Your child just finished a quest. Here is, for each one, what they actually understood, and two readings, no more, so you can go see for yourself.
Two, because nobody opens fifteen references. Each time: a short one for the coffee break, a longer one for an evening when you have time.
Everything reads without creating an account. The one exception: the book from the last quest, flagged as such.
Your child showed a machine a few objects they had measured and labeled themselves. The machine learned to sort new ones by searching its memories for the one whose numbers look the most alike.
Elements of AI, part 4 · University of Helsinki and MinnaLearn, 2018
This is exactly what Turbot does, told for an adult: keep examples, measure which one looks most like the new object, copy its label. The page then moves on to music and video recommendations, which run on the same idea.
For the parent in a hurry.
Class'Code IAI, module 2 · Inria, on Pixees, 2020
A ten-minute video on how a machine learns from examples, and an activity where you prepare, yourself, the bundle it gets. That is where you see why a single mislabeled object is enough to send Turbot off the rails.
For the parent who wants to keep going with their child.
A machine that talks never knows in advance what it is going to say. It picks one word at a time, drawing from the ones most likely to come next. The more you let it draw rare words, the wilder its sentences get.
Frédéric Alexandre, Inria and University of Bordeaux · The Conversation, 2023
In ten minutes, the heart of what your child just did: the machine knows nothing, it guesses the most likely word, one word at a time. The author also explains why a sentence can sound perfectly right and be completely wrong. That is the warning Plume gives at the end.
For the parent in a hurry.
Flowers team, Inria centre at the University of Bordeaux, 2023
Researchers explain in five minutes why the machine guesses the next word instead of knowing it. Watch it with your child right after the game; the episode on its weaknesses shows why it gets things wrong with so much confidence.
For the parent who wants to keep going with their child.
A machine does not think. It looks at a single square at a time, applies a small rule written in advance, moves one square over, and starts again. That is exactly how a computer works.
Hamdi Ben Abdallah, animation by Hugo Lehmann · Interstices (Inria), 2013
Your child just moved Robi forward square by square; here is the same machine, with the full list of its rules displayed beside the tape. The animation's first program is "add 1 to a binary number": the exact calculation they just did on screen.
For the parent in a hurry, with the child sitting alongside.
Sacha Krakowiak · Interstices (Inria), 2012
If your child asks "who was Turing?", the answer is here. How an idea written on paper in 1936 became the first real machines, and why the game is allowed to say that your tablet does the same thing, only much faster.
For the parent who wants to dig deeper.
A machine that knows no strategy at all can get good just by being punished when it loses and rewarded when it wins, hundreds of times in a row.
Pixees · Inria and the Société informatique de France, 2020
Your child's machine, for real: cups, marbles, and the same rule: take out the marbles that lose, add the ones that win. The game is not tic-tac-toe but the stick game; the mechanism is identical, and you can rebuild it at the kitchen table.
For the family that wants to build the machine at home.
Alain Dutech, Bruno Scherrer and Christophe Thiéry · Interstices (Inria), 2008
What your child was doing by removing and adding marbles has a name: reinforcement learning. Three Inria researchers show what the same principle does when plugged into Tetris. The first half reads with no background at all; the rest gets into the formulas.
For the parent who wants to put a name on what they saw.
Not every machine is trained the same way. Depending on the problem, you show it thousands of examples, reward it when it succeeds, write out rules for it in black and white, or make it guess what comes next. The problem picks the method, not the machine.
Melissa Racine and Tricia Grant · HabiloMédias (MediaSmarts), 2024
The ways of training a machine your child just handled, tied to what they already use: Netflix suggestions, face recognition, the car's cruise control. A Canadian organization that writes directly for parents, so the piece also answers "what do I do with this at home?"
For the parent in a hurry.
Jean-Paul Haton · Interstices (Inria), 2024
Your child just tried several ways of training a machine; here is where they come from, and why the field moved from hand-written rules to learning by example, then by reward. Longer than a blog post, but it reads in one sitting.
For the parent who wants to dig in afterward.
An artificial intelligence is not one big machine: it is several small programs that each know how to do a single thing. The real work is deciding which one to switch on, and when.
Maxime Morge, LIRIS (CNRS) · The Conversation, 2026
Your child just did by hand what this article describes: wiring together pieces that each know a single thing, and coordinating the whole. Written by a computer science professor for the general public, it puts the "AI agents" craze back into a story that is fifty years old.
For the parent in a hurry.
Philippe Mathieu, Sébastien Picault and Jean-Christophe Routier, CRIStAL · Interstices (Inria), 2016
Your child made four little programs move, each one only able to do a single thing. Here is the same machinery at full scale: the armies of The Hobbit, the citizens of SimCity, road traffic. And it reads without knowing anything about programming.
For the parent with half an hour to spare.
On the night of the great fire, your child checked four machines before lighting the fire: they tested an explanation instead of believing it, looked at whether a box was open or closed, caught a gatekeeper that leaned toward its examples, and kept the final decision for the human. The night’s lesson: you don’t take a machine at its word — you check it.
Interview with Aurélie Jean, researcher in computational sciences · Radio-Canada, 2019
The gatekeeper trial your child just passed — a machine turning someone away because they were missing from its examples — is exactly what this researcher explains here, with real cases. Ten minutes, and you can answer when they ask if it happens for real.
For the parent in a hurry.
Cathy O'Neil, translated by Sébastien Marty · Les Arènes, 2018, 352 pages
Your child fixed a gatekeeper that copied its examples; this book tells the same story when that mechanism truly decides a hiring, a loan or a school place. It reads without any computer background, and it can be borrowed before it’s bought.
For the curious parent. The only link on this page that is not free.