AI Foundations
Generative adversarial network
An AI setup where one part makes examples and another learns to distinguish them from real ones.
Example
A generator creates artificial images while a discriminator learns to distinguish them from training images.
Why people use it
It is one way to teach AI to create realistic-looking material.
What you'll hear
“One part creates images while another judges them.”
What this means for you
Assess generated results independently rather than relying on the training setup.
Can you control it?
Developer-only
The people building or running the AI choose this setup. An everyday user generally needs their help to change how this part works.
Common questions
- Does the part checking the examples verify real-world truth?
- No. It learns to distinguish examples, rather than checking every claim against real events.
- Can it produce many nearly identical images?
- Yes. Training can settle on a narrow set of results instead of learning the full variety.
- Does the judging part need to run afterward?
- Usually not. The creating part can often be used on its own after training.