The Marketer’s Guide to Artificial Intelligence Terminology
The field of artificial intelligence (AI) is comprised of many disciplines, technologies and subfields. There are dozens of terms that are used to describe AI technologies, and the definitions can be complex and confusing.
A large part of our focus with the Marketing Artificial Intelligence Institute is to make AI more approachable to marketers, so we thought a post about terminology was a logical place to start.
Artificial intelligence is the “science of making machines smart,” says Demis Hassabis (@demishassabis), founder and CEO of AI company DeepMind (which was acquired by Google). At a basic level, “smart” means achieving a goal by mimicking human cognitive functions. That goal could be winning a board game, correctly identifying a cat in a photo, adeptly using data from sensors to drive a car or anything else a human can do.
However, the three goals just mentioned are accomplished using different technologies and algorithms. All are examples of artificial intelligence, but each is a distinct field of study under the banner of artificial intelligence. In this way, artificial intelligence is “really a portfolio of technologies,” according to Guruduth Banavar, overseer of IBM’s research on AI. Some of these technologies include natural language generation (NLG), natural language processing (NLP), machine learning, deep learning and neural networks.
Artificial Narrow Intelligence (or “Weak AI”)
Artificial narrow intelligence, sometimes called “weak AI,” is the only type of AI that currently exists: it is purpose-built to either augment or replace humans for specific tasks. But don’t be fooled by the term “weak.” These narrow AI solutions result in highly advanced technologies, like the algorithms that power self-driving vehicles or detect cancer better than humans.
Artificial General Intelligence (or “Strong AI”)
Right now, artificial general intelligence (AGI) is science fiction. It’s what many people imagine when they conceptualize AI: an intelligence that is across the board as intelligent or more intelligent than a human being. Experts are divided on whether or not an AGI is even possible to build. The potential creation of an AGI raises fundamental questions about the nature of human intelligence that we as a species have not yet answered, including whether or not human intelligence can be reverse engineered and recreated in a machine.
An algorithm is a series of steps used to solve a problem. In computer science, algorithms are defined rules used to calculate a function. There are many types of algorithms that are designed to solve different problems.
An algorithm may use data from your Amazon purchase history to predict what items you might like to buy next. Or, an algorithm may plot the best and fastest course for a UPS truck to take when it delivers that purchase. Algorithms do this by making sense of datasets through various mathematical strategies.
This isn’t the fictional AI you see in the movies, but, rather, it’s real-world technology delivering practical solutions for specific business and marketing functions. Many algorithms are built by humans coding sets of instructions that tell machines what to do. But with advanced AI, machines can create their own algorithms, determine new paths and unlock unlimited potential to advance marketing, business and mankind.
For instance, Hassabis taught an artificial intelligence program how to beat video games from the 1980s. The system is “programmed to find a score rewarding, but is given no instruction in how to obtain that reward,” according to The New Yorker. To start, the system makes random moves, sometimes scoring and sometimes not.
However, the program’s algorithm assesses its past moves and determines which ones work best, putting that information into practice in its future games. In this way, it improves. DeepMind’s system went from knowing nothing about a given game to mastering it in a matter of hours using this methodology.
IBM uses the term cognitive computing to describe its approach to artificial (or augmented) intelligence.
In the company's words, “cognitive computing is a comprehensive set of capabilities based on technologies such as machine learning, reasoning and decision technologies; language, speech and vision technologies; human interface technologies; distributed and high-performance computing; and new computing architectures and devices.”
Robotics is a broad field that deals with creating and powering physical automatons. Artificial intelligence is not robotics, though it may be used to make robotic devices perform tasks. The robot or robotic device itself, however, is just a physical body for the artificial intelligence running on the processors within.
Big data is a broad term that means a “large volume of data—both structured and unstructured,” according to SAS. You may hear the term used widely—and loosely—to describe any influx of data into a company or organization. Big data fuels artificial intelligence.
Structured data is ordered data that is displayed in columns and rows.
Unstructured data is any data that is not organized in a specific way, such as Word documents, social media posts and emails.
Machine learning is the process of teaching algorithms to achieve better results when processing data through the use of learning techniques like supervised and unsupervised learning. With machine learning, algorithms can be altered—or alter themselves—to produce more desirable results.
Supervised learning is the practice of running algorithms through training data, then correcting their performance by giving them the “right” answer to the problems they are attempting to solve.
Unsupervised learning does not provide algorithms with the correct answer, or any answer. It is the process of allowing algorithms to analyze and find structure in a dataset on their own.
Pattern recognition is the label given to the activity of machines detecting patterns from data. It is often used synonymously with machine learning.
Natural Language Generation (or NLG)
Natural language generation (NLG) is an AI technology that takes structured data and turns it into text. How the data is translated depends on how the system’s creators code its translation rules. For instance, you would “teach” an NLG system the relationships between data points.
In one simplified example, you might tell the system that a rise of 20 percent or more when comparing data points over time should be referenced as “a significant increase.” Once that rule is written, the system would automatically refer to a rise of 20 percent or more as “a significant increase” when presented with properly structured data. Thousands of human-coded rules like this work together with the machine’s various other algorithms to create full written reports from data.
Right now, NLG is capable of producing thousands of stories in a fraction of the time it takes human writers—with the caveat that commercial solutions are mostly limited to data in spreadsheet form.
Natural Language Processing (or NLP)
Natural language processing (NLP) refers to a machine interpreting what human language means with an acceptable degree of accuracy. We might call this the machine “understanding” what is written. In some usages, NLP might also include elements of natural language generation, as a machine processes language, then generates more of it based on its understanding of a text.
For instance, Google Translate “understands” the text you type, then generates its translation in whatever language you select. NLP might also be used to detect—again, with varying degrees of accuracy—the emotional tone or sentiment of writing or produce a sensible summary of a longer piece of content. Apple’s Siri and Amazon’s Alexa are two popular consumer tools that use NLP.
Automation refers to the process of replacing an activity previously performed by a human either partially or wholly with the use of a machine. In the late 20th century, automation often referred to the replacement of human factory workers with machines in the manufacturing sector. With the dawn of artificial intelligence, however, higher-level cognitive tasks are now able to be—or could soon be—automated.
A neural network is a set of software, hardware or both that is modeled after aspects of the human brain. Right now, neural networks are often organized as layers of interconnected artificial neurons. Each layer’s neurons are “weighted” to prioritize some criteria over others, depending on the goal of the neural network.
Each layer analyzes the results from the previous layer, which means that, together, the layers can engage in more and more “sophisticated decision-making,” according to Neural Networks and Deep Learning. As the network sees the effectiveness rates of various weightings, these are changed and tested to improve results. In this way, the network “learns.”
Deep learning is one “particular approach to building and training neural networks,” according to Pete Watson, who works on AI for Google. This approach is a flexible set of rules and techniques that make it possible for the neural network to teach itself to achieve success on goals it was not trained to solve. For instance, Watson offers the example of an image-recognition neural network he worked on that taught itself to recognize classes of images it was never trained to identify.
These are some of the most important AI definitions that marketers should know. Are there other terms you would like to learn more about? Let us know in the comments and we’ll consider them for a future post. Be sure to subscribe so you don't miss a post.
About Mike Kaput
Mike Kaput is the Director of Marketing AI Institute and a senior consultant at PR 20/20. He writes and speaks about how marketers can understand, adopt, and pilot artificial intelligence to increase revenue and reduce costs. Full bio.