Back to Blog
•By Sumit Pandey, PhD•2 min read

ML 13: The Forger (GANs)

Two AIs fighting each other to create fake reality.

The Art of the Con

Generative Adversarial Networks (GANs). "Adversarial" means enemies. It is two Neural Networks fighting a war.

The Forger

The Players

  1. The Generator ( The Forger):

    • Trying to create fake paintings (images) that look real.
    • Starts by scribbling random noise.

  2. The Discriminator (The Detective):

    • Looks at an image and tries to guess: "Is this Real (from the dataset) or Fake (from the Forger)?"

The War

  • Round 1: Generator makes static noise. Discriminator says "FAKE". Generator learns nothing worked.
  • Round 100: Generator makes blob. Discriminator says "FAKE".
  • Round 1000: Generator makes something looking like a face. Discriminator says "Hmm, Real?". Generator learns "Aha! Eyes work!"
  • Round 10000: Generator becomes a master artist. Discriminator has to become a super-expert detective.

They push each other to get better. If the Detective sucks, the Forger creates garbage and gets away with it. If the Detective is too good, the Forger gives up. They need to improve together.

From noise to a pattern.

An illustrated generator and critic. Change the target, make a sample, and compare what improved.

Generated sample

Target: Face

0.0%pixel match · 0/64 cellsRound 0

Critic rule: not there yet — fewer than 85% of pixels match.

This is a simulation, not GAN training. Each round mixes more of the target into random noise; the critic uses a pixel-match rule. Real GANs learn both networks through competing losses.

Explore the full playground →

Deepfakes

This is how Deepfakes are made. This is how "This Person Does Not Exist" works. The Generator learned to make faces so realistic the Detective couldn't tell the difference.

Summary

We are done! We went from drawing a line (ML 1) to creating fake human beings (ML 13). The world of AI is huge, but you now know the core idiots running the show:

  1. The Line (Regression)
  2. The Committee (Neural Nets)
  3. The Tree (Decision Trees)
  4. The Mob (Random Forests)
  5. The Hiker (Gradient Descent)
  6. The Student (Overfitting)
  7. The Eyeball (CNNs)
  8. The Goldfish (RNNs)
  9. The Attention Seeker (Transformers)
  10. The Dog (RL)
  11. The Forger (GANs)

Go forth and build something dumb.

Share this article