What is the Difference Between AI, Machine Learning and Data Science?
Elena Samuylova
Chief Product Officer, Mechanica AI
In a newly hyped field, concepts change meaning every now and then. Some years ago, "big data" gained momentum. Now, it is all about AI. The term may be new for some, but is in fact quite familiar to the professionals in the industrial sector. Those witnessed the rise of expert systems and spoke about neural nets way before it became mainstream. Still, this new "AI spring" created some confusion: what exactly do all these technologies mean? How are they different from previously used approaches?

In this blog, we will try to explain the most popular terms of the new era of data analytics.
Artificial Intelligence (AI)
Simply put, AI refers to the ability of computers to perform complex tasks and exhibit human-like intelligence. This can mean pretty much anything - from playing games to estimating the probability of defects on a production line. The concept of AI by itself does not specify which exact technology is used, but rather refers to the complexity of the intellectual task that is now performed automatically. At the same time, it is machine learning that mostly powers AI capabilities today, and thus the two terms are often used interchangeably.
Narrow VS General AI
In popular media AI is often portrayed as some kind of autonomous "thinking machine" that has universal cognitive abilities. Despite having inspired multiple sci-fi plots, such General AI does not yet exist in practice.
Today we are dealing with so-called Narrow AI: it means that a machine can perform a specific intellectual task and reach, or surpass human-level results.
Examples include generating sales predictions, recognizing speech, or defining the correct set of parameters for an industrial process. However, each such AI is trained separately and is strictly limited to its context.

Industrial expert systems represent the narrow AI of the past - with knowledge and "if-then" rules programmed to represent human-like reasoning. In contrast, the key feature of the present-day AI is its ability to solve complex tasks without being explicitly explained how to do so. Here comes machine learning.
Machine Learning
Machine learning is a set of technologies that allow computers to learn from examples. Fed with available data, machine can identify underlying patterns and teach itself to perform the given task automatically. Instead of trying to explain a computer how a dog - or a surface defect in steel - looks like, one can show enough visual examples to let it figure out for itself. Same goes for being able to predict the production output learning from past process history, and so on.
The process of "teaching" the machine is guided by an expert data scientist. However, the resulting machine learning model is uniquely based on data as opposed to being explicitly programmed.
The mathematics behind this approach has long been known, but its practical use only became possible with the availability of large datasets and computing power, leading to the widespread adoption in the recent years.
Deep Learning
Highly popular, this concept refers to a family of methods used for machine learning. Neural Networks, in turn, describe the architecture of such models where signals can be transmitted between nodes, remotely resembling the way neurons connect in a biological brain.
The frequent use of these terms - and a number well-known cases, such as computer winning in the game of Go - led many to perceive deep learning as a superior machine learning method.
Still, it is just one of many techniques: well-fit for some tasks, and less for the others. Deep learning is typically the method of choice when a lot of training data is available, or can be artificially generated. For example, it is applied for the analysis of image and video streams, and - less frequently - for high density sensor data.
Data Science
This is an umbrella term that refers to the application of statistical methods for data analytics. The job of data scientist includes using advanced techniques such as machine learning, but does not limit to them exclusively. More traditional methods are also applied, and tasks may vary from building deep learning models to descriptive analysis and data visualization. This term is well-suited to define the job of an expert, but it makes sense to specify the scope and technologies when talking about specific "data science" use cases.
Big Data
So what about big data, one might ask? Is the term completely out of fashion?
Truth is, big data by itself is more of an infrastructure problem than a good way to describe what can be built with its help.
The term big data refers to the hardware and software solutions that are necessary to capture, store and organize access to such large datasets where traditional data processing means become inadequate. Sometimes, it is also used to refer to advanced data analytics and products and services powered by it. However, with the recent rise of AI, we suggest to stick to this term to avoid too much chaos in vocabulary!
Have any observations on how these concepts are used in practice? Let us know!
More posts you may like
Made on