How Does Generative AI Work?

drawn image of calendar icon

May 23, 2024

hand drawn writing icon

Anastassia Lauterbach

How Does Generative AI Work?

I would like you to picture a house with three rooms. Now imagine that you employ an art counterfeit.

This guy is talented, and he’s making paintings in the style of, let’s say, Marc Chagall. But this artist could be better.

So, he sits in one room of the house, and you call this artist “the Generator.”

But there is a second room where you employ another guy, a police officer trained to detect counterfeit art.

He inspects pictures closely and determines whether they are original or came from thieves.

So, you will call this guy “the Discriminator.”

And then you have a third room.

And in this room is a fabulous collection of authentic masterpieces by Marc Chagall.

You only show it to the Discriminator, so he learns the techniques of Marc Chagall.

Then you ask your artist – counterfeiter- to make a bunch of fakes.

So, you can bring one of his fakes to the Discriminator to inspect, or you can get one of the absolute masterpieces he has yet to see.

The Discriminator will try to determine each piece presented to him so he can shout ‘fake’ or ‘real.’

When he’s correct and identifies as fake, you go back to the counterfeiter – to the artist – and bring him the bad news.

And then you ask him to improve.

You do it repeatedly until the policeman – or the Discriminator – is wrong.

It means that he thinks that the fake piece is a real piece of art.

So here comes something significant.

You tell the policeman or Discriminator that he’s wrong and instruct him to improve his detection techniques.

So he might adjust something or his method, and he will become better at detecting the fakes. Then, he starts getting all of them correct again.

So, an internal competitive arms race accelerates until you have an impressive collection of counterfeit Marc Chagall pieces.

So, this technical process creates an algorithm that can’t even tell its own fakes. It presents a paradox of a third-party agency, like a social network administration, being able to easily detect what is right and what is fake.

There are countless examples of Generative AI in art and music. I particularly love the Beethoven Xth Symphony. If you have never heard about this piece of music, go to Spotify or another music platform and click on Beethoven Xth or the Beethoven AI project.

It’s a fabulous piece of music.

But here is an important thing. The team needed three years to create it.

Walter Werzowa, the team’s composer, spent three years first curating the database, including Beethoven’s Music and pieces of Beethoven’s contemporaries. Only then did the team start to train the model with this data.

So, you now hear that you can write like Leo Tolstoy. In that case, you will spend years curating a database of his writings and those literary influencers whom Tolstoy appreciated and even loved during his lifetime.

You might study who influenced Leo in his art, and then you will actually write with him.

Tim Boucher, a science fiction author from Australia, recently used ChatGPT, to bring his writing aspirations to life. He scribbled almost 100 books using this AI tool in less than a year. According to a newspaper, India Today, he earned a significant amount of money in the process.

It depends on what you want to achieve and your taste in art if you want to employ an AI today to work for you. There have always been bad writers before LLMs got democratized, but this amount will scale at a breathtaking pace now. The technology is still in the beginning. It will only replace great creators, and it will help new authors to find their voice.

There is one further issue my readers need to appreciate and learn about Gen. AI.

This technology creates a river of synthetic data.

Sometimes, these synthetics are tremendously important. For example, they can close the gap if we simply don’t possess enough data. At the same time, those synthetics could be harmful. They might change data behaviour and cause deviations where we don’t want them. In the upcoming years, I will specifically look into medical applications utilizing Gen. AI.

Besides, cybercriminals use generative technologies to create fake faces, people, and identities. In this context, Identity Resolution algorithms and further technologies will play a crucial role. I can only hope that they get democratized sooner rather than later. These are Dun & Bradstreet’s and Palanteer’s IPs and are widely utilized at Big Tech companies such as Alphabet, Meta, Amazon, and Tencent. In the upcoming months, I will write a blog on Identity Resolution and bring the topic to my podcast, “AI Snacks With Romy and Roby.”

romy and roby and the secrets of sleep book cover

Book 1

Romy & Roby And the Secrets Of Sleep.


Submit a Comment

Your email address will not be published. Required fields are marked *

Select your currency