So far, we’ve used the term AI in the broadest sense, but to understand AI more fully we need to understand machine learning.

Machine learning can be used to find patterns which are typically too complex for people to detect.

A good metaphor for machine learning is a very smart combine harvester.

Figure 21. Machine learning capabilities

As it cuts through the crop it applies machine learning capabilities. 

The first thing it does is classifies the crop in terms of whether it is wheat or chaff. The wheat is separated into a tank and the chaff is blown out of the back of the combine harvester.

Next, it looks for anomalies – are there other species of plants in the crop?

Then it analyses the size of the grains, and clusters them into groups – something that could be useful when combined with geographic and soil analysis data.

Finally, it can forecast ahead. It can use regression analysis to predict the grain sizes and the overall yield.

Driving the Combine Harvester is a diesel engine which in turn drives an electrical generator, which powers its sensors, computing and communication capabilities.

It’s really important to remember that the point of machine learning is to make predictions.

In the case of a combine harvester the outcomes could be a fully automated machine, real-time information to the markets, and information to form the basis of future crop planning.

Most machine learning tasks can be achieved using one of the following mathematical methods:

Figure 22. Main types of machine learning models

Machine learning is the application of one or more of these types of models to data, and the application of machine learning to a problem can be thought of as AI.  

There are two key types of machine learning – ‘supervised’ and ‘unsupervised’.

Supervised Learning

Supervised Learning requires data that has been labelled.

Classification and Regression methods can be used on labelled data.

An example of supervised learning is to predict whether a person’s income level would be greater or less than $50,000 based on labelled input variables like age, education, job type, marital status, race, and number of hours worked per week. In this example, people could be classified as likely or not to earn a given salary.

Unsupervised Learning

Unsupervised learning methods can be used on data that isn’t labelled.

Clustering and Anomaly Detection methods can be used on unlabelled data.

An example of unsupervised learning is to predict whether someone will purchase a given item based on data that is unlabelled. Here people can be clustered into groups according to their likelihood of making a purchase.

Machine Learning Process

Take, for example, a machine learning process designed to pick out faces in a landscape. In the Training Phase, labelled data (eg leaf, dog, car, face etc) is loaded into the system. The algorithm learns to extract those features (edges, shading, textures) that count as a face. Once trained, the algorithm is ready to be used with test data - ie the images that you need the faces to be identified in. The model can now apply a label to the data that best fits the required features.

Figure 23. Machine learning processes