recommends content; it doesn't do any intellectual task.
Limit: sometimes it recommends things that don't fit your taste or repeats one genre too much.

Step 4: Reflection (optional but recommended)

Answer in a short paragraph (5–10 lines):

What can AI do today in the products you use that it couldn't do (or didn't do well) about 5 years ago?

You can think about: text and image generation, conversational assistants, translation, recommendations, speech or image recognition, assisted coding, etc. You don't need to be exhaustive; it's enough to cite 2–3 examples and connect them with Module 1's idea of the "current state" (more data, more compute, large models, APIs).

Answer guide (only if you get stuck): Five years ago you didn't have an assistant that generated long coherent text or images from descriptions; Spotify's or Netflix's recommendations already existed but have improved. What changed is generation (GenAI) and the quality of conversation and assisted coding, thanks to large models (Transformers, LLMs) and APIs that let them be integrated into products. Connect that with "more data, more compute, new architectures" from lesson 03.

Self-assessment guide

When you finish, review:

You have at least 5 products listed.
For each one you've said what the AI does in one sentence.
You've classified each one (AI, ML, DL/GenAI when applicable) and given a brief justification.
You've chosen 3 products and answered the three questions (what would happen without AI, narrow vs general, limits).
(Optional) You've written the reflection paragraph about what AI can do today that it couldn't a few years ago.

If something doesn't add up, go back to the corresponding lesson (02 definitions, 04 types, 05 current state) and adjust your answer. There's no single solution; what matters is that you use the module's concepts consistently.

Criteria for "well done":

You have at least 5 products and for each one you've said what the AI does in a clear sentence.
You've classified each one (AI, ML, DL/GenAI when applicable) and given a brief justification (not just "yes/no").
For 3 products you've answered the three questions (what would happen without AI, narrow vs general, limits) consistently with lessons 02, 04 and 05.
You haven't said that any product is "AGI" or "general AI" (everything current is narrow; review lesson 04 if in doubt).
If you did the optional reflection, you've connected 2–3 examples with the "current state" (more data, compute, large models, APIs).

Frequently asked questions

Can I use products I don't use myself but that I know about?
Yes; what matters is that you can describe what the AI does in each one. If you're not sure, pick another one you know better.

What do I do if I don't know whether a product uses ML or DL?
Apply the criterion: if it personalizes itself with data (your history, clicks), it's usually ML. If you know or suspect it uses deep networks (speech, image, language, generation), note it as DL/GenAI. You can write "probably DL" if you're not sure.

Can I list more than 5 products?
Yes; 5 is the minimum. If you want to go deeper, you can expand the list and choose more than 3 for Step 3.

Is the Step 4 reflection mandatory?
No; it's optional but recommended. It reinforces the connection between the "current state" (lesson 05) and the products you use.

What do I do if I can't find limits in a product?
You can write "I haven't noticed obvious limits" or think about what could fail (e.g. recommendations that don't fit, answers that sometimes aren't correct). Not every product shows its failures obviously; what matters is that you reflect with the module's concepts.

Optional step: Compare with someone else

If you have the chance, share your list and classification with another person (a classmate, a friend) who has also done the exercise. Compare:

Do you agree on which products incorporate AI?
Do you agree on the classification (AI, ML, DL/GenAI) for the same products?
Have you noticed different limits in the same product?

There's no single correct answer; the goal is for you to use the module's concepts consistently and to see that different people can justify slightly different classifications with the same criteria. If you disagree, go back to the definitions in lesson 02 and the types in lesson 04.

Example of a complete answer (reference)

Step 1 – List:

Spotify 2. Google Search 3. ChatGPT 4. Waze 5. Instagram (explore/reels)

Step 2 – Classification (excerpt):

Product	What does the AI do?	AI?	ML?	DL/GenAI?	Justification
Spotify	Recommends music and creates personalized playlists	Yes	Yes	DL possible in recommendation	It learns from listens (ML); modern systems usually use networks.
Google Search	Ranks and displays results, answers direct questions	Yes	Yes	Yes (ranking, sometimes an LLM)	It learns from clicks and queries (ML); ranking and answers can use DL/LLMs.
ChatGPT	Generates text and holds a conversation	Yes	Yes	Yes (LLM, GenAI)	It generates new content (GenAI); it's a large language model (DL).
Waze	Predicts traffic and suggests routes	Yes	Yes	DL possible in prediction	It learns from traffic data (ML); prediction can use networks.
Instagram	Recommends reels and content in Explore	Yes	Yes	Yes (recommendation with networks)	It learns from interactions (ML); recommendation at scale usually uses DL.

Note: In every case the justification connects with Module 1's definitions: does it perceive, decide, learn or act? (AI) Does it personalize itself with data? (ML) Does it use deep networks or generate new content? (DL/GenAI). You don't need to know the internal technology with certainty; it's enough to apply the criteria and explain your reasoning.

Step 3 – Three cases (excerpt for ChatGPT):

Without AI it would just be a form with no generated responses. 2. It's narrow AI: very good at language but not at any task. 3. Limit: sometimes it hallucinates facts or makes up references.

