Artificial Intelligence is a broad term that applies to several technologies that learn based on experience. They process a large data set in search of patterns. Then, they use those results to improve their ability to recognize new patterns.
This is the feature that earns the designation ‘intelligent.’ By learning, I mean the ability to recognize patterns and the ways that they change. It’s not the full-featured learning that humans do. Artificial intelligence is to intelligence as artificial flowers are to flowers. They look similar, until you look closely.
These technologies have an almost infinite capacity to learn, but they will never understand anything. They have no empathy, compassion, or sense of right and wrong. They are pattern recognition machines. They can find patterns, but they can't tell you what the patterns mean.
The ability to find patterns everywhere is what seems intelligent. Because the systems can recognize patterns, they can also predict the next item in a set. Unless an unforeseen force (say a pandemic) interrupts things, they perform this task thoroughly.
Unfortunately, the world is a series of unforeseen events.
An AI Analogy (and alliteration)
Imagine that someone filled your local sports arena with 12-inch squares of fabric to a depth of 30 feet. The squares are every imaginable color, weave, and age. Some are new, some are old, some worn out. Some have hems and some have frays.
The area of the playing field is the ‘pitch,’ which is 75,000 square feet. So, one layer of 12-inch square pieces of cloth would have 75,000 squares.
A piece of shirt fabric is about fifteen thousandths of an inch thick. A pile of 66 would be an inch high. It would take around 800 to make a pile of 12 inches (66 x 12). A 30-foot-tall pile would take 2,400 (800 x 30) layers.
To fill the arena, you would need nearly 2 billion (2,400 x 75,000) squares of cloth.
Enter a sorting robot that can catalog each square and the way that it relates to all the other squares. It might match and categorize colors, thread size, degree of wear, proximity to other squares, location in the pile, or any other attribute of the pieces of cloth you could imagine and many you can't. It could tell you all the possible connections between those squares. It could show patterns that would help you see and understand what's in those 2 billion pieces of cloth more clearly.
Some of it may matter to you, like how many linen squares you have or whether some have holes in them. Some of it probably won't, like whether there is a square with a microscopic oil stain. It depends on what you care about. But neither the robot nor the 2 billion pieces of cloth will help you figure out what matters or why.
AI is artificial, but it's not intelligent.
All the things currently called AI are a form of Machine Learning (ML). ML does for data sets what our robot did for cloth. It identifies every imaginable connection between all the pieces of data in the set. It looks for patterns and calculates the features of those patterns. Like our robot, they help you understand things. They cannot understand anything.There is a way of thinking about AI that can understand things, Artificial General Intelligence (AGI). None of the current applications of AI are AGI. Many experts think that the current fascination with Machine Learning, while powerfully productive, is a dead-end for creating AGI.
In short, AI is a set of technologies that can identify patterns and the things that change them. This new way of doing math (AI is a form of math) is possible because we have so much computing power. By repeating very simple formulas many, many times (billions or trillions), data scientists are able use AI to do matching and predicting.