...

Lately, artificial intelligence has been talked about literally from every corner. On TV, they talk about smart razors with AI, some people are already convinced that neural networks have even been stuffed into refrigerators, others argue that real AI is something like ChatGPT, and some are downright afraid that AI will soon take over the world.

But let's calmly figure it out: what is this artificial intelligence of yours, and is it really as “artificial” as many like to think?

To put it simply, the field of knowledge responsible for this is called machine learning. So literally: how do you teach a machine (computer) something useful?

And here's an important point: there are a huge number of tasks in machine learning. And far from all of them involve some kind of neural networks, robots from the future, and “intelligent machines.”

A Simple Example

For example, you are selling a car on a well-known blue-and-white website. You post an ad, and the site tells you: “Your price is above market” or “Your price is below market.”

How does it know this? What kind of magic is this?

There’s no magic at all. In fact, there’s a mathematical algorithm behind it. Very often, such tasks are solved using machine learning models, like linear regression.

Roughly speaking, the model tries to predict how much your car should actually cost based on similar cars.

So the site is kind of saying:

“We’ve seen thousands of similar cars. And judging by them, your price looks inflated.”

or conversely

“You’re selling too cheap.”

Another Example

Another example from life — loans, credit cards, or mortgages.

Surely many have at least once wondered: how does a bank decide whether to approve my loan or not?

And here, machine learning is also frequently used. Banks usually apply more complex models, like gradient boosting.

But the idea is roughly the same: take past data and try to predict the outcome for a specific client.

How It Works

And here arises a perfectly logical question:

Why did the bank decide that I can or cannot get a loan?
Why did the site decide that my car is worth exactly that much?

In fact, the general principle is quite simple.

Imagine we have data on a multitude of objects. An object can be a person, a car, an apartment — anything.

Each such object has characteristics. In machine learning, they are usually called features.

For example:

For a bank client, the features might be:

age
income
marital status
number of children
credit history

For a car:

year of manufacture
mileage
color
number of owners
make/model

But besides features, there’s also what we want to learn to predict. This is called the target.

For example:

for a bank client, the target is whether they paid back the loan or not
for a car, the target is its price

What the Model Does

It takes a huge number of such examples from the past and tries to find patterns.

In simple terms, it looks like this:

the model seems to notice that among people who did not repay their loans, certain combinations of characteristics often occurred.

Or it notices that cars with certain parameters usually sold for about this price.

And when you come in with your application to the bank or with your car — the model compares your data with what it has seen before and draws a conclusion.

“Aha, we’ve had similar clients before. There’s a high risk that the loan won’t be repaid.”

“Aha, cars with such characteristics usually sold for N thousand rubles.”

The Main Idea

The algorithm doesn’t “pull things out of thin air.” It works based on real data from the past.

And based on this data, it tries to find dependencies.

Usually, a lot of data is needed — hundreds of thousands or even millions of examples.

Because the more data, the higher the chance that the model will find real patterns, not random coincidences.

Conclusion

Of course, I’ve greatly simplified everything right now. But at a basic level, the essence of machine learning is roughly like this.

In the next posts, we’ll delve deeper into how neural networks work.

Now you’re a bit in the loop that the trendy artificial intelligence isn’t so artificial after all 😉