Understanding what an artificial intelligence really does
This course explains, chapter after chapter, how an artificial intelligence model builds an answer, an image or a voice, why it gets things wrong with total confidence, and when not to trust it at all.
The course follows one single order, and that order matters. It starts with the most elementary step of all: how a text becomes, for a model, a string of units and then of numbers. Nothing that follows makes sense without that first step, neither the prediction of the next word, nor the confident invention, nor the biases a corpus passes on without saying so. The following chapters then look at what becomes of content handed to it, then at what a fabricated image or voice lets us see, and how far a watermark holds. The course only concludes at the end of that road: once the mechanism has been seen from the inside, deciding when a plausible answer is not enough, and holding back. That is why this chapter closes the course, and no other: discernment never comes before understanding, it follows it.
The chapters
01
Words in pieces
An algorithm that merges the most frequent pairs of characters until it has a vocabulary, applied live to a sentence and compared across several languages.
Open02
The next word
Watching, setting by setting, a model work out a probability distribution over the next word, then draw from it.
Open03
Why it makes things up
Putting an opening from the corpus to the same model, then one absent from it, and seeing that nothing in its answer says which is which, except the count it never shows.
Open04
Where the biases come from
A demonstration where you skew the make-up of a corpus and watch the model give the skew back to the exact figure, plus what the absence of a piece of data makes impossible.
Open05
What you give it
The four distinct things that happen to content handed to a service, what research has measured about it being given back, and the questions to ask before uploading a file.
Open06
A fabricated image
A demonstration that fabricates two images, reveals in their spectrum the regular trace of upsampling, then makes it disappear under noise.
Open07
A fabricated voice
A speech sound fabricated twice, roughly then carefully, to compare what the ear takes in with what its spectrum reveals, and how far a little ordinary noise wipes it out.
Open08
Watermarks
A green-list watermark placed on a text, a message hidden in the pixels of an image and then destroyed in front of you: two marking mechanisms, and how they differ from a declared provenance.
Open09
When not to use it
Three tests for deciding whether a model belongs in a situation, the cases where it belongs in none, and why an AI detector proves nothing.
OpenHow to follow it
The course reads from end to end, in the order the chapters are arranged above. Another way in exists for anyone who already knows what they are looking for: the paths page offers a different route depending on whether you are a teenager, a working adult or a teacher. A third way in leads straight to practice: the demonstrations page gathers every hands-on tool of the course, without going through the text that explains them.
No background needed
This course assumes no mathematics, no line of code and no account to create. Every technical term is defined at the point where it is needed, never before. Being able to read a web page is enough to follow it all the way through.
What this course does not do
This course explains how something works, not a product.
- It does not teach how to use any particular software or interface.
- It compares no artificial intelligence service with another.
- It does not say whether artificial intelligence is a good or a bad thing.
What it gives instead: enough to judge for yourself, once the mechanism has been seen from the inside.
The method
Every technical statement in this course carries its source, opened at the time of writing. What is still debated in research is reported as debated, never as settled. The detail, theme by theme, is gathered on the credits page.
The demonstrations
Every hands-on demonstration in the course runs entirely on this device, without a connection and without an account: you can open the browser’s Network tab while it works and see that no request goes out. Each one also explains its method in plain terms, so that it can be redone by hand, on paper, without it.