The field of machine learning revolves around programs and algorithms that have the ability to learn. Algorithms that can be trained on a data sets, learning their features, and can use their "experience" to make predictions on data they have never seen before.
These algorithms excel at solving problems that are difficult or cumbersome to model. A prime example is the problem of recognizing and classifying animals in pictures. While this is a task a human can perform rather easily, it is extremely difficult to create an algorithm that can find and identify animals in varying positions and postures, and with different individual appearances. Such an algorithm would first of all need a very detailed model of how different animals look from different angles, and in different poses. It must, of course, also take the lighting in the picture into account, and be able to recognize an animal even when parts of it are hidden.
Because of the immense complexity, constructing such algorithms by hand is not really an option; tasks like these require machines that can be taught to recognize the shapes and patterns in a picture as an animal. Google uses machine learning algorithms called neural networks in their image recognition software — see this article for some interesting and enjoyable reading on the subject. These neural networks have been trained on large sets of correctly classified images, to the point at which they can correctly classify images they have never been exposed to before.
Neural networks are programs with a structure that resembles the nervous tissue of the brain. Similar to a brain, a neural network consists of a number of artificial neurons that transmit information between each other, and whose output signals depend on the inputs from other neurons. Using specialized training algorithms, a neural network can made to recognize patterns in a data set, and based on what one could call experience, make predictions for data points it has never been seen before. Neural networks are very flexible, and have a multitude of uses - from function estimation and data classification to image and speech recognition.
A weather prediction neural network