Deepfakes: GANs and the War on Truth

A computer generates a face that has never existed. How? It uses a GAN (Generative Adversarial Network). It is two AI Brains fighting each other.

The Forger vs. The Detective

1. The Generator (Forger): Tries to draw a fake face.
2. The Discriminator (Detective): Looks at real faces and the fake one. “That looks fake.”
3. The Loop: The Forger gets better. “Okay, I’ll fix the eyes.” The Detective gets stricter. “The hair shadow is wrong.”
4. They fight millions of times until the Forger is so good that the Detective can’t tell the difference.

Robotics: Synthetic Data

We use this for good.

• We don’t have enough photos of the game elements on the field.
• We use a game engine (Unity/Unreal) to generate fake photos of the field.
• We use a GAN to make them look “Photo-Realistic.”
• We train our robot on these deepfake images.
• When the robot hits the real field, it already knows what to do. We use “Lies” to teach the robot the Truth.