The Lightwave

Practical Insights for Skeptics & Users Alike…in (Roughly) Two Minutes or Less

GANs: The Devil and the Angel on Reality’s Shoulders

This week, we talked a little bit about the power and danger of “Deepfakes.”

Here at the end of another week, we’re going to dive into the basics of how Deepfakes are made.

Specifically, we’re going to focus on Generative Adversarial Networks. Also known as GANs.

GANs are a type of artificial intelligence system that can create new, realistic data (like images, videos, or text) by learning from existing data.

They consist of two main parts: a generator that creates fake data and a discriminator that tries to distinguish real data from fake data.

Defining the Key Players

Generator:

This part of the GAN is like an artist trying to create a forgery. The devil on the shoulder, so to speak.

It starts with random noise, which you can think of as a blank canvas with random splatters of paint. The generator then uses its understanding of the target data (e.g., human faces) to gradually refine this noise into something that looks like the real thing.

It's as if the artist is slowly turning those random splatters into a portrait, adjusting and improving with each attempt.

Discriminator

The discriminator acts like an art expert or detective. Its job is to look at both real data (actual human faces) and the fake data produced by the generator, and try to tell which is which. It's constantly analyzing details and patterns to spot the fakes.

Team Generator and Team Discriminator Train Together

The training process is where it gets interesting. The generator and discriminator are locked in a constant battle of wits.

The generator is trying to create fakes so good that they fool the discriminator, while the discriminator is always working to get better at spotting even the tiniest flaws in the generated data.

As this competition continues, both the generator and discriminator become increasingly sophisticated. The generator learns to produce more and more realistic data, while the discriminator becomes better at spotting even the most subtle inconsistencies.

Eventually, the generator becomes so good at creating fake data that even the highly-trained discriminator has trouble telling the difference between real and fake.

The Face of the Unreals

There are numerous applications for this technology. Perhaps the most well known example is creating human faces that don’t actually exist.

The website ThisPersonDoesNotExist.com showcases this capability by generating a new, hyper-realistic face every time you refresh the page.

For example: This person does not exist.

GANs in Action Today

GANs have revolutionized numerous industries with their ability to generate realistic and novel data.

In the entertainment sector, they're crafting lifelike characters and immersive environments, elevating the visual quality of films and video games.

The fashion industry is uses GANs to design avant-garde styles and to forecast upcoming trends.

In healthcare, GANs are proving invaluable by producing synthetic medical images, which enhance AI diagnostic training without compromising patient privacy.

Perhaps most intriguingly, GANs are addressing privacy concerns by generating synthetic data that maintains the statistical integrity of sensitive information without exposing individual details.

Next Time…

Next week, we’ll look more at the implications of GANs, and how (if) we can balance Innovation and Ethics. We’ll also look at ways SMBs can use GANs in their own operations.

See you then.