[{"data":1,"prerenderedAt":328},["ShallowReactive",2],{"template:how-ai-works:en":3},{"title":4,"description":5,"readTime":6,"slug":7,"language":8,"kind":9,"coverUrl":10,"document":11},"How Artificial Intelligence Works","Data, training and neural networks in plain language — and where AI has limits","11 min","how-ai-works","en","explainer","\u002Ftemplate-media\u002Fhow-ai-works\u002Fcover.jpg",{"v":12,"kind":13,"design":14,"blocks":25},2,"longread",{"fontPairing":15,"baseTheme":16,"lineHeight":17,"blockGap":18,"showProgressBar":19,"accentColor":20,"viewPassThreshold":21,"quizPassThreshold":22,"quizShowCorrectness":19,"quizAllowRetry":19,"outlineSource":23,"tocPlacement":24,"language":8},"tech","dark","relaxed","normal",true,"#8b5cf6",80,70,"chapters","document",[26,35,42,48,53,70,77,81,84,113,140,151,155,162,165,180,204,220,226,233,238,242,248,263,286,291,300,313,325],{"id":27,"type":28,"title":4,"subtitle":5,"author":29,"date":30,"readTime":6,"imageUrl":10,"style":31,"overlayOpacity":32,"titleColor":33,"subtitleColor":34},"ai-cover","cover","Learning team","","dramatic-dark",40,"#ffffff","#e2e8f0",{"id":36,"type":37,"text":38,"dropCap":19,"alignment":39,"columns":40,"fontSize":41},"ai-lead","paragraph","\u003Cp>Artificial intelligence writes texts, recognizes faces and recommends what to watch next. It can feel like magic, but under the hood there is a surprisingly simple idea: learning patterns from examples. This explainer shows how that works — without formulas.\u003C\u002Fp>","left",1,"lg",{"id":43,"type":44,"title":45,"description":46,"chapterNumber":30,"level":40,"style":47},"ai-basics","chapter","What AI is — and what it is not","From hand-written rules to learning from data","classic",{"id":49,"type":37,"text":50,"dropCap":51,"alignment":39,"columns":40,"fontSize":52},"ai-basics-text","\u003Cp>For decades, programs followed rules written by people: if the email contains “win a prize”, mark it as spam. Modern AI works differently. Instead of rules, it gets thousands of examples — spam and normal emails — and finds the patterns itself. It does not understand the world the way people do; it calculates what is most likely given what it has seen.\u003C\u002Fp>",false,"base",{"id":54,"type":55,"columnsCount":56,"items":57},"ai-kinds","grid",3,[58,62,66],{"title":59,"text":60,"icon":61},"Rules","People write the logic. Predictable, but breaks on cases nobody thought of.","BookOpen",{"title":63,"text":64,"icon":65},"Machine learning","The program learns patterns from labeled examples and applies them to new data.","Cpu",{"title":67,"text":68,"icon":69},"Generative AI","Models trained on huge amounts of text or images create new content from a prompt.","Sparkles",{"id":71,"type":72,"text":73,"author":74,"role":75,"style":76},"ai-quote","quote","We can only see a short distance ahead, but we can see plenty there that needs to be done.","Alan Turing","Computing Machinery and Intelligence, 1950","italic-serif",{"id":78,"type":44,"title":79,"description":80,"chapterNumber":30,"level":40,"style":47},"ai-learning","How a model learns","Data in, patterns out",{"id":82,"type":37,"text":83,"dropCap":51,"alignment":39,"columns":40,"fontSize":52},"ai-learning-text","\u003Cp>Every AI system goes through the same stages. Click the markers to see what happens at each one.\u003C\u002Fp>",{"id":85,"type":86,"url":87,"caption":88,"aspectRatio":89,"markerStyle":90,"trigger":91,"markers":92},"ai-stages","hotspots","\u002Ftemplate-media\u002Fhow-ai-works\u002Fmap.png","Click a marker to see each stage","auto","number","click",[93,99,104,109],{"id":94,"x":95,"y":96,"title":97,"text":98},"ai-stages-1",12.5,12,"Data","Examples the model learns from: photos with labels, texts, purchase histories. Quality matters more than quantity.",{"id":100,"x":101,"y":96,"title":102,"text":103},"ai-stages-2",53.8,"Training","The model makes guesses, compares them with correct answers and adjusts itself — millions of times.",{"id":105,"x":95,"y":106,"title":107,"text":108},"ai-stages-3",57,"Prediction","A trained model gets new input and returns the most likely answer: a label, a number or a piece of text.",{"id":110,"x":101,"y":106,"title":111,"text":112},"ai-stages-4","Limits","A model only knows patterns from its data. Outside of them, it can be confidently