{"course":{"slug":"ai-for-teachers","title":"AI for Teachers","url":"https://addaly.com/learn/ai-for-teachers","level":"beginner","track":"work"},"lesson":{"slug":"the-sorting-game-for-younger-children","title":"The sorting game for younger children","url":"https://addaly.com/learn/ai-for-teachers/the-sorting-game-for-younger-children","position":44,"of":57,"minutes":8,"answer":"Classification, not generation, is the right first model for a child (a machine shown many labelled examples learns to sort new ones.","keyIdea":"Classification, not generation, is the right first model for a child (a machine shown many labelled examples learns to sort new ones, and gets it wrong at the boundary) and it can be taught with a pile of pictures and two boxes, with Teachable Machine as the free follow-up if a laptop appears."},"body":{"format":"markdown","text":"## Why start with sorting\n\nThe chatbot is the AI children see, and it is the hardest one to explain. Underneath every AI system that matters to a child's life (the one that tags faces in photos, the one that recommends videos, the one that decides which post they see) is a simpler idea: a machine shown many labelled examples learns to sort new ones. That idea is teachable at 8. It is also true, and it scales up: the chatbot is a sorter too, choosing the most likely next word from a very large pile.\n\nSo the first unit for younger children is sorting, and it needs no electricity.\n\n## The game\n\nMaterials: forty small pictures cut from magazines or drawn: twenty cats, twenty dogs, clearly labelled on the back. Two boxes. A screen (a large book standing up) and a volunteer to be the machine.\n\n**Training.** The class shows the machine one picture at a time, saying the label. \"Cat.\" \"Dog.\" The machine looks and puts it in the box. Twenty pictures. The machine is allowed to notice anything (ears, size, tail, whiskers) but not to ask questions. This is important: the machine learns only from what it is shown.\n\n**Testing.** New pictures, no label on the front. The machine sorts. The class checks the back. Most are right. Then the fox.\n\n**The fox.** The machine puts it in the dog box, and the class laughs, and that is the lesson. Ask: why did it do that? Someone will say: because it never saw a fox. Ask: what would fix it? Someone will say: show it foxes. Ask: and then a wolf? A raccoon? Now they have the idea that a machine knows only the kinds of thing it was shown, and that the world has more kinds than any training pile.\n\nTwenty-five minutes, and they have machine learning, the training set, the test set, and the out-of-distribution error: without any of those words.\n\n## The second round: what it was learning\n\nRun it again with a trick. All the training cats are on a white background; all the dogs on grass. Then test with a cat on grass. The machine, if the volunteer is honest about what they noticed, puts it with the dogs.\n\n> Figure: Forty pictures, two boxes, and the fox. Steps: 1. Show twenty labelled pictures (One at a time, saying \"cat\" or \"dog\". The machine may look but not ask.); 2. Test with new pictures (No label on the front. The class checks the back. Most are right.); 3. Show the fox (It goes in the dog box, and the class laughs.); 4. Ask why (\"Because it never saw a fox.\" Then: and a wolf? A raccoon?); 5. Run it again with a trick (All training cats on white, all dogs on grass. Then a cat on grass.); 6. It learned the background (It sorted perfectly and learned the wrong thing, and nobody could tell until a picture broke the pattern.). Twenty-five minutes gives a nine-year-old the training set, the test set and the out-of-distribution error, without any of those words. The second round is the one adults miss: with every training cat on white and every dog on grass, the machine sorts perfectly and has learned the background.\n\nThis is the most important lesson in the unit and it is the one adults miss. The machine did not learn \"cat\". It learned \"white background\". It sorted perfectly and it had learned the wrong thing, and nobody could tell until a picture broke the pattern. Real systems do exactly this (a famous case learned to spot huskies by the snow behind them) and a 9-year-old who has seen it happen with a book and forty pictures will remember it when they are 19 and reading about a hiring algorithm.