Greg Chaffins Part of the NerdBeach Series “Artificial Intelligence Explained: What It Is, How It Works, and Where It’s Going”
As futuristic as it sounded at one point, Artificial Intelligence (AI) has now become… normal. It now mundanely but efficiently:
- writes emails
- summarized reviews
- recommends movies
- answers questions
- Helps you decide what to eat, watch, and occasionally even what to think.
And we now rely on it for many facets of everyday life. But have you ever stopped and asked yourself how it actually works? I’ve got news for you – those once popular images of AI consisting of weird glowing blue light (always blue, not sure why) and self awareness is just not the case. No, it’s not Skynet (that’s more like AGI.)
But the real answer as to what AI is exactly is no less impressive on its own merits. Especially considering the impact it is having on modern civilization.
First: AI Is Not a Brain (It’s More Like a Super-Powered Pattern Machine)
We need to get something out of the way first. AI is not sentient. It doesn’t think like a human (unlike Artificial General Intelligence (AGI), and it doesn’t exist yet.) AI does not:
- Have thoughts
- Have awareness
- Understand things the way you do
Instead, AI is really good at one thing: Finding patterns in data. That’s it. Everything else – writing, talking, recommending, predicting – comes from that one ability.
Step 1: Feed It a Ridiculous Amount of Data
AI systems are designed to learn from assimilating (like the Borg, but as dramatic) data – a lot of data. And that data can come in a wide variety of form and format. AI is trained on different types including:
- Text
- Images
- Videos
- Code
- Anything digital (including human fear, if properly digitized. But I digress.)
The idea is actually a simple one: If you present an AI system with enough examples, it can start recognizing the patterns in them. For example:
- Show it millions of sentences → it learns how language works
- Show it millions of images → it learns how objects look
- Show it millions of behaviors → it learns what people tend to do
AI does not take this information and just record it. Instead it uses the data to learn the relationship between things, how they relate.
Step 2: Training (a.k.a. Controlled Trial and Error)
Once the AI has received all the data, it goes through a process called training. This is basically:
- Guessing
- Being wrong
- Adjusting
- Rinse and repeat – a lot. We’re talking a lot of iterations to fine tune the guess.
Think of it like teaching someone blindfolded how to shoot a basketball.
- First shot, they miss. (They are blindfolded, after all.)
- You tell them how far away they are from the basket, and in which direction.
- They shoot again. Go back to step 2 and repeat.
In theory they would eventually be able to swish the ball through the net every time. Assuming, of course, the shooter never tires of the game and their muscles never get too tired to shoot properly. Well, AI is that never tiring shooter, and it will do thousands of attempts in order to fine tune its blindfolded shooting prowess.
Step 3: Neural Networks (The Fancy Middle Layer)
Here’s where things sound complicated, but we’ll keep it simple for now. Most modern AI incorporates a neural network in order to operate. And no, despite the name it is not any kind of brain.
It’s more like a system of layers, with each layer looking at the data and trying to find a pattern. It notes if a given pattern is found, and then passes it on to the next layer beneath it. The pattern seeking gets more and more specialized as the data passes through its depth. The end result could be:
- a sentence
- a prediction
- an answer
- even a pixel
While the output is quite useful, the process is still basically a rather complex chain of “if this, then maybe that.”
Step 4: Prediction (This Is the Whole Game)
This is where things really come together to make AI a practical tool. AI is basically a prediction engine, always predicting what comes next. When it writes a sentence, it predicts what the next word is the most likely to come. Once that word is written, it predicts the next (based on what it’s written.) and so on.
AI uses its prediction engine to function.
- When AI suggests a video for you, it’s predicting which one you will be most likely to click on.
- When it recognizes (through its layer of pattern recognition) an image, it looks at the closest match.
AI is not thinking about a given image or video, it’s simply making some very educated guesses based on pattern matching.
Why It Feels So Smart
You may ask why AI feels so smart if it’s just predicting things. Well, that’s because its predictions are really good. When you combine:
- Massive amounts of data
- Advanced training
- Fast processing
You get outputs that feel:
- Natural
- Relevant
- Sometimes even insightful
Your brain looks at the result and decides that it is the result of intelligence. No doubt your brain is making that decision based on its own functions, but it’s actually just the result of High-quality pattern matching.
Where AI Gets It Wrong
Here’s an important distinction to make. Remember, AI doesn’t actually understand anything. So it goes through its process and spits out a prediction to the best of its training. But it can easily:
- Be confidently wrong
- Misinterpret context
- Miss obvious details
When this happens, it is called a “hallucination.” I guess you could say that AI was just making up something that wasn’t there, but it did so very confidently. Since it doesn’t actually reason, AI doesn’t even know it’s wrong. And that is a very big limitation.
Remember our basketball example earlier? AI could be convinced that it nailed the shot, even though the ball went into the stands instead.
What AI Is Really Good At
Despite its limitations, AI can be incredibly useful in many applications. It really excels at such things as:
- Processing large amounts of information
- Finding patterns humans might miss
- Automating repetitive tasks
- Generating ideas or content (as you might expect a prediction engine to be)
It’s like having a super-fast assistant that never gets tired and knows a lot, but sometimes asks something basic and questionable.
What AI Is Not (Let’s Kill the Sci-Fi Myths)
No, AI is not going to take over the world. It’s simply not capable. After all:
- It’s not self-aware
- It’s not conscious
- It’s not plotting anything
- It doesn’t have intentions
It’s not sitting there and contemplating world domination. Instead it’s just quietly going about its business doing a lot of math – very quickly. Boring in comparison, isn’t it?
Patterns On Repeat
AI is often presented as something complicated and advanced, at least in the marketing brochures. But at its core it’s built on a simple concept:
- Learn patterns
- make predictions
- improve over time
Everything else is just layers of complexity built on top of that simplicity. You can say that it stacks well.
Impressive, But Not Magical
As you can see, AI isn’t magic and it’s not a digital mind. It’s simply a tool that is very good at its job – turning data into predictions.
And when those predictions are good enough… They do start to feel to us like intelligence. It’s easy to imagine that there is a lot more going on than there really is. But once you understand what’s really going on, it becomes a lot less sci-fi, but perhaps a lot more fascinating.
This article is part of the NerdBeach series: Artificial Intelligence Explained: What It Is, How It Works, and Where It’s Going
