You don't have to be a computer programmer to understand machine learning

One of the biggest buzzwords in technology today is "machine learning." It is being used all around us, and we usually do not even notice.

Recommender systems, search engines and your email all use machine learning to make your life easier. Machine learning involves an algorithm that goes through a large set of data, learns something from it and then makes decisions or predictions based on what it learned. The big difference between this and other computer programs is that it is not a set of explicit instructions. Instead, giving the machine some general guidelines and training it with large sets of data allows it to learn and revise itself.

A common example is spam filters. A spam filter has a set of criteria that it uses to decide whether an incoming email is spam. It might look at things like length, the frequency of certain words or images. When you mark something as spam or move a piece of spam back into your inbox, the program updates its criteria. It learned from the new information and will make future decisions based on the new criteria.

Another example is teaching a machine to recognize stop signs, something self-driving cars have to be able to do. The machine is given thousands of photos of stop signs, from different angles, in different weather, etc. It decides how to weight various criteria of what it sees in the images, such as color, shape and size. After it has this template, it can then apply the criteria to new images and give a probability that they contain a stop sign.

For a short overview of machine learning, take a look at “Machine Learning: The New AI” by Ethem Alpaydin. It starts with a general description and history. Then it goes deeper into important methods used for different tasks such as facial recognition and language translation. It is not aimed at programmers or mathematicians, and does not have detailed math to decipher. At the end there is also a little bit about the future of machine learning.

As helpful as it can be, machine learning does have its limits and problems. It is reliant on data. If the data it looks at is biased then the results will be, too. You cannot train it with photos of yield signs and then expect it to identify stop signs correctly. If you give it the wrong set of data to look at or use a poorly modeled algorithm, then the results can have negative consequences for society. These downsides are addressed in “Weapons of Math Destruction” by Cathy O’Neil.

It may sound scary or difficult to learn but you can create your own machine that is capable of learning. If you want to find out how to get your computer setup with appropriate software, see what math is involved, locate data sets and examine different practical uses for it, take a look at “Machine Learning for Dummies” by John Paul Mueller and Luca Massaron.

You’ll find these titles in the nonfiction collection at the Iowa City Public Library.

Todd Brown is senior librarian at the Iowa City Public Library.