wrong.",{"id":114,"type":115,"items":116,"orientation":137,"numbering":138,"style":139},"ai-steps","timeline",[117,121,125,129,133],{"value":118,"title":119,"text":120},"1","Collect data","Gather examples and label them: “cat”, “dog”, “spam”, “not spam”.",{"value":122,"title":123,"text":124},"2","Train","Show the examples to the model over and over until its guesses stop improving.",{"value":126,"title":127,"text":128},"3","Test","Check the model on examples it has never seen to measure real accuracy.",{"value":130,"title":131,"text":132},"4","Use","Put the model to work on new data in a product.",{"value":134,"title":135,"text":136},"5","Improve","Collect mistakes, add new data and retrain.","horizontal","manual","line",{"id":141,"type":142,"question":143,"options":144,"selectionType":145,"pairs":146,"correctIndex":147,"correctIndexes":148,"explanation":149,"feedbackIncorrect":30,"variant":150},"ai-quiz-steps","quiz","Put the stages of building an AI model in order.",[119,123,127,131,135],"sequence",[],null,[],"Testing on unseen data comes before real use — otherwise you do not know how good the model actually is.","interactive",{"id":152,"type":44,"title":153,"description":154,"chapterNumber":30,"level":40,"style":47},"ai-network","Inside a neural network","Layers of simple calculations",{"id":156,"type":157,"url":158,"alt":159,"caption":160,"credits":30,"layout":161,"aspectRatio":89},"ai-figure","image","\u002Ftemplate-media\u002Fhow-ai-works\u002Ffigure.png","Three columns of connected circles with one highlighted path from left to right","A neural network: inputs on the left, hidden layers in the middle, answers on the right. The highlighted path is one signal.","contained",{"id":163,"type":37,"text":164,"dropCap":51,"alignment":39,"columns":40,"fontSize":52},"ai-network-text","\u003Cp>A neural network is a chain of layers. Each “neuron” takes numbers from the previous layer, weighs them and passes the result on. Training adjusts the weights so that the final answer matches the examples more and more often.\u003C\u002Fp>",{"id":166,"type":167,"items":168,"style":178,"defaultIndex":179},"ai-layers","tabs",[169,172,175],{"title":170,"text":171,"imageUrl":30,"videoUrl":30},"Input layer","\u003Cp>Receives raw data as numbers: pixels of a photo or words of a sentence turned into numbers.\u003C\u002Fp>",{"title":173,"text":174,"imageUrl":30,"videoUrl":30},"Hidden layers","\u003Cp>Find patterns step by step: edges, then shapes, then objects. Nobody programs these patterns — they emerge during training.\u003C\u002Fp>",{"title":176,"text":177,"imageUrl":30,"videoUrl":30},"Output layer","\u003Cp>Gives the answer: the probability of “cat” or “dog”, or the next most likely word.\u003C\u002Fp>","underline",0,{"id":181,"type":182,"headers":183,"rows":187,"caption":203,"hasHeaderRow":19,"hasHeaderColumn":19,"striped":19,"bordered":19},"ai-tasks","table",[184,185,186],"Task","What it does","Example",[188,192,195,199],[189,190,191],"Classification","Puts input into categories","Spam filter",[107,193,194],"Estimates a number","Delivery time forecast",[196,197,198],"Recommendation","Suggests what you may like","Next video to watch",[200,201,202],"Generation","Creates new content","Drafting an email","What AI does in everyday products",{"id":205,"type":142,"question":206,"options":207,"selectionType":208,"pairs":209,"correctIndex":147,"correctIndexes":218,"explanation":219,"feedbackIncorrect":30,"variant":150},"ai-quiz-tasks","Match each product feature with the type of AI task.",[],"matching",[210,212,214,216],{"left":211,"right":189},"Sorting support tickets by topic",{"left":213,"right":107},"Estimating next month’s sales",{"left":215,"right":196},"Suggesting courses to an employee",{"left":217,"right":200},"Writing a first draft of a summary",[],"The same idea — learning from examples — powers all four types of tasks.",{"id":221,"type":222,"title":223,"text":224,"variant":225},"ai-warning","callout","AI can be confidently wrong","Generative models produce the most likely text, not checked facts. They can invent sources, numbers and names. Always verify important information.","warning",{"id":227,"type":228,"label":229,"scope":230,"style":231,"align":232},"ai-continue","continue","Continue to using AI responsibly","next-step","primary","center",{"id":234,"type":235,"style":236,"spacing":237},"ai-separator","separator","dots","md",{"id":239,"type":44,"title":240,"description":241,"chapterNumber":30,"level":40,"style":47},"ai-responsible","Using