Inputs, patterns, predictions

Here’s a simple way to think about machine learning. It has four parts.

Training data. People collect huge numbers of examples. For a recommender like Instagram, that means videos people watched, skipped, liked, or saved. This is the training data (show definition).

Training. An algorithm (show definition) goes through the examples and adjusts itself to find patterns. It might notice that people who finish cooking videos often watch more of them.

The model. After training, you have a model (show definition). Give it a new video and a viewer, the input. It gives back a prediction: how likely you are to watch, like, or save it.

Feedback. When you tap “not interested” or watch something all the way through, that becomes new data. This is a feedback loop (show definition).

An algorithm is step-by-step instructions, like a recipe. Machine learning is one way to build AI: it learns patterns from examples instead of following typed-in rules. AI is the bigger category: systems built to do tasks that usually need human thinking.