Step 4 – Reflection (example):
"Five years ago I didn't have an assistant that generated long coherent text or images from descriptions; Spotify's or Netflix's recommendations already existed but have improved. What changed is generation (GenAI) and the quality of conversation and assisted coding, thanks to large models and APIs. I connect that with lesson 03 (more data, more compute, architectures like Transformers) and with lesson 05 (current state: what's possible today in real products)."

Why this example is good: It uses the module's concepts (GenAI, large models, APIs), connects with the history (lesson 03) and with the current state (lesson 05), and gives concrete examples (text, image, recommendations). You don't need to be exhaustive; 2–3 well-connected examples are enough.

Connection with the rest of the guide

This exercise closes Module 1 and connects with what's coming:

Module 2 (Machine Learning): When we talk about "learning from data", you'll already have seen in real products how ML is used (recommendations, personalization, classification). This exercise gives you concrete examples to contrast with the ML concepts (supervised, unsupervised, training vs inference). When you read "training", think about how Gmail's spam filter learned from labeled emails; when you read "inference", think about how the already-trained classifier now decides whether a new email is spam.
Modules 4 and 5 (Transformers, LLMs): The products you classified as GenAI or that use language (ChatGPT, Copilot, etc.) are the ones you'll see in detail in those modules.
Module 6 (APIs): Many of the products you listed consume models via API; when you see how an LLM is integrated into an application, you'll recognize the pattern.
Module 7 (AI Engineering): Your job will be to design and integrate systems like the ones you identified: narrow AI on specific tasks, with known limits and careful design.

Tips for the reflection (Step 4)

If you do Step 4 (reflecting on what AI can do today that it couldn't a few years ago), these points can help you structure the paragraph:

Content generation: Five years ago you didn't have assistants that generated long coherent text (summaries, drafts, code) or images from descriptions at today's quality. That changed with large models (Transformers, LLMs) and APIs that let them be integrated into products.
Conversation and code: Today's chatbots (ChatGPT, Claude, etc.) and code assistants (Copilot, Cursor) are a qualitative leap over what existed a few years ago (rule-based assistants, basic autocomplete). Connect that with "more data, more compute, new architectures" from lesson 03.
Recommendations and personalization: Recommendations (Netflix, Spotify, Amazon) already existed years ago, but they've improved with more data and more capable models (Deep Learning, embeddings). Not everything is "new"; part of it is the evolution of what was already there.
Speech and image recognition: Speech transcription and image recognition (OCR, scene description) have improved a lot; today it's reasonable to use voice or images as input in products. Connect with the "current state" from lesson 05.

You don't need to cover everything; choose 2–3 examples and connect them with the module's concepts (definitions, history, current state).

Submission template (optional)

If you want to structure your answer clearly, you can use this template. It isn't mandatory; it's an aid so you don't forget any step.

Step 1 – List of products (minimum 5):

Product 1: [name] — [what the AI does, in one sentence]
Product 2: [name] — [what the AI does, in one sentence]
Product 3: [name] — [what the AI does, in one sentence]
Product 4: [name] — [what the AI does, in one sentence]
Product 5: [name] — [what the AI does, in one sentence]

Step 2 – Classification (table or list):

For each product: What does the AI do? | AI? (yes/no) | ML? (yes/no) | DL/GenAI? (if applicable) | Brief justification.

Step 3 – Three cases to go deeper:

Product A: (1) What would happen without AI? (2) Narrow or general? (3) Any limit or error noticed?
Product B: (1) What would happen without AI? (2) Narrow or general? (3) Any limit or error noticed?
Product C: (1) What would happen without AI? (2) Narrow or general? (3) Any limit or error noticed?

Step 4 – Reflection (optional):

[A paragraph of 5–10 lines: What can AI do today in the products you use that it couldn't a few years ago? Connect with the module's definitions, history or current state.]

When you finish, review it with the self-assessment guide and with this lesson's "well done" criteria.

More product examples to expand your list

If you want to expand your list beyond the examples already cited, here are more ideas by category:

Communication and writing: Slack (summaries, semantic search), Notion (generation, summarization), Grammarly (grammar and style), Google Docs (suggestions, proofreading), Outlook (reply suggestions).

Search and knowledge: Google Search, Bing, Perplexity (answers with sources), Wikipedia (sometimes AI tools for editing).

Image and design: Canva (generation, suggestions), Adobe Firefly, DALL·E, Midjourney, Stable Diffusion, Lensa (portraits).

Voice and audio: Siri, Alexa, Google Assistant, Otter (meeting transcription), Descript (audio/video editing with AI).

Code and development: GitHub Copilot, Cursor, Codeium, ChatGPT/Claude for code, Tabnine.

Productivity: Notion AI, Microsoft Copilot (Word, Excel, etc.), Google Workspace (Docs, Sheets, Gmail with AI).

Shopping and recommendations: Amazon, Mercado Libre, eBay (recommendations, image search), Shein, Zalando.

