Eye on Emerging Tech: Generative AI Limited on Creativity as yet

About The Author

Generative AI

Generative AI has come a long way and is poised to steam ahead with exponential growth in computing power.

This is a technique to create original/new content by leveraging existing content like text, photos, videos, images or audio. By utilizing generative AI, computer models try to decipher the underlying pattern signaled by the input with an aim to produce similar ‘new’ content.

One of the techniques of Generative AI is to build Generative Adversarial Networks (GANs). That entails taking two neural networks: a generator and a discriminator and then pitting one against the other to find a balance between the two networks.

The generator network generates new data based on the source data. The ultimate aim of the generator is to produce new data with an intent to fool the discriminator. On the other hand, the discriminator network tries to match/differentiate between the original source and the ‘new’ generated data to find a closer match to the original.

With the generators/discriminators of GANs, transcoders, Reinforced Learning (RL) or variational auto-encoders, it is now possible to deeply fake. Many of these unsupervised learning algos have been applied to images, photos, and videos to much success.

Some good examples of these are here and here.

It is entertaining and frightening at the same time.

If you want to see more, then go to the entire Channel Dedicated to Tom Cruise Fake Videos.

Here is the entertaining part with Tom Cruise’s, deep fake TikTok page, that incidentally may come in handy if you are STILL waiting for Top Gun 2 to hit the theaters.

If you are looking for the scary part, look no further than a host of political pages available worldwide on the internet. They come in all flavors.

My skepticism of Generative AI relates to producing high quality, creative original text from copious amounts of related text input. While it may be possible to generate text brochures, slogans, catchy headings, or keywords from text, when it comes to original creativity, the machines will be subservient to humans.

I will make my case with an excerpt from ‘Return of the Native’ by Thomas Hardy.

Yeobright and Eustacia looked at each other for one instant, as if each had in mind those few moments during which a certain moonlight scene was common to both. With the glance the calm fixity of her features sublimed itself to an expression of refinement and warmth: it was like garish noon rising to the dignity of sunset in a couple of seconds.

‘Thank you; it will hardly be necessary,’ she replied.
‘But if you have no water?’ ‘Well, it is what I call no water,’ she said, blushing, and lifting her long-lashed eyelids as if to lift them were a work requiring consideration.

Now, let’s input the complete works of Thomas Hardy as source data and try to generate some creativity as the above excerpt. My bet is that it will not come even close. Even with the exponentially growing power to compute, the machines are no match for the ~86 billion neurons in the human brain when it comes to creativity.
Well, if/when that scale tilts in the favor of the machines, I will subscribe to that channel.
How far do you think we are from a day when AI can create highly creative and original work?