Q. Can you please explain what artificial intelligence (AI) is? I've heard a bunch of different definitions, but I'm not sure which one is correct.
A. The major dictionaries offer a variety of definitions for AI, but a common theme is that AI must emulate intelligent human behaviors or be able to perform tasks that previously required human intelligence.
Some people misread the second part of that definition as saying that AI refers to technology performing tasks that humans previously had to handle with their own brains. That view misses the mark. To make the distinction clear, consider the following example.
A calculator can complete certain mathematical tasks that previously only humans could handle. However, the calculator is not applying any cognitive intelligence to the problem. A human provides the inputs and "tells" the calculator what functions to perform. The calculator does not have to apply any intelligence to make any decisions. As such, calculators are not AI.
Now consider a situation where you want to compare the number of cars and motorcycles in a parking lot (the type of thing you might do in an inventory audit). By applying image-recognition technology (a subset of AI that we cover in more detail below), a computer can scan a parking lot, tell the difference between cars and motorcycles, and provide accurate counts of each. The image-recognition technology that identifies the shapes and characteristics of the vehicles requires the application of intelligent decision-making. This is only possible with AI.
It's important to understand that AI is an umbrella term covering quite a few technologies. Among these are machine learning, deep learning, natural language processing, speech recognition, the aforementioned image recognition, and robotics. These are all AI, but all AI is not, say, machine learning. To further complicate things, these technologies are subcategories of AI that often have their own subcategories.
AI: General vs. narrow
Fortunately, you don't have to know every subcategory, niche, and nuance of AI. You should know the following.
There are two main categories of AI: artificial general intelligence (AGI) and narrow artificial intelligence (NAI).
AGI refers to a machine that, in simple terms, can think on its own like a human does. Such a machine would be capable of learning to solve any number of problems without human input and would be able to adapt and evolve on its own. You've seen this type of AI exclusively in fiction, as it does not yet exist in the real world. Among the most notable examples are Skynet in the Terminator movies and HAL 9000 in 2001: A Space Odyssey. (Those technically are examples of artificial super intelligence, but the distinction is not important for this discussion).
There is significant debate regarding how close we are to achieving AGI, but most experts and futurists believe we are decades away from the creation of an actual AGI.
In contrast, NAI is in wide use today. As the name implies, NAI refers to machines created to handle a specific task or a limited range of tasks. NAI can make autonomous decisions in its area of specialty and often outperform humans in that area. For example, the initial version of IBM's Watson defeated the top human competitors on the game show Jeopardy! back in 2011, but it couldn't play, much less win at, any other games — or do much else really.
NAI includes a sizable number of subcategories, but only a few have achieved enough adoption that you are likely to interact with them on a regular basis.
Let's take a brief look at those.
Machine learning refers to the method of providing specific structured data points to an application and programming the application to determine a specific defined outcome based on the data points that have been provided. Structured data is data that is organized in a logical and consistent format, such as a database or a tax form.
An example of machine learning you've almost certainly seen are the recommendations made by online stores such as Amazon or online content providers such as Netflix and Spotify. The recommendations are a specific prediction being made by the machine learning algorithm, based on data such as your viewing habits and ratings combined with those of similar customers.
Deep learning is a subset of machine learning with algorithms designed to emulate human neural networks. What sets deep learning apart is its ability to learn and complete tasks such as structuring unstructured data (data not organized in a specific format, such as research articles). Deep learning makes it possible for machines to handle many types of tasks with little human guidance. It's like asking someone to organize a shoebox full of receipts, and then they just do it. They don't need specific directions because they can figure it out on their own.
You can provide deep learning machines with a combination of structured and/or unstructured data to process, learn, and draw conclusions from in order to produce a variety of requested outputs that can dynamically change.
We are far less likely to regularly interact with deep learning models than their machine learning counterparts, but deep learning has been successfully applied to a number of medical challenges, including making cancer treatment recommendations and early detection of heart disease.
Speech recognition refers to the process of "hearing" words and translating those words to text. Anyone talking into their phone to dictate a text message is leveraging speech-recognition technology.
Natural language processing
Natural language processing is a broad term for recognizing natural language to execute a task. Whereas speech recognition involves hearing and recognizing words, natural language processing refers to the ability of a machine to hear the word and understand its meaning well enough to respond with an appropriate action.
For example, virtual assistants such as Google Home, Apple's Siri, Amazon's Alexa, and Microsoft's Cortana can "hear" a question or command and respond accordingly. If you ask Siri to tell you the football scores, it can. If you tell Google Home, "Google, turn on the outside lights," Google Home will do so.
Image recognition refers to the application of AI to recognize an object based on what is seen and categorize that image.
Facial recognition is a popular but controversial example of this technology (see "What CPAs Should Know About Facial-Recognition Technology"). Another common example is Google Photos' ability to categorize your photos based on the objects in the photos. If you store images in Google Photos, search for "Animals" and it will retrieve all photos that have an animal in them. Take it a step further and search for "dog" and you will only be presented pictures that have a dog in them.
Robotics uses AI to sense surroundings based on a wide variety of sensors and variables, and subsequently respond to those environmental factors.
The auto industry has started using these technologies to improve the safety of our cars. For example, light detection and ranging, or LIDAR, is like RADAR but wit h light pulses. This technology can enable cars to sense road lines and keep the vehicle in the middle of the lane or to "see" the car ahead suddenly stopping and apply the brakes faster than a human can, increasing the odds of avoiding an accident.
Only the beginning
These real-world examples show that AI is here to stay. The breakthroughs and innovations that have so rapidly occurred over the last decade are only the beginning of what is surely going to have a significant impact on the way we live in the future.
— By Byron Patrick, CPA/CITP, CGMA
About the authors
Kelly L. Williams, CPA, Ph.D., MBA, is an assistant professor of accounting at Middle Tennessee State University. Byron Patrick, CPA/CITP, CGMA, is senior applications consultant at botkeeper.
Submit a question
Do you have technology questions for this column? Or, after reading an answer, do you have a better solution? Send them to firstname.lastname@example.org. We regret being unable to individually answer all submitted questions.