Health and wellness: Meditation apps with personalization, wearables with pattern detection, triage chatbots (with limits and supervision).

Finance: Fraud detection (banks), roboadvisors, customer service chatbots.

Choose products you know and can describe what the AI does in each one; they don't have to come from this list.

One-page summary

Goal: Identify at least 5 products with AI, classify them (AI, ML, DL/GenAI) and reflect on what the AI does and what limits it has.
Step 1: List of at least 5 products; for each one, a sentence about what the AI does.
Step 2: Classification (table or list): What does the AI do? AI? ML? DL/GenAI? Brief justification.
Step 3: Three products: (1) What would happen without AI? (2) Narrow or general? (3) Any limit or error noticed?
Step 4 (optional): Reflection: What can AI do today that it couldn't a few years ago? Connect with definitions, history or the current state.
Criteria for "well done": At least 5 products; classification with justification; 3 products with the three questions; not saying any product is AGI; optional reflection connected with the module.
Reminder: Everything current is narrow AI and weak AI; if you classify something as "general", review lesson 04.
Connection with the guide: This exercise closes Module 1 and connects with ML (Module 2), Transformers/LLMs (4-5), APIs (6) and AI Engineering (7).
Optional template: Use the submission template if you want to structure your answer; it isn't mandatory.
Compare with someone else (optional): If you can, share your list and classification with another person and compare criteria.
Resources: Elements of AI, AI For Everyone (the global course platform), Product Hunt (AI tools), McKinsey (use cases), AI Index (Stanford), EU AI Act. See the "Additional resources" section at the end of this lesson for links and descriptions.

Notes for going deeper

Why isn't there a single correct answer? Because the classification (AI, ML, DL/GenAI) depends on what you know or suspect about the internal technology; certainty isn't required. What matters is that you use the module's concepts consistently and justify briefly. Two people can classify the same product slightly differently with the same criteria.

What to do if you're not sure whether a product uses ML or DL? Apply the criterion: if it personalizes itself with data (history, clicks), it's usually ML. If you know or suspect it uses deep networks (speech, image, language, generation), note it as DL/GenAI. You can write "probably DL" if you're not sure; what matters is the justification.

How to use this exercise in the bootcamp or in another guide? When you integrate a model (API, RAG, agent), remember that you're integrating narrow AI on a specific task. This exercise helps you be clear about what kind of system it is and how to describe it (to product, business or in a report).

What to do if you compare with someone else and you don't agree? Go back to the definitions (lesson 02) and the types (lesson 04). There's no single correct answer; what matters is that each of you justifies with the module's criteria. If you disagree on "is it ML?", ask: does it personalize itself with data? If you disagree on "is it narrow or general?", remember: everything current is narrow.

Questions to reflect on

Why at least 5 products? So you practice the classification several times and see that different products fit different combinations (AI without ML, AI + ML, AI + ML + DL/GenAI). With 5 or more you cover variety.

Why three products to go deeper? So you reflect on what would happen without AI, narrow vs general and limits in concrete products. Three is a good balance between depth and time.

Why is the reflection (Step 4) optional but recommended? Because it reinforces the connection between the "current state" (lesson 05) and the products you use. It isn't mandatory but it helps lock in the ideas.

What to do if you can't find limits in a product? You can write "I haven't noticed obvious limits" or think about what could fail (recommendations that don't fit, answers that sometimes aren't correct). Not every product shows its failures obviously.

How to connect the reflection (Step 4) with the module? Cite 2–3 concrete examples (text generation, images, recommendations, assisted coding) and connect them with "more data, more compute, large models, APIs" (lesson 03) or with the "current state" (lesson 05). You don't need to be exhaustive; 2–3 well-connected examples are enough.

Summary

This exercise applies the definitions (AI, ML, DL, GenAI), the types (narrow AI) and the current state from Module 1.
Goal: identify at least 5 products with AI, classify them with criteria and reflect on what the AI does and what limits it has.
There's no single correct answer; what matters is using the concepts consistently and justifying briefly.
With this you close Module 1 and have a foundation for Module 2 (Machine Learning) and the rest of the guide.

Additional resources

Elements of AI – Chapter 1: What is AI? — A free course with similar exercises identifying AI in products; several languages; complements this exercise.
AI For Everyone (Coursera) — Andrew Ng's non-technical introductory course; Module 1 includes examples of AI in industries (health, transport, etc.) you can add to your reflection (in English).
Product Hunt – AI tools — A listing of products that incorporate AI; useful for expanding your list if you want more examples (in English).
McKinsey: AI use cases by industry — Examples of AI use by industry; it helps connect "products I use" with "products in companies" (in English).
Stanford HAI: AI Index – Applications — Data on AI adoption across different sectors; context for your reflection on "what AI can do today" (in English).
EU AI Act – High-risk applications — A listing of AI applications considered high-risk; useful for seeing which products you use (or similar ones) may be in the regulatory spotlight (in English).

← Previous5. AI in 2024-2026: Current state, what's possible and what isn't Next →1. Module Introduction: Machine Learning Fundamentals