How a Neural Network Learns

{
  "version": 1,
  "title": "How a Neural Network Learns",
  "subtitle": "From a pile of numbers to a machine that recognises a cat: the idea behind modern AI, in plain words.",
  "topic": "How neural networks learn from examples",
  "domain": "ai",
  "emoji": "🧠",
  "kicker": "AI explained",
  "chips": [
    "No maths needed",
    "Beginner"
  ],
  "style": {
    "palette": "galaxy",
    "background": "ai",
    "fonts": "technical",
    "captions": "neon",
    "mascot": "robot",
    "effects": [
      "glow"
    ],
    "enter": "slide-blur",
    "transition": "blur-zoom",
    "header": [
      "host",
      "title"
    ],
    "progress": "dots"
  },
  "narration": {
    "voice": "M3",
    "speed": 0.96
  },
  "scenes": [
    {
      "kind": "intro",
      "say": [
        "Your phone can pick out a cat in a photo.",
        "Nobody wrote a rule that says what a cat looks like. So how does it know?"
      ]
    },
    {
      "kind": "motion",
      "heading": "Inside a neural network",
      "lead": "Let us open the box.",
      "canvas": "wide",
      "header": false,
      "shapes": [
        {
          "id": "links1",
          "type": "path",
          "d": "M 28 14 L 80 12 M 28 14 L 80 24 M 28 14 L 80 36 M 28 14 L 80 48 M 28 30 L 80 12 M 28 30 L 80 24 M 28 30 L 80 36 M 28 30 L 80 48 M 28 46 L 80 12 M 28 46 L 80 24 M 28 46 L 80 36 M 28 46 L 80 48",
          "stroke": "text",
          "strokeWidth": 0.25,
          "opacity": 0.4,
          "enter": "draw",
          "at": "line1",
          "delay": 1.1,
          "duration": 1.4
        },
        {
          "id": "links2",
          "type": "path",
          "d": "M 80 12 L 132 20 M 80 12 L 132 40 M 80 24 L 132 20 M 80 24 L 132 40 M 80 36 L 132 20 M 80 36 L 132 40 M 80 48 L 132 20 M 80 48 L 132 40",
          "stroke": "text",
          "strokeWidth": 0.25,
          "opacity": 0.4,
          "enter": "draw",
          "at": "line1",
          "delay": 1.5,
          "duration": 1.4
        },
        {
          "id": "in",
          "type": "circle",
          "x": 28,
          "y": 30,
          "size": 7,
          "fill": "sky",
          "fill2": "violet",
          "glow": true,
          "repeat": {
            "count": 3,
            "layout": "column",
            "gap": 16
          },
          "enter": "pop",
          "stagger": 0.15,
          "at": "line1"
        },
        {
          "id": "hid",
          "type": "circle",
          "x": 80,
          "y": 30,
          "size": 7,
          "fill": "violet",
          "fill2": "fuchsia",
          "glow": true,
          "repeat": {
            "count": 4,
            "layout": "column",
            "gap": 12
          },
          "enter": "pop",
          "stagger": 0.12,
          "at": "line1",
          "delay": 0.5
        },
        {
          "id": "out",
          "type": "circle",
          "x": 132,
          "y": 30,
          "size": 8,
          "fill": "emerald",
          "fill2": "teal",
          "glow": true,
          "repeat": {
            "count": 2,
            "layout": "column",
            "gap": 20
          },
          "enter": "pop",
          "stagger": 0.15,
          "at": "line1",
          "delay": 1
        },
        {
          "id": "tagIn",
          "type": "text",
          "text": "Photo goes in",
          "x": 28,
          "y": 57,
          "size": 3.6,
          "fill": "sky",
          "enter": "fade",
          "at": "line1",
          "delay": 1.6
        },
        {
          "id": "tagHid",
          "type": "text",
          "text": "Neurons mix the signals",
          "x": 80,
          "y": 57,
          "size": 3.6,
          "fill": "violet",
          "enter": "fade",
          "at": "line1",
          "delay": 1.8
        },
        {
          "id": "tagOut",
          "type": "text",
          "text": "Answer comes out",
          "x": 132,
          "y": 57,
          "size": 3.6,
          "fill": "emerald",
          "enter": "fade",
          "at": "line1",
          "delay": 2
        },
        {
          "id": "signal",
          "type": "circle",
          "x": 28,
          "y": 30,
          "size": 3.4,
          "fill": "amber",
          "glow": true,
          "enter": "pop",
          "at": "line2"
        },
        {
          "id": "verdict",
          "type": "text",
          "text": "Dog?",
          "x": 132,
          "y": 4,
          "size": 6,
          "fill": "rose",
          "enter": "pop",
          "at": "line3"
        }
      ],
      "script": "tl.to(\"signal\", { x: 104, duration: 2.6, ease: \"power1.inOut\", repeat: -1, repeatDelay: 0.5 }, \"line2\"); tl.to(\"links1\", { strokeWidth: 0.9, stroke: \"emerald\", opacity: 0.9, duration: 1.4, ease: \"power2.inOut\", yoyo: true, repeat: -1 }, \"line4\"); tl.to(\"verdict\", { text: \"Cat!\", fill: \"emerald\", duration: 1, ease: \"none\" }, \"line4+=2\");",
      "lines": [
        {
          "say": "A neural network is a stack of layers, made of tiny calculators called neurons."
