Creativity is an illusion. I’ve covered that in the past.
Yet, there’s still a barrier for many people…
How can computers create something new?
As the argument goes…
Humans build every computer. We train them on our creative output.
We feed large language models (LLMs) our words. That creates a limit, right?
Sure, there’s a limit of text for training. But we’re pushing that limit to new highs. And we’re seeing a faster feedback loop…
As the world’s population marches forward, literacy rates have climbed. And the internet, along with low cost computers and phones, has spurred more words. On top of that, auto transcription from audio adds to the mix.
LLMs have limits today… however, with the right models, computers can find patterns to create original content. And already, we’re seeing combinations of words humans have never seen before. Is that not creative?
Going one step further, these new words are funneling back into training new AI models. The feedback loop is picking up speed. And throw RLHF into the mix.
But that’s just scratching the surface.
AI with Humans as a Guide
We’re still in early stages but a huge change is underway.
Multi-sensory models will match human creativity. Or with a more optimistic lens, upgrade human potential.
Sight, sound, smell, touch, taste…
This is how we understand our world (the map is not the territory). It’s how we make memories and change our environment. And with new AI models, we’re building in more human senses.
With humans as a guide, machines are becoming more creative. However, emulating humans is far from the finish line. Both you and I are limited in scope.
Outpacing Human Creativity
We’re organic machines. And intelligence is relative. For example, a dog is more intelligent with smell. And for sight, we only see a small slice of the electromagnetic spectrum.
… a corollary to the curse of knowledge, it’s easy to see our senses are limited.
We’ll continue adding and expanding the senses in the models we create (and the models our models create). We’re allowing for and prompting more variability (an essence of creativity?).
Looking forward, there’s plenty of room for exploratory learning. That comes with greater breadth and depth. And as always, there are many challenges to overcome.
There are data and sensory limits today. On top of that, computational costs tend to climb in a non-linear way. However, we’re increasing total compute capacity and getting better with compression. And more breakthroughs are on the horizon.
But I digress…
With technologies today, we guide innovation… but in a not-too-distant future, we could be pushed out of the loop.
In a narrow view, I admit AI creativity is limited by humans. We’re building and enabling its future. However, as we continue to improve the models, they’ll far surpass our agency.
We have an opaque bottle in our hands. And once we let the genie out, there’s no turning back.
It’s symbiotic today… but will this change?