For much of the history of artificial intelligence (AI), computational power has limited the potential for technological development. Computers have only been able to store and process information at a sufficient rate to make real-time AI feasible for the last several years. Once that aspect of technology advanced far enough, though, it made possible another key aspect of AI: machine learning.

Machine learning is a subfield of AI. It refers to a process by which a computer or program is able to learn and synthesize new information without explicitly being programmed.

Today, machine learning governs nearly every AI application, from chatbots like ChatGPT to social media algorithms to autonomous vehicles and much more. A 2020 survey by Deloitte found that two-thirds of companies were using machine learning at that point, with close to 100% planning to integrate machine learning within the following year.

Understanding Machine Learning

The broader goal of AI is to imitate human intelligence and solve problems in a way akin to the way that humans do. An AI program might have the goal of understanding text written in a natural language, recognizing a visual scene in a photograph, or responding to spoken prompts with coherent answers, for example.

The concept of machine learning goes back at least to the 1950s, when computer scientist pioneer Arthur Samuel defined it as “the field of study that gives computers the ability to learn without explicitly being programmed.” This definition has remained largely unchanged in the decades since.

In traditional computer programming, a programmer must create a detailed list of instructions for a computer to follow in order to achieve the set task. This process works in many cases but is insufficient when attempting to train a computer to perform a particularly complex task, like responding to a spoken prompt with written language. This is where machine learning comes in.

Machine learning utilizes a pool of data to train a computer program. Programmers set a machine learning model to match this data pool and initiate it: the computer trains itself to find patterns and solve problems using the data provided. Human programmers can step in at various points in the process to correct problems, tweak the model, and encourage particular types of results.

Applications of Machine Learning

There are seemingly endless potential uses for machine learning. Broadly, machine learning algorithms can be divided into several categories:

1. Descriptive: a descriptive machine learning system uses data to explain what happened.
2. Predictive: a predictive machine learning system uses data to predict something that will happen in the future.
3. Prescriptive: a prescriptive machine learning system uses data to make suggestions about a set of actions to take.

Machine learning systems can also be supervised, meaning that they are trained with labeled data sets to become more accurate and to “learn” as they process more data. They can also be unsupervised, meaning that they train themselves by looking for patterns in unlabeled data. Finally, so-called “reinforcement” machine learning uses a trial-and-error system to train models with a reward system.

Machine learning depends on a robust data set. The larger the pool of data that the system has to analyze, the more accurate and powerful the result. For this reason, machine learning applications work best in situations that have thousands or even millions of examples of data. These can range from bits of text to photographs, ATM transactions, recordings, and more.

Some of the primary applications of machine learning include:

1. Natural language processing

Natural language processing is the system by which machines learn to understand language that humans speak or write. While this may not seem like a substantial feat, it differs significantly from the data and numbers that computers typically analyze in lines of code. Natural language processing makes it possible for computers to understand language that is received and to create new responses. It can also allow AI programs like Google Translate to create translations from one language to another. Natural language processing is behind chatbots like ChatGPT and virtual assistants like Amazon’s Alexa or the iPhone’s Siri.

2. Neural networks

Neural networks are machine learning algorithms modeled after the human brain in some way. Like the brain, cells or nodes are connected to be able to process inputs and relay outputs, just like neurons. Data can move through the network from one cell to the next, and each cell performs a specific function. Neural networks may be used for visual identification AI, among other things. In an example of a neural network in which AI aims to identify whether a picture contains a human face, each node would analyze the picture for different information and work to produce an output to indicate whether or not there is a face included.

3. Deep learning

Deep learning is an application of machine learning akin to neural networks, but with many additional layers. Generally, the more complicated and layered the deep learning process, the more complex the tasks that system can address. Deep learning is used to power autonomous vehicles, medical diagnostic AI systems, and similarly complicated programs.

Machine learning has numerous applications across many businesses and industries. It powers suggestion algorithms in popular search engines, streaming sites, and more. Social media companies often use machine learning to create news feeds, while facial recognition programs use machine learning to identify features. Even investment firms like hedge funds use machine learning to analyze stocks or other securities.

Finally, machine learning has tremendous potential to revolutionize a host of industries, but it also comes with several areas of concern. One of these is so-called “explainability.” Machine learning systems can be delicate and susceptible to failing on tasks that may seem simple to human beings. The reason for this is that machine learning models must be designed incredibly carefully to be sure that they are performing actions and making decisions according to the programmers’ desired specifications. Machine learning systems can also reflect conscious or unconscious biases—one popular example is the phenomenon of social media algorithms feeding users extreme content.

Cheat Sheet

• Machine learning is a subfield of artificial intelligence (AI) that provides computers a way to learn new information without explicitly being programmed.
• Most recent AI developments in the last decade have largely been due to machine learning technology.
• In 2020, Deloitte found that about two-thirds of companies already used machine learning.
• Machine learning systems can be descriptive, predictive, or prescriptive, and they can also be supervised, unsupervised, or provide learning reinforcement.
• Generally, the larger and more robust the data set that a machine learning system has to work with, the better the results.
• Examples of machine learning processes include natural language processing, neural networks, and deep learning.