Generative Adversarial Networks — Explain Like I'm 5

How two AI networks competing against each other created the technology behind deepfakes, AI art, and synthetic data — the forger vs. the detective.

The Forger and the Detective

Imagine a very talented art forger and a very experienced art detective locked in an ongoing competition.

The forger starts out terrible — their fake paintings are obviously wrong, wrong colors, wrong style. The detective immediately spots them all.

But here’s the twist: every time the detective catches a forgery, the forger studies what gave them away and gets slightly better. And every time the forger gets better, the detective is forced to develop sharper skills to keep up.

After months of this back-and-forth, the forger has become so good that the detective genuinely can’t tell the difference between a real painting and a fake.

That’s basically how a GAN (Generative Adversarial Network) works.

Two Networks, One Goal

A GAN has two neural networks:

  • The Generator: The forger. It starts with random noise and tries to generate realistic-looking images (or audio, or text).
  • The Discriminator: The detective. It examines images and tries to decide: “is this real or fake?”

They train together, constantly competing. The generator tries to fool the discriminator. The discriminator tries not to be fooled. Each time one improves, it forces the other to improve too.

After enough rounds, the generator can create images so realistic that the discriminator — and often humans — can’t tell them apart from real ones.

What Came Out of This

Ian Goodfellow (then at Google) invented GANs in 2014, reportedly writing the first working version in a single night after a debate at a Montreal bar. Within a few years, GANs could:

  • Generate photorealistic human faces that don’t exist (ThisPersonDoesNotExist.com)
  • Age a photo of your face or swap hairstyles
  • Turn a sketch into a photo
  • Create deepfake videos
  • Generate synthetic training data for other AI systems

The Catch

GANs are notoriously tricky to train. Sometimes the generator collapses to making just one type of output (called “mode collapse” — like a forger who only ever copies one painting). Getting the balance right between the two networks is an art form.

Newer approaches like diffusion models have largely taken over for image generation tasks, but GANs pioneered what was possible.

One thing to remember: GANs work through competition — two networks trying to outsmart each other — and that adversarial pressure is what drives them to get remarkably good.

deep-learninggansgenerative-aicomputer-visionsynthetic-data

See Also

  • Activation Functions Why neural networks need these tiny mathematical functions — and how ReLU's simplicity accidentally made deep learning possible.
  • Attention Mechanism The trick that made ChatGPT possible — how AI learned to focus on what actually matters instead of reading everything equally.
  • Batch Normalization The 2015 trick that let researchers train much deeper neural networks — why keeping numbers in the right range makes AI learn 10x faster.
  • Convolutional Neural Networks How AI learned to see — the surprisingly simple idea behind face recognition, self-driving cars, and medical imaging.
  • Dropout Regularization How randomly switching off neurons during training makes AI models that generalize better — the counterintuitive trick that stopped neural networks from memorizing everything.

Related Topics

ab-testing How tech companies run thousands of experiments at once to improve their products — the scientific method applied to everything from button colors to recommendation algorithms. 3 levels → activation-functions Why neural networks need these tiny mathematical functions — and how ReLU's simplicity accidentally made deep learning possible. 3 levels → ai-agents-architecture How AI systems go from answering questions to actually doing things — the design patterns that turn language models into autonomous agents that browse, code, and plan. 3 levels →