No, GenAI will not kill human creativity

My favorite photo is the surrealist "Dalí Atomicus", taken by Philippe Halsman. The photograph shows three flying cats and one flying Salvador Dalí, surrounded by an arc of floating water. The photograph is a technical masterpiece, but cannot strictly be called creative. All the photographer did was press the shutter button - anyone can do that! At least that was people's opinion when the camera was first invented. At the time, photography was not considered something one could be creative with, as it only depicted reality, and lacked “something beyond mere mechanism at the bottom of it.”

.In many ways, the camera is a good analogy to generative AI (GenAI). A camera can work as a tool that allows people with creative ideas to express themselves, even if they lack classical artistic skills. Hence the camera enables creativity!

Here in Norway, my favorite author, Erlend Loe, claims that AI will never be able to write something original and good. He is not alone in this conviction. The award-winning author Ray Nayler, claims that a GenAI tool, while able to effortlessly produce predictable content, lack the ability to truly innovate since what it creates is “not an original idea, but a mash-up of our old tropes, repackaged for our consumption”. And Nick Cave, the Australian musician, writer, and actor, famously said that "Writing a good song is not mimicry, or replication, or pastiche, it is the opposite". But while we would love to think that humans have the capacity to create truly novel art, it this the case? As Picasso himself said: “Good artists borrow, great artists steal.”

The main issue in the debate about AI creativity is that the words “original” and “creative” are not well-defined, and the criterion used to judge the “originality” or “creativity” of a piece will differ depending on whom you ask.

Creating “mash-ups” vs. “truly new” things

In the field of machine learning and statistics, we often make a distinction between interpolation (creating “mash-ups”) and extrapolation (creating “truly new” things). The current GenAI models only interpolate, meaning that the content they create is situated inside the space of things they have seen before. A true "paradigm shift"-level of creativity and originality would probably require extrapolation, which the current models cannot reliably do.

Imagine a world where colors have not yet been invented. An interpolating machine would be unable to describe or produce any colors in this world. But let’s then say a human comes along and invents the color “blue” – a truly new and original discovery – a paradigm shift. Our interpolating machine would now be able to use the construct of “blue” for simple things, like manipulating its intensity. It would not, however, be able to invent a new color based on its knowledge of “blue”.

Let’s then say another human invents a new color, “yellow”. This, of course, is also an amazing accomplishment, though no longer a paradigm shift, since the concept of colors had already been established. Still, it constitutes a new dimension within the concept of colors. The interesting part is that with these two human discoveries, an interpolating machine would now be able to discover the color “green” since it is a combination of the two already discovered colors. The critics would, of course, say that “green” is not an original idea, since it is just a derivative mash-up of “blue” and “yellow”.

It should be noted that our interpolating machine would not be able to create the color “red”. This would constitute extrapolation into yet another new dimension, the “red”-dimension. But if these three primary colors were discovered, our interpolating machine would now be able to find all other combinations, and thereby all other colors we know of. This leaves us with three distinct levels of creativity:

1. Paradigm shifts
2. Discovery of new dimensions within a paradigm
3. Discoveries within a space of established boundaries or dimensions.

Who is the true original?

Returning to the issue of what constitutes creative work, allow me to build the following case: imagine that people's creativity followed a Bell curve most people are not highly creative, but maybe 10 percent can produce work that we generally would consider “creative” (the actual numbers are not that relevant for the example).

But how creative are these people really? Aren’t they all influenced by people who came before them, peers who inspire them, the time they live in, and the medium they are practicing in? So, what would true originality look like? Let’s say that just one percent of the total population can create true innovations, something that has never been done before. Would this degrade the work of the remaining top nine percent to such an extent that it can no longer be considered creative? If so, most of the people we currently consider “creative” would also be stripped of this pride.

I’ll give some examples, starting with Andy Warhol?—?a famous visual artist described by many as a creative genius. He is known for paintings like the Campbell Soup Can and Marilyn Diptych (portraits of Marilyn Monroe). An especially interesting piece in our case is the Orange Prince, a portrait of the singer Prince. The painting is one of twelve silkscreen portraits on canvas depicting Prince. While many consider these images to be creative masterpieces, Warhol created them based on an original photo taken by the photographer Lynn Goldsmith. A derivative, in other words, inspired by others' work?—?or simply copyright infringement, as the Supreme Court decided in May 2023

We see the same in music, where sampling of other songs has become widespread. Some examples of recent hits that are based on old classics include:

• M.I.A?—?“Paper Planes”, which samples The Clash?—?“Straight to Hell”
• The Notorious B.I.G.?—?“Mo Money Mo Problems”, which samples Diana Ross?—?“I’m Coming Out”
• Madonna?—?“Hung Up”, which samples ABBA?—?“Gimme Gimme Gimme”.

While sampled songs are obvious examples, derivative art doesn’t have to be this explicit either. Ever wonder why “all pop songs sound the same”? Well, music theory has given us a few chords that for some reason just… work. Axis of Awesome demonstrated this beautifully with their legendary song “Four Chords”, where they played nearly 50 different songs using the same four chords.

