AI Course — Lesson 2: how AI reasons
It doesn't think: it processes. And understanding how it processes means knowing what to expect - and what not to.
AI Does Not “Think”, It Processes
It has no awareness, nor a model of the world: it processes data according to mathematical patterns it has learned.
The Scheme
Input → Model → Output. Everything in modern AI comes down to these three boxes.
Example: the Spam Filter
Input: the text of the email. Model: an algorithm trained on millions of labeled messages. Output: a probability that it is spam. The model does not know what a scam is; it only knows that certain combinations of words often appear in emails marked as spam.
The Two Lives of a Model
In training it sees millions of examples with the correct answer: it tries, gets it wrong, corrects itself. Slow and expensive. In inference it receives new input and answers: fast and immediate.
Error as a Compass
Every time it makes a mistake, the model calculates how wrong it was and uses that number to correct itself. Improvement comes from error, not from success.
Why AI Makes Mistakes
It has no common sense. It does not understand context and learns shortcuts: if in the data cats are almost always indoors and dogs outdoors, it will learn indoors = cat.
The Technical Curiosity: “But It Seems to Understand Everything!”
It seems so. In reality it recognizes language patterns. It has no model of the world: it has a model of language. It is intelligent in a functional sense, not a human one – very good at specific tasks.
Lesson Materials
Watch the full lesson on YouTube
Download the Handouts
