Key Learning Points:

  • GAN (Generative Adversarial Network) is a system where two AIs compete with each other to generate new data.
  • The “generator” creates data that resembles the real thing, while the “discriminator” tries to determine whether it’s real or fake.
  • GAN is a powerful technology, but because it can be misused, how we use it is extremely important.

How Does AI “Draw” Images?

When you hear that “AI draws pictures,” what kind of image comes to mind? Maybe an AI that draws a cat when shown a photo of one. Or perhaps an AI that can create a realistic-looking photo of a person who doesn’t actually exist. Behind this ability to “create something from nothing” lies a technology called GAN.

The name might sound technical, but its mechanism is surprisingly similar to how people interact. There’s a kind of back-and-forth or rivalry at play—something very human-like in nature.

What Is GAN? Two Competing AIs Like a Counterfeiter and an Inspector

GAN stands for “Generative Adversarial Network.” In Japanese, it’s translated as “enemy-based generation network.” This technology involves two AI models with different roles that compete with each other to create new data.

One is called the “generator,” which creates data—like realistic images—from scratch. The other is the “discriminator,” which judges whether that data is real or fake.

These two AIs go through repeated rounds of training together. At first, the generator’s images are rough and easily detected by the discriminator. But over time, as they keep challenging each other, the generator improves and eventually becomes capable of producing images so realistic that even humans have trouble telling them apart.

This relationship is often compared to that between a counterfeiter and an inspector. The counterfeiter (the generator) tries to make fake bills look as real as possible, while the inspector (the discriminator) sharpens their skills to spot fakes. Through this ongoing battle, both sides become more skilled—a helpful way to picture how GAN works.

Where Is GAN Used? Real-World Applications and Cautions

One well-known example of GAN in action is the website called “This Person Does Not Exist.” The photos there look like ordinary portraits, but every face shown was generated by GAN—they don’t belong to any real person.

GANs are also used in restoring old black-and-white photos by adding natural-looking color or sharpening blurry images. Thanks to this technology, even historical photographs displayed in museums can be given new life.

However, this convenience comes with challenges. Because these images look so real, there’s also potential for misuse. For instance, it’s technically possible to create fake videos impersonating celebrities. That’s why it’s becoming increasingly important to think carefully about how we use and manage such technologies.

What GAN Reveals About Human-Like Learning

Just hearing the word “GAN” might make it seem intimidating at first. But when you look inside its workings, you’ll find elements that feel very familiar—like something from everyday human experience. The desire to improve one’s skills or stay one step ahead of others—that drive for growth is something we all recognize.

As generative AI continues to evolve, GAN remains one of its foundational technologies. It brings together two key abilities: the power to create and the power to discern. And it’s precisely this balance—the push and pull between these forces—that leads us toward new forms of value and expression.

In our next article, we’ll shift focus toward how we communicate with AI. To get the image or idea you’re imagining from an AI model, you need clear instructions—called prompts. Let’s explore how those work together next time.

Glossary

GAN: Short for “Generative Adversarial Network.” A system where two AIs compete with each other while generating new data (mainly images).

Generator: An AI model within GAN that creates realistic-looking data (such as images) from scratch.

Discriminator: An AI model within GAN that determines whether data created by the generator is real or fake by comparing it against genuine examples.