In this lesson: Place any AI product you encounter into the right category.
The word "AI" is used for systems that have almost nothing in common. Four distinctions will let you classify anything you meet.
1. Narrow versus general
Narrow AI does one task. A model that detects tuberculosis in a chest X-ray cannot play chess, write an email, or even read a different kind of scan. Every AI system in commercial use today is narrow — including the chatbots, which are narrow at an unusually wide task.
General AI — one system that can learn any intellectual task a human can — does not exist. When you read predictions about it, you are reading opinion, not a product announcement.
2. Predictive versus generative
| Predictive AI | Generative AI | |
|---|---|---|
| Output | A label, a number, a score | New text, images, audio, code |
| Question | "Which category is this?" | "Make me one of these." |
| Examples | Fraud detection, credit scoring, diagnosis | Chat assistants, image generators |
| Checking it | Compare against known answers | Much harder — there is no single right answer |
Predictive AI has quietly run banks and hospitals for two decades. Generative AI is the newcomer that made everybody notice.
3. What the model works on
- Text — language models. Writing, summarising, answering, translating, coding.
- Vision — images and video. Detecting disease, reading documents, recognising crops from drone photos.
- Audio — speech to text, text to speech, translation.
- Multimodal — several at once. You show a photo and ask a question about it in words.
4. Where the model runs
This matters more than beginners expect, because it decides cost, speed and privacy.
- On the device — face unlock on your phone. Fast, private, small.
- In a data centre, reached over the internet — the large chat assistants. Powerful, costs money per use, and your input leaves your building.
What is genuinely new since 2022
Not the mathematics — the transformer architecture behind today's language models was published in 2017, and neural networks are older than most people reading this. What changed was scale: far more data, far more computing power, and the discovery that models keep getting better as you add both. That is why AI felt sudden when it had been coming for forty years.
Try it yourself
Pick three AI features you have used — the recommendations on a video app, a translation tool, a chatbot. Classify each along all four dimensions. Where you are unsure, that uncertainty usually means the company has not said, which is itself worth noticing.