AI responsibly","What to keep in mind at work",{"id":243,"type":244,"url":245,"caption":246,"credits":30,"layout":161,"aspectRatio":247,"autoplay":51,"loop":19,"muted":19},"ai-video","video","\u002Ftemplate-media\u002Fhow-ai-works\u002Fclip.mp4","A signal passes through the layers of a network","16:9",{"id":249,"type":250,"items":251,"multiple":51,"icon":261,"style":262,"firstOpen":19},"ai-faq","accordion",[252,255,258],{"title":253,"text":254},"Does AI understand what it says?","\u003Cp>Not in the human sense. It finds statistical patterns. That is why it can write fluently and still make obvious mistakes.\u003C\u002Fp>",{"title":256,"text":257},"Where does the data come from?","\u003Cp>From public texts, licensed datasets and data users provide. Never paste confidential company data into tools that are not approved by your company.\u003C\u002Fp>",{"title":259,"text":260},"Can AI be biased?","\u003Cp>Yes. If the training data is unbalanced, the model repeats that imbalance. Check results for fairness, especially in decisions about people.\u003C\u002Fp>","chevron","card",{"id":264,"type":265,"columnsCount":56,"flip":266,"items":267,"frontBg":30,"frontColor":30,"backBg":30,"backColor":30},"ai-terms","flashcards","hover",[268,271,274,277,280,283],{"frontTitle":269,"frontText":30,"frontImageUrl":30,"backTitle":30,"backText":270,"backImageUrl":30},"Model","A program that has learned patterns from data.",{"frontTitle":272,"frontText":30,"frontImageUrl":30,"backTitle":30,"backText":273,"backImageUrl":30},"Training data","The examples a model learns from.",{"frontTitle":275,"frontText":30,"frontImageUrl":30,"backTitle":30,"backText":276,"backImageUrl":30},"Parameter","A number inside the model adjusted during training.",{"frontTitle":278,"frontText":30,"frontImageUrl":30,"backTitle":30,"backText":279,"backImageUrl":30},"Prompt","The instruction or question you give a generative model.",{"frontTitle":281,"frontText":30,"frontImageUrl":30,"backTitle":30,"backText":282,"backImageUrl":30},"Hallucination","A confident but invented answer from a generative model.",{"frontTitle":284,"frontText":30,"frontImageUrl":30,"backTitle":30,"backText":285,"backImageUrl":30},"Bias","A systematic error that comes from unbalanced data.",{"id":287,"type":288,"text":289,"level":290,"alignment":39,"withAccent":19},"ai-summary","heading","In a nutshell","h3",{"id":292,"type":293,"style":294,"items":295},"ai-summary-list","list","bullet",[296,297,298,299],"AI learns patterns from examples instead of following written rules","Training adjusts the model until its answers match the data","Neural networks are layers of simple calculations","AI can be wrong or biased — check what matters",{"id":301,"type":142,"question":302,"options":303,"selectionType":308,"pairs":309,"correctIndex":147,"correctIndexes":310,"explanation":311,"feedbackIncorrect":312,"variant":150},"ai-quiz-true","Which statements about AI are true?",[304,305,306,307],"It learns from examples","It can produce confident but wrong answers","It repeats biases present in its data","It understands the world exactly like a person","multiple",[],[179,40,12],"AI is powerful pattern matching — useful, but not understanding in the human sense.","AI does not understand the world like a person; it calculates likely answers.",{"id":314,"type":142,"question":315,"options":316,"selectionType":320,"pairs":321,"correctIndex":179,"correctIndexes":322,"explanation":323,"feedbackIncorrect":324,"variant":150},"ai-quiz-overfitting","A model is 99% accurate on its training data but often wrong on new data. What most likely happened?",[317,318,319],"It memorized the training examples instead of learning general patterns","It needs a faster computer","The new data is too small","single",[],[],"This is called overfitting: the model learned the examples by heart. Testing on unseen data reveals it.","Speed and data size are not the issue — the model did not generalize.",{"id":326,"type":37,"text":327,"dropCap":51,"alignment":39,"columns":40,"fontSize":52},"ai-closing","\u003Cp>Next time you use an AI tool, think about the data behind it and check the answers that matter. Understanding how AI works is the best protection against both hype and fear.\u003C\u002Fp>",1792666346383]