AI for Doctors / Theory / Lesson 1

AI, machine learning, deep learning, GenAI: which is which

Four words that get used as if they mean the same thing. They don’t, and the difference decides what a tool can and cannot do.

8 min

When a vendor says “AI-powered”, the phrase can mean anything from a fixed rule written by a programmer to a model trained on a billion images. The four terms nest inside one another, like the rings below.

Tap a ring
Artificial intelligenceMachine learningDeep learningGenAI

Machine learning

The computer learns the rule from examples instead of being told it. You give it many labelled cases; it finds the pattern that best separates them.

Example: A sepsis risk score learned from thousands of past ICU admissions.

Two practical questions follow from where a tool sits. First, did it learn from data? If yes, ask whose data, because it will be best on patients like the ones it learned from. Second, does it give you a label or score (predictive AI) or does it write new text or images (generative AI)? A score can be checked with sensitivity and specificity. Generated text has to be checked line by line, because it can be fluent and wrong at once.

Model: the trained thing itself: a big set of numbers (weights) that turns an input into an output.
Algorithm: the recipe used to train the model. People often say “algorithm” when they mean the model.
Foundation model: a very large model trained on broad data, then adapted to many tasks. MedGemma and GPT-class models are examples.

Watch

AI, ML, deep learning, GenAI: which is which
5 min · narrated

Check yourself

1. A hospital system flags any potassium above 6.5 mmol/L using a fixed rule. This is best called:
2. Which question matters most for a model that learned from data?
3. An LLM drafts a discharge summary. How should its output be checked?

Educational material, not clinical advice. Interactive numbers are teaching examples unless a source is named.