The technology sector loves its buzzwords. Sometimes a precise definition can be difficult to tease out because a buzzword can mean something totally different, depending on whom you ask. 

Artificial intelligence (AI) is a catchall term that describes how software can be used to simulate human thinking. Capturing data from Internet of Things (IoT) devices to make decisions based on preset parameters is a common example of AI (think Nest thermometers or automatic reordering of items in your refrigerator). Ultimately, AI would function nearly like a human. The most prominent example is Sophia, the humanoid developed by Hanson Robotics that has been granted citizenship in Saudi Arabia. 

Two subsets of AI that often get confused are machine learning (ML) and deep learning (DL). The underlying technology that allows software to sort huge amounts of data and extract useful information is affordable for a wide variety of companies. However, you need to understand your needs on the front end, whether you need ML or DL. 

ML great at pattern recognition
Machine learning does a great job at what it’s trained to do. Let’s say we build a machine that can archive pictures of dogs. Naturally, we need to “train” the machine to recognize the features of a dog (tail, snout, ears, legs, colorings, etc.). Once properly trained, the machine can successfully detect dog in pictures, distinguishing them from humans and archiving the results. 

But how will a machine trained to recognize dogs react to a picture of a cat? The machine likely will mischaracterize a cat unless it is taught the differences between the two. And once that additional training takes place, what would the machine do with a photo of a horse? 

Machine learning can be a step toward true AI, but it is not AI on its own. Unfortunately, many machine learning products or efforts are being called AI. While this is technically correct, since ML is part of AI, it sets the wrong expectations in the minds of customers. 

More accuracy with DL
Now what if a machine, when provided a picture of  horse, understood it was neither a dog nor a cat, researched different animals on the internet, determined it was a horse and added that information to its knowledge database? That’s an example of a true AI system. 

Deep learning is a more recently evolved subtype of machine learning. In machine learning, the programmer has to teach the system what to look for, and the features have to be specifically pointed out. It’s more like object recognition. 

With deep learning, however, the programmer skips the process of manually entering specific features. Instead, the programmer feeds the images directly into the deep learning system, where the machine categorizes each one according to its own parameters. With DL, the calculations involved are much more complex, requiring the ingestion of far more data in order to return accurate results. 

Moreover, you would not have total control over which features to select. The machine would do that automatically to form a learning process that the programmer would have to verify. Deep learning also has a higher training time.  

If your computational resources are low or if the training materials are limited, machine learning is your best bet. But if you want extremely accurate calculations, deep learning should be your choice. 

But before engaging any technology company for your AI project, be sure you understand your needs and a company’s capabilities to provide either ML or DL (or both).  

By Randall McCroskey April 1, 2018
Tags: AutomationBusiness Intelligence