\n\n## Words to give them, and words to withhold\n\nGive them: examples, pattern, mistake, \"it only knows what it was shown\". Withhold: algorithm, neural network, probability. There is time for those. The concept is what matters, and the words come easily later once the concept is there.\n\n## The free follow-up if a laptop appears\n\nGoogle's Teachable Machine is free, runs in a browser, and does the sorting game for real: children show a webcam twenty pictures of one thing and twenty of another, press train, and then test it with new objects. It sorts. It fails on the fox. It fails on the background. Everything they learned with the boxes happens on screen in ten minutes, and it is one of the few AI tools built for a classroom.\n\nIt needs a laptop with a camera and a connection. If the school has one such laptop, one demonstration at the front, after the paper version, is the right order: the paper version is the lesson and the screen is the confirmation. Many schools have neither, and the paper version is complete on its own.\n\n## Connecting it to their lives\n\nEnd with a question: where have you seen a machine sort things? Photos on a phone that find faces. Videos suggested after the one you watched. The spam folder. Each is a sorter trained on examples, each makes mistakes at the edges, and each was trained by somebody on a pile that somebody chose. \"Who chose the pile?\" is the question that turns a mechanism lesson into a citizenship one, and 10-year-olds ask it unprompted once the game has been played.\n\n## The honest limit\n\nSorting is the right first model and it is not the whole of AI. A child who has only the sorting model will, at 12, meet a chatbot and be puzzled by how a sorter writes a poem. The bridge is the prediction activity, and the progression lesson placed it at 11 to 13. Sorting first, prediction second. Both are true, and the second builds on the first."},"check":{"question":"In the sorting game, a class trains a \"machine\" (a student behind a screen) with twenty pictures of cats and dogs, then tests it. It sorts a fox as a dog. What is the teaching point, and what would fix the machine?","options":[{"text":"The machine never saw a fox, so it used the nearest pattern; labelled foxes would teach a third pile.","correct":true},{"text":"The machine was not paying enough attention during training, and a more careful student in the role would have sorted it correctly.","correct":false,"tempting":"Misses the mechanism; the student sorted by the features the training pictures taught, and no fox was among them."},{"text":"Foxes are simply too similar to dogs for any machine, real or pretend, to tell apart, so the error was unavoidable.","correct":false,"tempting":"Real classifiers separate foxes from dogs given examples; the point is what the machine was trained on, not what is possible."},{"text":"The training set should have been much larger: 200 pictures of cats and dogs instead of 20 would have fixed the error.","correct":false,"tempting":"More cats and dogs teaches more about cats and dogs; it teaches nothing about foxes, which is the point."}]},"prev":{"slug":"what-to-teach-at-which-age","title":"What to teach at which age","url":"https://addaly.com/learn/ai-for-teachers/what-to-teach-at-which-age"},"next":{"slug":"image-generators-and-deepfakes","title":"Image generators and deepfakes","url":"https://addaly.com/learn/ai-for-teachers/image-generators-and-deepfakes"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'The sorting game for younger children', in AI for Teachers. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-teachers/the-sorting-game-for-younger-children (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"The sorting game for younger children","courseTitle":"AI for Teachers","module":{"index":6,"slug":"teaching-ai-itself","title":"Teaching AI itself"},"answer":null,"keyIdea":"Classification, not generation, is the right first model for a child (a machine shown many labelled examples learns to sort new ones, and gets it wrong at the boundary) and it can be taught with a pile of pictures and two boxes, with Teachable Machine as the free follow-up if a laptop appears.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"type":"flow","steps":[{"label":"Show twenty labelled pictures","detail":"One at a time, saying “cat” or “dog”. The machine may look but not ask."