A rose by any other name…Machine Learning, Deep Learning, Artificial Intelligence
What’s in a name? A rose by any other name will smell just as sweet.
The above aphorism is true. Yet we all appreciate the importance of nomenclature. Without clear definitions of terms it becomes hard for people to exchange ideas. We are apparently in the age of Artificial Intelligence(AI). Or is it machine learning? Is machine learning just a fancy name for statistics that has been around for decades? Is deep learning a kind of machine learning or is it AI? There is a lot of confusion surrounding these terms and this article is a humble attempt at clarification.
First, let us understand the source of the confusion. One factor is the fact that these terms are somewhat new. Well, these concepts have been around but their utility was unclear and few paid attention to them. Now they are making measurable impacts on our lives and so we hear these terms a lot.
Another factor is the way marketing of technology products typically works. Marketing managers of companies selling products based on these concepts look for “messaging” that will improve sales of their products. They may not be very interested in following a shared or standard nomenclature. When I worked for one such company, a question we used to get a lot from our salespeople: Does our product do deep learning? It didn’t but due to the confusion around these concepts, we lost some opportunities where customers were looking for “deep learning”, even though our product would have been useful for them.
Now, here is my attempt at clarification:
Machine learning is a collection of methods that allow machines i.e. computers to learn “laws of nature” from data that is collected about it. For example, given data about mass, velocity etc from experiments of moving objects a computer might learn Newton’s laws. Newton’s laws are relatively simple, so an intelligent human like Newton could figure those out without using computers (rumor is that he had help from Apple). I used the word “nature” above very loosely (ironic as I am attempting to clarify terminology!) as the laws or concepts that a machine can learn from data need not be about nature. Machine learning methods can learn about stock trading and car driving for example.
There are several methods or algorithms that let machines learn. Linear regression and decision trees are some simple ones. Neural networks are more complicated. Deep neural networks are neural networks that have many “layers”. Without getting into details of how neural networks work, deep learning is basically the use of deep neural networks, which are a subset of neural networks, which is one of many machine learning algorithms we have found useful.
Artificial intelligence is the imitation or simulation of human-like intelligence in machines. For example, a self driving car (actually a computer driven car!) is a feat(or a soon to be feat) of AI. AI is typically based on machine learning and that is the reason why some people get confused between the two. But AI can and often does include human-learnt concepts programmed into computers. For example, a self driving car may use machine learning to read letters on road signs seen by its camera. But its collision avoidance scheme may be human programmed in the form of a rule like: if LIDAR senses object in car’s trajectory apply brakes. It is important to make the distinction here that this rule or “law of nature” was provided by a human and the machine did not learn this law from data. One might argue that trying to learn this rule from data would be analogous to using a fork-lift to move a paper envelope.
Given the above definitions, one might ask: isn’t machine learning always used for AI? The answer the no. Often machine learning is used to do things humans cannot do e.g. weather forecasting. Since humans don’t have this “intelligence” I would argue that this is not AI.
Did the above clarification help? Let me know if your comments.