What Is AI, Really?
Everybody needs to understand what artificial intelligence means. Let’s start our Byte-Sized Insights journey here.
One of the tech world’s most recognizable buzzwords, artificial intelligence, or AI, was once confined to pop culture staples like The Terminator and The Matrix. Now, it has permeated into our daily lives to become an unmistakable presence in the world of technology. But what exactly does the term AI mean? Should we really be worried that robots are going to become sentient and try to control our minds in a dystopian future? Probably not for the time being, which this engineer who got fired from Google learned the hard way.
Part of our mission in this blog is to demystify the terms you’re hearing every day — AI, machine learning, automation — and show how accessible they can, and should, be. We’ll start right here, with our first question you might ask if you’re living in the 21st century: “What is AI, really?”
Artificial intelligence is the study of how computers can carry out tasks that a human being could perform. What this boils down to is the question of how to best simulate human decision-making and problem-solving into probabilistic processes that can be interpreted by a machine that only understands ones and zeroes. That’s what we’re referring to when we say something is “AI”.
Today, it’s impossible to avoid AI. Take, for example, the new Google Pixel Super Bowl ad featuring Amy Schumer erasing her exes from photos. That’s AI. The technology that results in your next Uber from Vegas taking you home without a driver in the front seat. That’s AI, too. And the algorithm that Netflix uses to recommend your next binge? Yup, that’s AI… as well as some strategic marketing, but that’s a different story.
This brings us to another question: how do we go from this incredibly complicated goal of mocking human intelligence to actually really smart machines like this? That’s what we call machine learning, or the practice of “teaching” the computer how to do the right thing.
In classical computer programming, a rule-based command is given to the computer — for example, “if 47+53 is equal to 100, tell me the statement is true; otherwise, tell me it is false.” This command is broken down, or compiled, into the binary language that the computer is built to interpret, and the computer spits out the correct answer. Where machine learning comes into play is problems that the computer has never seen before, or can’t answer with a defined set of rules.
Let’s continue with the addition example by imagining that you’re teaching a child how to add. Assuming the child knows how to count, and understands the concept of larger and smaller numbers, you ask the child how to compute 47+53. Since this is a new task for the child, they may not know where to start. So, we encourage them to just take a random guess. “47?” they ask, and you tell them that the answer is actually greater than 47. The child’s next guess is much larger now, perhaps “250?” But you let them know that this time it is smaller than that. This back-and-forth, guess-and-check process can continue until the child gets closer and closer to the correct answer.
Computers aren’t really using AI to do basic arithmetic — in fact, some AIs are actually worse at simple calculations than a calculator (see how ChatGPT gets some basic facts wrong in this article from the Wall Street Journal) — but the scenario we just walked through encompasses the general idea behind all your favorite technologies. What all of these engineers and researchers are doing is feeding data to algorithms (albeit much more involved than what I just presented) that learn by making guesses and comparing them to the right answer until they have developed complex formulas to predict the answer to never-before-seen questions.
For example, your favorite chatbot will answer your questions by taking all of the news articles, books, and blog posts it has seen on the Internet to iteratively guess what the most likely next word in its response will be. Similarly, a self-driving vehicle will train on its huge database of images of roads, street signs, pedestrians, and other vehicles to identify that the object in front of your car is a traffic cone.
That’s really all it is, folks. Okay, not really. There’s a lot more we have left to learn and understand about AI. But for now, if someone asks you what artificial intelligence is, you can tell them it’s just like teaching a kid how to add numbers.
So, have you decided whether or not to believe the AI hype yet? Read our next article to see why we totally buy it … for the most part.