},{"label":"Test with new pictures","detail":"No label on the front. The class checks the back. Most are right."},{"label":"Show the fox","detail":"It goes in the dog box, and the class laughs."},{"label":"Ask why","detail":"“Because it never saw a fox.” Then: and a wolf? A raccoon?"},{"label":"Run it again with a trick","detail":"All training cats on white, all dogs on grass. Then a cat on grass."},{"label":"It learned the background","detail":"It sorted perfectly and learned the wrong thing, and nobody could tell until a picture broke the pattern."}],"title":"Forty pictures, two boxes, and the fox","caption":"Twenty-five minutes gives a nine-year-old the training set, the test set and the out-of-distribution error, without any of those words. The second round is the one adults miss: with every training cat on white and every dog on grass, the machine sorts perfectly and has learned the background."},"alt":"Figure: Forty pictures, two boxes, and the fox. Steps: 1. Show twenty labelled pictures (One at a time, saying “cat” or “dog”. The machine may look but not ask.); 2. Test with new pictures (No label on the front. The class checks the back. Most are right.); 3. Show the fox (It goes in the dog box, and the class laughs.); 4. Ask why (“Because it never saw a fox.” Then: and a wolf? A raccoon?); 5. Run it again with a trick (All training cats on white, all dogs on grass. Then a cat on grass.); 6. It learned the background (It sorted perfectly and learned the wrong thing, and nobody could tell until a picture broke the pattern.). Twenty-five minutes gives a nine-year-old the training set, the test set and the out-of-distribution error, without any of those words. The second round is the one adults miss: with every training cat on white and every dog on grass, the machine sorts perfectly and has learned the background."}},"blocks":[{"t":"h2","id":"why-start-with-sorting","runs":[{"text":"Why start with sorting"}]},{"t":"p","runs":[{"text":"The chatbot is the AI children see, and it is the hardest one to explain. Underneath every AI system that matters to a child’s life (the one that tags faces in photos, the one that recommends videos, the one that decides which post they see) is a simpler idea: a machine shown many labelled examples learns to sort new ones. That idea is teachable at 8. It is also true, and it scales up: the chatbot is a sorter too, choosing the most likely next word from a very large pile."}]},{"t":"p","runs":[{"text":"So the first unit for younger children is sorting, and it needs no electricity."}]},{"t":"h2","id":"the-game","runs":[{"text":"The game"}]},{"t":"p","runs":[{"text":"Materials: forty small pictures cut from magazines or drawn: twenty cats, twenty dogs, clearly labelled on the back. Two boxes. A screen (a large book standing up) and a volunteer to be the machine."}]},{"t":"p","runs":[{"text":"Training.","b":true},{"text":" The class shows the machine one picture at a time, saying the label. “Cat.” “Dog.” The machine looks and puts it in the box. Twenty pictures. The machine is allowed to notice anything (ears, size, tail, whiskers) but not to ask questions. This is important: the machine learns only from what it is shown."}]},{"t":"p","runs":[{"text":"Testing.","b":true},{"text":" New pictures, no label on the front. The machine sorts. The class checks the back. Most are right. Then the fox."}]},{"t":"p","runs":[{"text":"The fox.","b":true},{"text":" The machine puts it in the dog box, and the class laughs, and that is the lesson. Ask: why did it do that? Someone will say: because it never saw a fox. Ask: what would fix it? Someone will say: show it foxes. Ask: and then a wolf? A raccoon? Now they have the idea that a machine knows only the kinds of thing it was shown, and that the world has more kinds than any training pile."}]},{"t":"p","runs":[{"text":"Twenty-five minutes, and they have machine learning, the training set, the test set, and the out-of-distribution error: without any of those words."}]},{"t":"h2","id":"the-second-round-what-it-was-learning","runs":[{"text":"The second round: what it was learning"}]},{"t":"p","runs":[{"text":"Run it again with a trick. All the training cats are on a white background; all the dogs on grass. Then test with a cat on grass. The machine, if the volunteer is honest about what they noticed, puts it with the dogs."