What is the Difference between Artificial Intelligence, Machine Learning and Deep Learning?

By Will Bodansky, Head of Operations at TrueFace.ai

Look at any tech news site, blog feed or forum and it’s likely you’ll come across Artificial Intelligence, Machine Learning or Deep Learning. They’re such hot terms at the moment that they seem overused and almost meaningless, solely employed for the purpose of bringing attention to a product or article. However, they’re real terms with amazing real world impact, and if you’re one of the curious souls that wondered what they meant and what’s the difference between them, then you’ve come to the right place.

Artificial Intelligence is more of a general term, that can be sub-categorised, and it is from this general term in which Machine Learning and Deep Learning were born. I’ll get into this birth in a bit, but first let’s dive into what exactly is artificial intelligence (AI). AI, literally speaking, is intelligence exhibited by machines, but this can be broken down into two main areas; applied and general. General AI is the development of machines that have the same capabilities, and so act in a similar way, as human intelligence. These are the ones you see in the movies, whereby some hero manages against all odds to defeat a machine with equal or superior capabilities than that which a human displays. Narrow AI, or Applied AI, is a more realistic construct that programs and enables computers to perform specific tasks at least as well as a human could perform them.

Now you might be thinking, well surely these programs are just following predetermined rules, how do they exhibit intelligence, and how does this apply to the world we live in today? This is where Machine Learning comes in.

Creating software has historically been about setting rules for the program to follow, an example of this in practice this may be an if-then-else function that states what action to perform when a specific event occurs. Machine learning breaks away from this thought process and allows the program to learn from provided examples, extracting knowledge from large datasets, and then applying this knowledge to recognise patterns it uses to identify the most probable outcome given the information available. With this new found intelligence, the machine can learn complex patterns and produce predictions with a higher degree of accuracy than previously achieved.

Deep learning is a subset of Machine Learning. It makes use of Neural Networks to solve real world applications of Machine Learning, in the sense that it mimics human decision-making. This process can be extremely complex as it requires a lot of information (i.e. a dataset), to be able to properly understand a concept, and then determine what this information means and provide a meaningful outcome.

Let’s take TrueFace.ai’s facial recognition APIs, where you can use deep learning through language agnostic apis and recognise a face to an almost perfect degree of accuracy. Via training the algorithm with datasets of faces, varying the lighting, the distance the photo is taken from the face, the angle at which the photo is taken, and of course the person who’s face it is, the program can learn where to measure specific points on the face, how to process this information, and thus give the result of a match or identification of an individual.

In short, it is more constructive to understand how the different forms of machine learning relate, and how, to put it generally, it can be used to help us make predictions, and thus forms part of the more all-encompassing category of Artificial Intelligence.