        },
        {
          "say": "A photo goes in on the left. Each neuron mixes what it receives, and passes a signal to the next layer."
        },
        {
          "say": "At the far end comes an answer. The first time, it is mostly a guess, and it is usually wrong."
        },
        {
          "say": "So the network checks its mistake, and strengthens the connections that helped. Then it tries again, thousands of times."
        }
      ]
    },
    {
      "kind": "cycle",
      "heading": "The learning loop",
      "center": "Practice",
      "transition": "swirl",
      "items": [
        {
          "icon": "🎲",
          "label": "Guess",
          "tone": "sky",
          "say": "First, the network looks at a photo and makes a guess."
        },
        {
          "icon": "📏",
          "label": "Measure the mistake",
          "tone": "rose",
          "say": "Then it measures how far the guess was from the right answer."
        },
        {
          "icon": "🎚️",
          "label": "Adjust",
          "tone": "amber",
          "say": "It nudges every connection a tiny bit, in the direction that would have helped."
        },
        {
          "icon": "🔁",
          "label": "Repeat",
          "tone": "emerald",
          "say": "And it repeats this with the next photo, and the next, and the next."
        }
      ]
    },
    {
      "kind": "chart",
      "heading": "The mistakes shrink",
      "chart": "area",
      "items": [
        {
          "label": "Round 1",
          "value": 92,
          "display": "92%",
          "tone": "rose",
          "say": "In the first round, almost every answer is wrong."
        },
        {
          "label": "Round 10",
          "value": 58,
          "display": "58%",
          "tone": "amber",
          "say": "After ten rounds, it is already getting some right."
        },
        {
          "label": "Round 100",
          "value": 24,
          "display": "24%",
          "tone": "sky",
          "say": "After a hundred rounds, the mistakes are fewer than one in four."
        },
        {
          "label": "Round 1000",
          "value": 6,
          "display": "6%",
          "tone": "emerald",
          "say": "And after a thousand rounds, it is wrong only a few times in a hundred."
        }
      ],
      "outro": "These numbers are illustrative."
    },
    {
      "kind": "versus",
      "heading": "Two ways to build a program",
      "left": {
        "title": "Write the rules",
        "icon": "📜",
        "tone": "amber"
      },
      "right": {
        "title": "Learn from examples",
        "icon": "🖼️",
        "tone": "violet"
      },
      "lead": "Cats are the perfect test.",
      "transition": "flip",
      "points": [
        {
          "side": "left",
          "text": "Pointy ears? Whiskers? Fur?",
          "say": "With rules, you would list pointy ears, whiskers and fur, and miss every cat in a box."
        },
        {
          "side": "right",
          "text": "Thousands of labelled photos",
          "say": "With learning, you show it thousands of photos, each labelled cat or not cat."
        },
        {
          "side": "left",
          "text": "Breaks on every exception",
          "say": "The rules break on every exception."
        },
        {
          "side": "right",
          "text": "Finds the patterns by itself",
          "say": "The network finds the patterns by itself, including ones we would never think of."
        }
      ]
    },
    {
      "kind": "timetravel",
      "heading": "An old idea, finally ready",
      "from": 1950,
      "header": false,
      "lead": "None of this is new. What changed is the amount of data and computing power.",
      "events": [
        {
          "year": 1958,
          "title": "The perceptron",
          "text": "The first machine that learns from examples.",
          "tone": "amber",
          "say": "In nineteen fifty-eight, the perceptron became the first machine that learned from examples."
        },
        {
          "year": 1986,
          "title": "Backpropagation",
          "text": "A way to adjust every layer at once.",
          "tone": "violet",
          "say": "In nineteen eighty-six, backpropagation showed how to adjust every layer at once."
        },
        {
          "year": 2012,
          "title": "Deep learning takes off",
          "text": "Networks beat the field at recognising images.",
          "tone": "sky",
          "say": "In twenty twelve, deep networks suddenly beat every other method at recognising images."
        },
        {
          "year": 2022,
          "title": "AI you can talk to",
          "text": "Chatbots that write, translate and explain.",
          "tone": "emerald",
          "say": "And in twenty twenty-two, networks learned to hold a conversation with millions of people."
        }
      ]
    },
    {
      "kind": "quiz",
      "question": "What does a neural network adjust while it learns?",
      "options": [
        "The strength of its connections",
        "The brightness of the photos",
        "The number of cats in the world"
      ],
      "answer": 0,
      "explanation": "Learning means tuning millions of connection strengths.",
      "say": [
        "Quick quiz: what does a neural network adjust while it learns? The strength of its connections, the brightness of the photos, or the number of cats in the world?",
        "It adjusts the strength of its connections. Learning means tuning millions of them."
      ]
    },
    {
      "kind": "statement",
      "eyebrow": "Remember",
      "text": "It does not follow **rules**. It finds **patterns**.",
      "icon": "🧠",
      "header": false,
      "transition": "zoom-in",
      "say": "Remember: a neural network does not follow rules. It finds patterns."
    }
  ]
}