}]},{"t":"p","runs":[{"text":"This is the most important lesson in the unit and it is the one adults miss. The machine did not learn “cat”. It learned “white background”. It sorted perfectly and it had learned the wrong thing, and nobody could tell until a picture broke the pattern. Real systems do exactly this (a famous case learned to spot huskies by the snow behind them) and a 9-year-old who has seen it happen with a book and forty pictures will remember it when they are 19 and reading about a hiring algorithm."}]},{"t":"h2","id":"words-to-give-them-and-words-to-withhold","runs":[{"text":"Words to give them, and words to withhold"}]},{"t":"p","runs":[{"text":"Give them: examples, pattern, mistake, “it only knows what it was shown”. Withhold: algorithm, neural network, probability. There is time for those. The concept is what matters, and the words come easily later once the concept is there."}]},{"t":"h2","id":"the-free-follow-up-if-a-laptop-appears","runs":[{"text":"The free follow-up if a laptop appears"}]},{"t":"p","runs":[{"text":"Google’s Teachable Machine is free, runs in a browser, and does the sorting game for real: children show a webcam twenty pictures of one thing and twenty of another, press train, and then test it with new objects. It sorts. It fails on the fox. It fails on the background. Everything they learned with the boxes happens on screen in ten minutes, and it is one of the few AI tools built for a classroom."}]},{"t":"p","runs":[{"text":"It needs a laptop with a camera and a connection. If the school has one such laptop, one demonstration at the front, after the paper version, is the right order: the paper version is the lesson and the screen is the confirmation. Many schools have neither, and the paper version is complete on its own."}]},{"t":"h2","id":"connecting-it-to-their-lives","runs":[{"text":"Connecting it to their lives"}]},{"t":"p","runs":[{"text":"End with a question: where have you seen a machine sort things? Photos on a phone that find faces. Videos suggested after the one you watched. The spam folder. Each is a sorter trained on examples, each makes mistakes at the edges, and each was trained by somebody on a pile that somebody chose. “Who chose the pile?” is the question that turns a mechanism lesson into a citizenship one, and 10-year-olds ask it unprompted once the game has been played."}]},{"t":"h2","id":"the-honest-limit","runs":[{"text":"The honest limit"}],"role":"limit"},{"t":"p","runs":[{"text":"Sorting is the right first model and it is not the whole of AI. A child who has only the sorting model will, at 12, meet a chatbot and be puzzled by how a sorter writes a poem. The bridge is the prediction activity, and the progression lesson placed it at 11 to 13. Sorting first, prediction second. Both are true, and the second builds on the first."}]}],"check":{"question":[{"text":"In the sorting game, a class trains a “machine” (a student behind a screen) with twenty pictures of cats and dogs, then tests it. It sorts a fox as a dog. What is the teaching point, and what would fix the machine?"}],"options":[{"text":[{"text":"The machine never saw a fox, so it used the nearest pattern; labelled foxes would teach a third pile."}],"correct":true,"tempting":null},{"text":[{"text":"The machine was not paying enough attention during training, and a more careful student in the role would have sorted it correctly."}],"correct":false,"tempting":[{"text":"Misses the mechanism; the student sorted by the features the training pictures taught, and no fox was among them."}]},{"text":[{"text":"Foxes are simply too similar to dogs for any machine, real or pretend, to tell apart, so the error was unavoidable."}],"correct":false,"tempting":[{"text":"Real classifiers separate foxes from dogs given examples; the point is what the machine was trained on, not what is possible."}]},{"text":[{"text":"The training set should have been much larger: 200 pictures of cats and dogs instead of 20 would have fixed the error."}],"correct":false,"tempting":[{"text":"More cats and dogs teaches more about cats and dogs; it teaches nothing about foxes, which is the point."}]}],"order":[2,3,0,1],"why":null},"prev":{"slug":"what-to-teach-at-which-age","title":"What to teach at which age","minutes":9},"next":{"slug":"image-generators-and-deepfakes","title":"Image generators and deepfakes","minutes":10}}}