Data and simple patterns
AI often starts with data: numbers, lists, categories. A simple 'pattern' might be: the average of a list, or the most common item. Here we work with a list of numbers and find the average—like a tiny step toward what ML does with lots of data.
Appy Says…
AI doesn't think. It finds patterns. Show it a million pictures of cats and it learns what 'cat' looks like. Show it a billion sentences and it learns how language works. That's the whole secret — data and patterns.
What is AI Data and Pattern Learning?
Machine learning is the process of training a model on data so it can make predictions or decisions on new data. The model finds patterns — mathematical relationships — that generalise from the training set.
- •Training data: examples the model learns from
- •Features: the input values (pixels, words, numbers)
- •Labels: the correct output (cat/dog, spam/not-spam, price)
- •Pattern: a generalised rule learned from many examples
- •Prediction: applying the learned pattern to new, unseen data
- •More data + better features = better patterns
Think of it like learning from Minecraft maps
Imagine showing a player 10,000 Minecraft maps and asking 'is this map good or bad?' After enough examples, they spot patterns: good maps have variety, interesting landmarks, resource balance. An AI does the same — but with maths instead of instinct.
How It Works
- •1. Collect data: thousands/millions of labelled examples
- •2. Choose a model architecture (linear regression, neural network, etc.)
- •3. Train: the model tries predictions, compares to correct answers, adjusts its parameters
- •4. Evaluate: test on data the model hasn't seen to check generalisation
- •5. Deploy: use the trained model to make predictions on new inputs
- •6. The model learns by minimising a 'loss function' — a measure of how wrong it is
Real-World Examples
- •Spam filter: trained on millions of emails labelled spam/not-spam
- •Spotify Discover Weekly: trained on your listening patterns vs millions of users
- •TikTok For You: trained on every swipe, watch time, and share
- •Netflix recommendations: patterns in your watch history vs similar users
- •Google Photos face recognition: trained on billions of labelled face images
Key Facts
- •GPT-4 was trained on roughly 1 trillion tokens of text data
- •The ImageNet dataset (14 million labelled images) transformed computer vision AI in 2012
- •Garbage in, garbage out — biased training data produces biased models
- •Most AI breakthroughs in the last decade came from more data + more compute, not new algorithms
Watch Out!
AI models learn patterns from data — including biased patterns. If training data over-represents certain groups or contains historical discrimination, the model will reproduce those biases. Understanding training data is crucial for responsible AI.
Remember
AI = data + patterns. The model doesn't understand — it finds statistical regularities in data. The quality and diversity of training data determines how well it works.
What You Learned
- •AI learns patterns from labelled training data — not explicit rules programmed in
- •Train → evaluate → deploy; quality of data determines quality of predictions
- •Unlocks: understanding how every AI product from Spotify to TikTok to ChatGPT works
Key Facts
- →GPT-4 was trained on roughly 1 trillion tokens of text data
- →The ImageNet dataset (14 million labelled images) transformed computer vision AI in 2012
- →Garbage in, garbage out — biased training data produces biased models
- →Most AI breakthroughs in the last decade came from more data + more compute, not new algorithms
Real-World Examples
Remember
AI = data + patterns. The model doesn't understand — it finds statistical regularities in data. The quality and diversity of training data determines how well it works.
Quick Quiz
What do we need before we can find patterns?