Put It Together: Sort Real Products on the Map

You now have the full map: AI on the outside, machine learning inside it, deep learning inside that, and generative AI at the center. The last step is to use it.

In this lesson you will learn three quick questions that place almost any product on the map, practice on everyday tools, and turn the skill into something useful: reading "AI-powered" claims with a sharper eye.

What You'll Learn

• Three questions that place any system on the AI map
• How to sort ten everyday products, with the reasoning for each
• Why one product often uses several layers of the map at once
• Questions to ask when a product claims to be "AI-powered"

Three questions that place anything

Ask these in order and stop at the first "no."

1. Did it learn its behavior from examples?
2. Does it work on raw images, audio, or free text?
3. Does it create new content?

Question 2 is a rule of thumb, not a law. Some classic models work on text and some deep models work on tables. But in practice it points you to the right box most of the time.

Practice: sort ten everyday products

Try placing each one yourself before reading the answer.

1. A calculator app. Did it learn? No. It follows exact math rules. It is not even usually called AI. Rules, not AI.

2. Finding the fastest route in a maps app. The route search itself is a classic algorithm working through a road network. Rules-based AI. (Hold that thought for number 3.)

3. The traffic prediction in that same maps app. It learns from past traffic patterns to guess how busy each road will be. That is learning from examples, so it is machine learning, and often deep learning at large companies.

4. Your email spam filter. It learned from huge numbers of emails people marked as spam. Modern filters read message text and are increasingly deep learning, while older ones used classic machine learning on hand-picked signals. Machine learning at minimum.

5. A bank's credit risk score. It learns from past loan records stored as tables: income, repayment history, amounts owed. Decisions often have to be explained. Classic machine learning, frequently wrapped in written rules.

6. Movie or video recommendations. It learns from what millions of people watched and skipped. Machine learning, with the largest platforms leaning heavily on deep learning.

7. Face unlock on your phone. It works on raw camera images and recognizes, rather than creates. Deep learning (predictive).

8. Voice typing on your phone. It turns raw audio into text. Deep learning. It creates text, but only a transcript of what you said, so most people class it as recognition rather than generative AI.

9. "Remove background" in a photo app. It detects which pixels are the subject. Deep learning (predictive). But "generative fill," which paints in brand-new scenery, is generative AI. Same app, two boxes.

10. ChatGPT, Claude, or Gemini. Learned from data, works on raw text, creates new content. Generative AI, built on a large language model.

Real products use several boxes at once

Notice how often a single product showed up in more than one place. The maps app used classic route search and machine learning traffic prediction. The photo app used predictive deep learning and generative AI.

This is normal. Engineers pick the right tool for each part:

• Written rules where the logic is known and must be exact.
• Classic machine learning for tidy data and explainable scores.
• Deep learning for images, sound, and language.
• Generative AI where new content is actually the goal.

So the most accurate description of a product is often not one label but a short list: "rules for X, a learned model for Y, a generative model for Z."

Reading "AI-powered" claims

"AI-powered" can mean anything from a few if-then rules to a large generative model. The map gives you better questions to ask:

• Which part actually uses AI? A product may have one small learned feature and a lot of ordinary software around it.
• Does it learn, or follow rules? If it learns, from what data? Old or narrow data leads to blind spots.
• Is it predictive or generative? Predictive systems can be checked against right answers. Generative ones need a human to check that the content is correct.
• Can it explain its decisions? This matters most for anything touching money, health, jobs, or legal outcomes.
• Would a simpler approach have done the job? Not every problem needs deep learning, and bigger is not automatically better.

You can also use a chatbot to help you think through a product. Try a prompt like this:

Here is how a product describes its "AI" features: [paste the description].
Based only on this text, which parts sound like rules-based software,
classic machine learning, deep learning, or generative AI? For each,
explain your reasoning and list what information is missing to be sure.

Treat the answer as a starting point, not a verdict. The chatbot only knows what the description says, which is exactly why the "what is missing" part is useful.

Key Takeaways

• Three questions place almost anything: Did it learn from examples? Does it work on raw, messy input? Does it create new content?
• Everyday tools span the whole map, from rules (route search, calculators) to classic ML (credit scores) to deep learning (face unlock) to generative AI (chatbots).
• One product often uses several boxes at once, choosing the right tool for each part.
• "AI-powered" is vague; ask which part uses AI, what it learned from, whether it is predictive or generative, and whether it can explain itself.
• The map is a thinking tool: it helps you predict how a system will behave and where it is likely to fail.