Back to Basics: AI Edition

By Taryn Engmark

Associate Editor

Embedded Computing Design

November 01, 2023


Back to Basics: AI Edition

Artificial intelligence (AI) is another one of those nebulous concepts that people throw around pretty often as an overarching umbrella term for a lot of things that, technically speaking, are not quite AI.

Welcome to Back to Basics, a series where we’re going to be reviewing basic engineering concepts that may require a more complex explanation than a quick Google search could provide.

I’m sure you’ve heard of machine learning, deep learning, and neural networks before — all terms generally used interchangeably with AI, but there are slight differences that distinguish them, both from each other and from the AI umbrella.

To Think, or Not to Think, That is the Question

Let’s first look at the highest level of our AI hierarchy: artificial intelligence. Basically, artificial intelligence is any computer system that can do one of two things: appear to think like a human, or actually think rationally.

The difference between appearing to think like a human versus thinking rationally is very important, and unfortunately, also very philosophical. Currently, any AI that we use only mimics the behavior of human thought, but it does not actually “think” (more on that later). The question that arises is whether or not a computer that perfectly mimics human thought is actually rational. Tricky, right?

We can split AI into two subcategories: weak AI and strong AI. Weak AI is what we commonly use today, and it’s designed for a specific, narrow purpose. For example, DALL-E, Siri, or ChatGPT are all designed to do one thing, whether that be creating art, acting as a digital assistant, or cheating on your English papers.

Strong AI can actually think for itself, as a fully complete rational being. For now, this type of AI only exists in fiction. Any intelligent robot you’ve seen in Sci-Fi books or movies probably falls under this category.

Neural Nets: If you teach a machine to fish…

Machine learning (ML) is a discipline that exists under the umbrella of AI. While AI is all about mimicking thought, ML is the part of AI that focuses specifically on data collection and predictive behavior. Normally, machine learning is accomplished by using neural networks.

Neural networks, or neural nets, are a specific “shape” of algorithm that mimic human thought. They are made up of lots of nodes of code that can do a small set of things: take in an input, set the importance of the input, and if the importance is greater than a set threshold, give an output. Each of these code nodes can talk with a whole slew of other nodes arranged in layers, giving the functionality of a code “brain” — hence the name “neural” net.

Essentially, a neural net is really good at taking a complicated set of data, like a picture, and learning what correlations of that data lead to what the algorithm is told is a “good result.”

When looking at a picture, it is broken up into individual pixels. Each pixel has two pieces of data: its location in the picture and its color. A neural net takes each pixel and then correlates it to data of other pictures that it has been told fall in the same category.

When you ask a neural net to generate art, it doesn’t think about what art is, or even what it looks like. Instead, it simply correlates a slew of pixels to different ideas of what “correct” is.

Will Artificial Overlords Rule Us All?

When looking at our AI today, it’s a little easier to understand that it only mimics human thought patterns. We laugh when it spits out some random gobbledygook with twisted spaghetti fingers or nonsensical prose, because we never would have associated the result with the prompt. However, to the AI, this result makes perfect sense, because it has no idea what the picture or sentence really means. It only knows that when it does similar actions, the result is often used or accepted.

As we move into a time where weak AI is a commonly used tool, where does that leave us? It can be unsettling to see AI creating realistic art, writing compelling words, and becoming increasingly adept at assisting us in our day-to-day lives.

But, you can take comfort in the fact that AI currently cannot replace one thing: creativity. Neural nets only synthesize what they’ve seen from data patterns that already exist. When you do something unique, never-before-seen, you’ve done something that AI cannot do. As we learn to live in an AI-assisted world, creativity will be more and more valuable.


Check out our Back to Basics series on coding fundamentals!