I’m not even going to pretend to explain the technical details, because I barely understand what the words even mean. I just want you to understand the concept in its absolute bare-bones form. That way, when you throw around commentary about artificial intelligence, you can do so with shaky confidence. With that goal in mind, here is artificial intelligence explained simply, as in really simply. To help guide me, I used this article from the super smart people at MIT.
Artificial Intelligence as a Broad Concept
Artificial intelligence (AI) is a term that simply refers to machines imitating intelligent human behavior. It is therefore a very broad term. It is also a rather nebulous term, and debating what is and isn’t AI is a favorite pastime for some people. (I tend to stick to the Who’s the greatest NBA player of all time – Michael Jordan or Lebron James debate. It’s easier on my fragile neurons.)
The application of AI as we know it is thus far limited to specific tasks (like a car driving itself), a concept known as narrow AI or weak AI. The idea of artificial general intelligence (AGI), where an AI system would have full human cognitive ability and be able to perform many unfamiliar and unrelated tasks, is referred to as strong AI (and hopefully far off).
Machine Learning
Because most advances in AI have involved machine learning, the terms are often used interchangeably.
Machine learning is actually a subset of AI that allows computers the ability to learn without being explicitly programmed. In other words, the computer can learn to accomplish a prespecified goal via trial and error. The rules ultimately used by the machine to accomplish the prespecified goal may not be perfectly clear. (Contrast this to traditional computer programming where a machine is given a precise step-by-step list of commands (rules) to accomplish a specific task.)
To make machine learning work, you need to provide the machine (a) a lot of data and (b) a goal. As an example, you can feed a machine a bunch of images (the data) and request that it be able to identify cows (the goal). Alternatively, you can provide a bunch of sales data with the goal of identifying what client characteristics correlate with sales. Or you can provide a ridiculous amount of different types of data to eventually have a machine play a game or a vehicle drive itself. In all cases, the more data the better.
These examples lead us to three general types of machine learning.
Supervised Learning
In this model, the machine is trained with labeled data sets. In the above example, training images of cows would be labeled as such, and the machine would eventually be able to identify cows in unlabeled test images on its own.
Unsupervised Learning
In this model, data is unlabeled, and the machine finds patterns in the data on its own. The sales data example would fit this picture.
Reinforcement Learning
In this model, a machine is rewarded when it makes a right decision. When trained with enough different scenarios over time, the machine can reliably learn what actions to take under different circumstances. The examples of game-playing machines and autonomous vehicles can fit this category.
Again, this is a ludicrously basic introduction to a not particularly simple topic. Even so, all this thinking is taxing my natural intelligence. Isn’t there some sort of machine that can do this for me?