Generative AI

Language models have been progressing rapidly over the past 4 years, most notably with ChatGPT 1, 2, 3, and in it’s current form 3.5. ChatGPT is believed to be based on a 175 billion parameter neural network trained on multiple data sets and then refined with multiple techniques. Some of the data sets are thought to be based on Common Crawl, Wikipedia, newspapers such as the NYTs, magazines, books, and reddit links. (Many of the datasets related to the content of the Internet have their origins in the crawl created by a non-profit organization called Common Crawl.) ChatGPT is 2 year old model with a limited knowledge of the world and events after 2021 — with an expected ChatGPT4 soon to be released. ChatGPT 3.5 is added more data sets trained on codex, teaching it reading and logic. The system trained on these data sets then were fined tuned by human labeling ­– supervised labeling ranking of the output and then reinforcement learning with the goal of fine tuning the nuances of language style transfer, query of things.

AI technical advancements are propelling us into a brave new world where human creativity will flourish via Generative AI capabilities for text, images, videos and 3D.

Plato said the purpose of humanity is to obtain knowledge. Friedrich Nietzsche had a different take and said it is to obtain power. Ernest Becker thought the purpose is to escape death and Darwin thought it is to propagate our genes. On the other hand, the nihilists said there is no meaning, and Steven Pickard said the meaning is beyond our cognitive capabilities

I would argue that the answer is none of the above. Instead, it is human creativity for innovation to improve the human condition.

Creativity, like intelligence and consciousness, is hard to define. There are varying denfitions, including the one from Steve Jobs Creativity is just connecting things. When you ask creative people how they did something, they feel a little guilty because they didn’t really do it, they just saw something. It seemed obvious to them after a while. That’s because they were able to connect experiences they’ve had and synthesize new things. And the reason they were able to do that was that they’ve had more experiences, or they have thought more about their experiences than other people.

Much of the advancement of AI can be attributed to GANs. GAN stands for generative adversarial networks and are ‘deep neural net architectures comprised of two nets, pitting one against the other (thus the “adversarial”)’. These were introduced by Ian Goodfellow at the University of Montreal in 2014. GANs are important because ‘they can learn to mimic any distribution of data’ including images, music, speech, prose and anything with unique attributes and characteristics. Examples include everything from creating anime characters to posing three dimensional images to creating new backgrounds on videos and movies

In plain English, two different neural networks combine to work together to create different content. Let’s say you have a base photograph and a reference of a Van Gogh painting: the GAN can essentially render your photo in the style of Van Gogh. That’s a simple example, and conceptually the technique can be applied to just about anything.

There are also CANs — creative adversarial networks — which are GANs with the intent of independently creative thinking. In June 2017, Rutgers released a research paper that introduced the concept. AI systems recently created a perfume for the first time, and an art piece painted by AI recently sold for hundreds of thousands of dollars. Since 2015 Associated Press has used AI to assist in generating articles. But we must remember these are not replacements for people; they are tools to amplify and augment our natural abilities.

GANs and CANs have fueled the excitement around Generative AI, such as the ability to convert an image into the style of any artist.

Yet with these recent advancements have raised a number of questions such as: Who gets the credit for the AI generated content: the artist or the AI. Have you heard about the monkey selfie? A lawsuit contended that because a monkey took the picture, the monkey should own the copyright. Ultimately, the courts decided animals don’t have rights. The intent was with the photographer who set up the camera. The monkey just played with the buttons and wasn’t responsible for the creative aspect of the photo. The fundamental principle is the results of AI activity can and should be traced back to its creator, the human — and we must be focused on giving rights to the creators.

There is also consideration on how to ensure the distributed nature of artificial intelligence as not to be centralization and controlled by a small amount of entities. These are just a few examples of the new considerations we must be mindful of.

Ultimately, it’s not what AI can create; rather it’s what humans can create with AI.