What’s the difference between AI and machine learning?
Valérie Bécaert Valérie Bécaert
August 22 6 min

What’s the difference between AI and machine learning?

Artificial intelligence (AI) and machine learning (ML) are two terms that loom large in today’s technological world. AI has long held a place in the public consciousness and pop culture, from fantastic visions of tomorrow to the ever-increasing number of applications we see in everyday life.

Machine learning, although not as well-known, is steadily beginning to gain traction. Many people are aware of it even if they don’t know what it precisely means. In this article, we explore the definitions of AI and machine learning, and the differences between them. We also consider how they fit into humanity’s present technology and future innovations.

Machines that think as we do

The term “artificial intelligence” was first coined by American academic John McCarthy in 1956. McCarthy defined it as “the science and engineering of making intelligent machines.” He was one of a host of pioneering thinkers, which included Alan Turing and Isaac Asimov, who theorized that machines might “think” in similar ways to humans. This meant the ability to acquire knowledge and solve complex problems. AI has come to refer to a broad range of technologies with this aim.

Machine learning is sometimes used interchangeably with AI, but this is incorrect. It’s a subset within the broad field of AI. It is a type of AI, not a synonym for it. Let’s explore what machine learning really entails.

The ability to learn and be trained

Typical computer programs solve tasks using explicit instructions that have been provided by a programmer. However, some tasks are difficult to solve with an explicit sequence of instructions. For example, recognizing human faces in images or translating text while retaining its original meaning. These require an implicit understanding of the task at hand.

Machine learning is the study of algorithms that allow computer programs to perform specific tasks using data instead of explicit instructions. These programs rely on previously-seen patterns to infer their next prediction or decision. They do so by building a mathematical model based on sample data, known as “training data,” which includes inputs and the desired outputs.

This mathematical model is “learned” through an iterative process called “training,” where the machine learning program is repeatedly presented with training data and is tasked with predicting the desired output. With each prediction, the program is corrected such that its next prediction is more accurate. After a certain amount of time, the program will have learned a complex implicit understanding of the task such that its decisions appear subjective.

Today’s advanced learning machines can do all this because they have what’s known as an artificial neural network: an approximation of the workings of a biological creature’s brain, complete with artificial neurons transmitting data back and forth. It allows them to classify information, analogous to human sensory data like sights and sounds, according to the things they recognize, inspired by the way we do – by “making connections”.

In fact, neural networks (aka. the connectionist approach) started by making high level observations about the neo cortex and how it seems to be constructed from a network of simple units (the neurons). However, today's neural networks have evolved tremendously since these early days and have very little to do with brains.

Better translation by machines

One excellent example of machine learning in action is machine translation, where a machine is given various examples of how words, phrases and sentences have previously been translated from one language to another and it learns from them.

And it’s very good at it – a perfect application of machine learning. In fact, many of those who study linguistics veer toward descriptivism rather than prescriptivism, in recognition of this lack of objective certainty. For example, babies can learn a language simply by listening to it and they can master a lot of subtleties without having to learn grammatical rules.

Instead of consisting solely of hard and fast rules like mathematics does, language has evolved in a much more organic way. With so many variations and idiosyncrasies to deal with, it’s better to have a machine learn them all by absorbing large quantities of text, than for us to try to program them one by one.

In other words, it learns how humans actually use language in their everyday lives, instead of merely applying relatively simple pre-programmed rules such as “word X in French is word Y in English”. This is why machine translation services are now able to provide much more nuanced and realistic results than ever before.

AI’s most fruitful area

This discovery – that machines, like people and animals, learn best from a cycle of repeatedly observing and doing – has dramatically changed the field of AI and the entire technology landscape. Instead of attempting to program a machine to be “smart” and anticipate every eventuality, now the machine teaches itself using countless examples, provided by humans, of what does and doesn’t work under various circumstances.

It’s a technology that’s formed the foundation for countless revolutionary developments such as self-driving cars. These autonomous vehicles use pre-made maps or satellite navigation, combined with data from a camera and other sensors, to analyze their environment, and navigate it. They can also work with real-time cloud data, making decisions based on traffic information, weather, and other factors.

Our hyper-connected world of the Internet of Things (IoT) means that machines now have an abundance of “study materials” to learn from. The two developments drive each other– machines that learn, and an increasingly rich digital habitat for them to “experience” and learn from.

Extending this idea a little further, you could consider machine learning as a form of evolution, albeit highly accelerated: the Darwinian self-optimization of a machine to meet the demands of its environment. As a matter of fact,there is a whole set of machine learning algorithms, called genetic algorithms, based on this. A digital expression of one of the most fundamental natural laws. It seems the saying “practice makes perfect” applies to all life on Earth – even the artificial.