Machine Learning (ML) and artificial intelligence (AI) are hot — but they aren’t the same. Knowing the difference and some ML basics can help you make smarter business decisions.
Machine learning (ML) is one of the hottest buzzwords around today. With Microsoft founder Bill Gates saying ML will be “worth ten Microsofts” and Defense Advanced Research Projects Agency (DARPA) Executive Director Tony Tether calling it “the next internet,” it’s easy to see why people are excited.
Another hot term is artificial intelligence (AI). Often, people use these terms interchangeably, but they aren’t quite the same things.
Understanding the differences between ML and AI will help you sort through vendors’ claims when they approach you with new technologies. That will help you avoid purchasing services you may not need and make sure you are using the right ML approach.
In this post, we’ll separate the facts from the hype about ML and AI. We’ll start with a brief history of ML and AI, discuss the differences between the two, and give some basics about how ML works. In a later post, we’ll go a little deeper into how you can make ML work for you.
An Overview of Machine Learning and Artificial Intelligence
Though some think ML is a brand-new idea, Arthur Samuel coined the term in 1959. In his words, ML is a “field of study that gives computers the ability to learn without being explicitly programmed.”
With ML, the computer programs itself by using input data to create output data. With the help of an algorithm, the output data then creates a new program.
AI is a more general term that refers to creating computer systems that perform tasks more intelligently or in a more human way. AI works the way people work—it tries things, learns from mistakes, and changes its behavior for the future.
AI also relies on people to function. Human beings must create the datasets and algorithms to make it work correctly. As Gartner writes, “The rule with AI today is that it solves one task exceedingly well, but if the conditions of the task change only a bit, it fails.”
Think of the relationship between ML and AI like this: You can program a computer to demonstrate AI without necessarily using ML, but a computer that uses ML is also using AI.
A device like Alexa is AI. It accepts input in the form of a human voice asking a question (no more keyboarding!) and then uses the data from that voice to find an answer to the question.
ML allows the device to adapt to your unique speech patterns. Alexa may not recognize your speech at first, but after a few tries—thanks to ML—it will figure it out for future requests, making false answers less frequent.
ML vs. Traditional Programming: A Deeper Dive
In 1998, Tom Mitchell defined ML like this: “A computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T, as measured by P, improves with experience E.”
Let’s unpack this by looking at ML vs. traditional programming.
In traditional programming, the programmer uses input data and a set of code rules to create a program. The coded program runs on the computer and produces a desired output. Below is how traditional programming works:
With ML, in contrast, the programmer creates a predictive model by identifying the data samples for a true/false condition and then manipulates the data to pass to a predictive algorithm that creates new rules. Below is how ML works, where a model is built from example inputs to make data-driven predictions:
Types of Machine Learning
Broadly speaking, three types of ML algorithms exist:
Supervised learning is the most popular ML paradigm. It is also the easiest to understand and the simplest to implement. You can think of supervised learning as “task-oriented” because it typically focuses on a single task. Programmers feed a lot of data into algorithms until they can accurately perform the desired task.
Supervised learning can be two types – regression or prediction and classification. Predicting the number of items that will sell over the next three months from an inventory of similar items is an example of regression or prediction. Predicting the weather from one of three possibilities (Sunny, Cloudy or Rainy) is an example of classification.
In unsupervised learning, programmers again feed large amounts of data into an algorithm, but they also give it the tools to understand the data’s properties. The program can then learn to group, cluster or organize the data so that a human (or another intelligent algorithm) can make sense of it.
One example of unsupervised learning might be recommendations of what to watch next from your streaming service. These are known as recommender systems. Another example could be algorithms that use customers’ buying habits to group them into similar purchasing segments so companies can market to them better.
Reinforcement learning relies on recognizing mistakes and successes or hits. The algorithm is “rewarded” for performing correctly, and it is “penalized” for performing incorrectly. This model requires no human intervention. The algorithm teaches itself to maximize reward and minimize penalty.
AI and ML represent the future of computing, but you must understand how they are alike and different to make the best decisions and to communicate well with vendors. You can then begin using some basic ML concepts to determine which type of approach will meet your needs. We’ll take a deeper look at more definitions and how to start writing simple ML programs in our next post.