AI vs Machine Learning Explained in 7 Minutes

Hey everyone! Austin here. If we haven't met, I'm a software engineer and graduate student studying artificial intelligence at Georgia Tech. If you’d like to learn more about AI, check out my website aiwithaustin.com.
Today, we're diving into two terms that are transforming billions of lives, whether people realize it or not: Artificial Intelligence, and Machine Learning. These terms often get confused, conflated, and used synonymously. But what's what? That's what we'll be uncovering in this video.
Artificial Intelligence
Let's start with Artificial Intelligence. At its core, AI is an attempt to answer the philosophical question: Can we make machines - which are, metal and electricity - respond to situations in ways we'd consider intelligent without explicitly programming the response?
Imagine this scenario: Your doorbell rings when you're out of the house, and you get a text saying, "Hey Austin, Mark just dropped off a package." Any human could easily write that text, but getting a computer to do it? That's where things get a little bit more interesting.
Here's the thing about computers: they're 100% literal. They do exactly what we tell them to do - nothing more, nothing less. This is why programming can be both powerful and frustrating.
Think about it: when we make tiny errors in code, like missing a semicolon, the entire program can crash and we call it a "bug." We might wonder, "Why doesn't the computer just fix it?" Well, because we never told it how to fix semicolons or even that it should try.
This traditional approach is called rule-based or explicit programming. It's like saying, "If Mark drops off the package, send the text 'Mark dropped off a package'. If Lucy does it, send 'Lucy dropped off a package'." You can see how this gets cumbersome quickly. You'd have to write a rule for every possible person who might drop off a package!
Artificial Intelligence aims to go beyond this limitation by using a function to calculate the response. This function is called a “model”. An AI model is quite simply just a function. Just like from your algebra classes. This model receives input, and gives back output. The input could be the pixels from a camera of Lucy or Mark. These pixels show up in front of the AI’s camera, it figures out that it’s Mark, and it sends the text. Yet, even in this case computers are still literal even when they’re being controlled by an AI model. The difference is that the model is dynamic. It exists along a continuum. If the face size is 900 pixels, it knows it’s Mark, if it’s 892 it knows it’s Lucy. The key difference is that we didn't have to program every input and output. Just like algebra class, using a model, allows us to draw a graph much quicker. That’s the difference. That’s what this huge term is about. It’s simply about trying to create a function that accurately predicts what the output should be, based on an input. And then we use it to do work. That’s intelligence, that we’ve artificially created.
Machine Learning
Now that we understand AI, let's dive into Machine Learning. Machine learning is just one technique to develop AI models.
Remember how we said AI uses models, which are essentially functions? Well, Machine Learning is a big fancy word for the process of creating and improving those functions. Making the models more closely predict the correct output. Making their predictions better. It’s looking at billions of bank transactions, and trying to define: how could I have spotted these 100 fraudulent transactions when they first came through? Was it the time that the transactions were sent? The amount? That the sender had never interacted with the recipient before? A combination? Ok, now, prove it. Make a function that given all those inputs, says whether it’s fraudulent or non-fraudulent. That’s machine learning.
Think back to your algebra classes. You had equations like y = mx + b, right? Machine Learning is like having a bunch of points on a graph and trying to find the best line that fits through all those points. But instead of simple lines, we're dealing with data that has perhaps millions of parameters which produces pretty complex functions.