What is industrial AI?

To define industrial AI, we must first define AI itself. Although the field of artificial intelligence has existed for over half a century, it has no clear and all-encompassing definition. Further, the lines between AI and adjacent fields like machine learning, big data, predictive analytics, and IoT are often blurred, as are the lines between AI and subfields like deep neural networks and cognitive computing.

For our purposes, artificial intelligence refers to those computer science techniques and technologies that allow software to exhibit ‘smarts’—in other words, to do things that seem human-like. This can include things like making decisions, recognizing objects, or understanding speech. It really is a very broad term.

Strictly speaking, machine learning (ML) is a subset of AI. ML refers to a set of techniques that allow us to create AI software by training that software with data) to display some desired intelligent behavior. This is as opposed to, for example, explicitly programming our software with a bunch of rules to generate our desired behavior—and it’s a very powerful concept.

It is for this reason that, while machine learning is only one way to build an artificially intelligent system, for all practical purposes ML and AI are used interchangeably today. All the interesting activity in AI is in machine learning.†

What about cognitive computing? It’s a bit more esoteric a term, usually used
to highlight capabilities akin to humans’ higher level thinking and reasoning skills. An example would be the ability to determine the sentiment expressed in text or images, or what objects are present in pictures. But again, for all practical purposes, the term is most often used interchangeably with AI—in fact, it’s the preferred term in some regions of the world—and the work in this field is based upon machine learning.

How does this relate to big data? Well, data is used to train the machines,
and the more you have of it the better (assuming it’s high quality data).9 And how about predictive analytics? Well, machine learning can be a more powerful way to make predictions, and one that can learn from patterns in the data. But simple averages and other formulas can be used for predictions as well... these need not be based on ML/AI.

Finally, for enterprises whose operations involve the physical world, the industrial internet of things (IoT) is an increasingly important source of insight into the status, location and performance of enterprise assets (see Figure 1). Because IoT devices and sensors can number into the millions, and can report status with millisecond resolution, the resulting data volumes can quickly become voluminous, lending themselves to the application of machine learning techniques.

To what then does the term “industrial AI” refer? Well, certainly the word industrial has certain immediate connotations, primarily of manufacturing and heavy industry. But to limit our scope to just those industries would be to miss the less obvious connections between a broad set of related use cases, the environments they exist within, and the common challenges and requirements that they give rise to.

Rather than referring to a set of vertical industries, by industrial AI we’re referring to a class of applications that can exist within any vertical:

We define industrial AI as any application of AI relating to the physical operations or systems of an enterprise. Industrial AI is focused on helping an enterprise monitor, optimize or control the behavior of these operations and systems to improve their efficiency and performance (see Figure 2).