But we can go even broader and look at music genres. Most of us recognize "rock" as a distinctive music genre, but rock drew inspiration directly from the rhythm and blues genres of African American music and country music. In other words, rock is also a derivative.

We see the same in other art forms, for example, the film industry, with Quentin Tarantino’s work being a classic example. His movie “Kill Bill” is a tribute to martial arts cinema, spaghetti westerns, and Japanese samurai movies. It contains specific references to films like "Lady Snowblood" and "Fistful of Dollars". Tarantino himself is even quoted as saying “I steal from every single movie ever made. If my work has anything, it’s that I’m taking this from this and that from that and mixing them together.”

Being creative with "mash-ups"

Our current GenAI models are unable to extrapolate. They cannot create ideas at the level of a paradigm shift, like inventing the concept of colors or creating new dimensions, like a new primary color. However, I find it almost elitist to assert that the only way to be creative is to extrapolate. There are plenty of ways to creatively mix existing colors and ideas in an interpolating way: recontextualizing old ideas with new perspectives, combining existing things never combined before in new and interesting ways, for example satire and parody. If we consider artists like those mentioned above to be original and creative, we should apply the same standard when talking about GenAI.

On the other hand, as Ray Nayler points out in his original article, "AI thrives when our need for originality is low and our demand for mediocrity is high". A plausible outcome of this democratization of GenAI is the proliferation of work made with little to no effort. This resembles the rise in photos being generated after the smartphone entered our lives. The fraction of what we would consider creative, high-quality photos appears diluted compared to the time of analog cameras when only a few had access to such expensive equipment. However, this increase in quantity does not necessarily diminish the value or quality of fine art. In fact, with more individuals able to explore their creative ideas, there may be an abundance of unique and high-quality works to discover. It is even worth considering that the availability of inexpensive tools can also contribute to a more diverse and flourishing artistic landscape. So maybe let a thousand flowers bloom?

A plausible outcome of this democratization of GenAI is the proliferation of work made with little to no effort.

Creativity was traditionally seen as a distinctly human trait, encompassing both the conception and execution of ideas. While ordinary GenAI already handles the execution phase, this conventional wisdom is now being further challenged. Colleagues of mine successfully as SINTEF have eliminated the human component in the GenAI image generation process, removing both the conception and execution phases, and prompting a reevaluation of what constitutes creativity.

Moreover, associating creativity with extrapolation?—?extending beyond existing knowledge?—?might not be enough to fend off the machines for long. Google DeepMind ’s FunSearch has, for example, demonstrated capabilities of generating insights that surpass known human understanding, potentially redefining the boundaries of creative thought and human-machine interaction.

As we stand at this technological crossroads, it is apparent that the definition of creativity is not just expanding, but fundamentally transforming. With the emergence of GenAI, more people will be capable of producing works of apparently high quality (though most of it will presumably be shallow). This raises important questions about what we value in creativity and how we assess its quality.

I predict we may soon find ourselves flooded by uninspired pieces of “art”, much like the era of snapshot photography. But in this vast sea, there will also be novelty, creativity, and new ways of expression – for those who care to look. Humans, for better and for worse, are prone to boredom. If everything looks the same, people will crave something new. And while it feels like we are currently stuck in an uninspiring, never-ending era of Marvel movies right now, bear in mind that the "Golden Age of the Western films" lasted for 20 years (!), and we got out of it eventually.

Humans, for better and for worse, are prone to boredom. If everything looks the same, people will crave something new.

Finally, while I hope to have demonstrated that creativity can be achieved through interpolation (the mixing of blue and yellow to create green), there is a far more interesting question we must dare to face: Is it really the interpolation itself we are against, or is it simply the fact that a machine made something and that it made it so effortlessly? Think of it like a Turing test for creativity: Would we still be hesitant to declare a machine “creative” if it could produce true novelty that sparked emotions and challenged assumptions, while still lacking “something beyond mere mechanism at the bottom of it”? Are we brave enough to use such terms for something deprived of intent and a “soul”, or is creativity something we inherently define to reside within the realm of human expression? As we grapple with these questions, rather than pitting machine-generated creativity against human expression, we should instead be asking ourselves if we are truly open to embracing new forms of creativity and artistic expression.

If you're interested in a deeper dive into the concept of originality, there is a great podcast episode from NPR's TED Radio Hour called “What Is Original?”. The episode looks at music, fashion, and innovation, demonstrating how every idea, invention, and song is built on something that came before it. It was first aired in 2014, way before GenAI was a thing. However, it is as relevant as ever, and I highly recommend it.

Also, if you are located in Trondheim and are reading this before June 3rd 2024, consider going to the talk "D?r forfatteren med den kunstige intelligensen?" with Inga Strümke and B?r Stenvik, hosted by Litteraturhuset i Trondheim! (I have no affiliation)

Viggo Tellefsen Wivestad is a data scientist, doing research in the field of Software Engineering for SINTEF in Norway, one of Europe’s largest independent research organizations. This post is a response to the recent article AI and the Rise of Mediocrity in TIME magazine, where Ray Nayler criticizes GenAI (generative artificial intelligence) for fostering mediocrity in creativity.?

?