If you’re like most marketers, you’re probably trying to get in on a little AI action to raise your game and keep up with your competition. And if you’re like most marketers, you might not understand exactly how it all works yet. Join the club.
As you discover new smart tools for your company, the first step towards making smart buying decisions is to understand the difference between machine learning and artificial intelligence. These terms are often used interchangeably, but they are definitely not the same thing.
“AI is any technology that enables a system to demonstrate human-like intelligence,” explained Patrick Nguyen, chief technology officer at 7.ai. “Machine Learning is one type of AI that uses mathematical models trained on data to make decisions. As more data becomes available, ML models can make better decisions.”
Let’s break that down, shall we?
You don’t have to have a smart home to come into contact with machine learning. In fact, companies like Facebook and Google have been using it for some time now to organize Big Data, speed up search or optimize advertising.
According to the University of Maastricht, “Machine learning algorithms are widely employed and are encountered daily. Examples are automatic recommendations when buying a product or voice recognition software that adapts to your voice.” Sounds familiar, right?
Machine learning is based on what is known as “neural networks.” If it sounds complicated, that’s because it is. But in a nutshell, neural networks are built for training and learning. They rely on certain factors of importance to determine the probable outcome of a situation and need to be programmed by humans first.
A neural network programmer must adjust the factors of importance (otherwise known as weights) in the outcome until the network reaches the required result from the information it has.
Now, just imagine a human programmer manually setting up a neural network for every possible outcome of a Google search! That’s where machine learning comes in.
Once the neural network has been perfected and the machine understands how to adjust the factors of importance on its own, it can train itself to improve accuracy without human intervention. And once the machine is trained, it can sort new inputs through the network and produce accurate results in real-time (think voice search).
It’s an incredibly complex and clever technique, but still, machine learning doesn’t possess any real intelligence.
Algorithms don’t need to understand why they self-correct and improve, they are merely programmed to do so. However, once machine learning reaches a point where it can reflect and interact with humans in a convincing way and make decisions by itself, that’s when artificial intelligence is at play.
The reason we hear the two definitions interchanged is that AI cannot exist without machine learning—although machine learning can exist without AI. Think about an algorithm that can identify patterns in data based on specific weighted factors, or perhaps identify all types of images that are the same.
“If we plug several photos of cats doing different things or in different places into a computer, but all the photos are still tagged as cats, then the computer will learn from each photo it is shown,” said Kamelia Aryafar, Ph.D., director of machine learning at Overstock. “Eventually, it will recognize that the cat is the common denominator in each set of data, in turn helping the computer learn to identify cats.”
There isn’t really anything humanly intelligent about that. However, when that algorithm is connected to cameras and speakers, detecting objects in front of it and given a voice that responds to questions, it mimics human intelligence. It has become artificial intelligence.
When a machine can tell the difference between objects and make a choice to discard or accept them, based on understood criteria, AI is born. In fact, any time a decision is being made by a machine, that is artificial intelligence and has gone beyond mere machine learning.
Two types of AI
Artificial intelligence can be further broken down into two major types: general or applied. General AI is a lot harder to achieve than applied AI. And in fact, applied AI is very much tied to the given examples of machine learning, in which computers make a decision for themselves.
Consider LinkedIn Messaging for a moment. The application predicts possible answers to a message, showing everyday applied AI in use. “Predicted responses are generated by machine learning models trained on large amounts of message data: these models find the most common responses to messages whose linguistic characteristics (i.e., sequences of words and phrases) are similar to an input message,” Nguyen explained.
“This is called predictive Natural Language Processing, NLP for short,” said Aryafar. “It is reading the language or text written like a promotion, new baby or job changes and formulating suggestions based on what it has scanned in the text.” Pretty cool, right?
General AI is a much broader category and requires the machine to understand, interpret and respond to a wide range of tasks and stimuli, much as humans do, mimicking the human brain.
While we’re still scraping the surface of the possibilities of general AI today (it may be awhile before all humans live alongside Sophia the robot) applied AI is already very much in force. And the root of it all is the neural networks necessary for machine learning that powers so many of the devices we already use.