Photography, Go, & AI: "The value of creativity itself has diminished"

This year at NeurIPS, there was a workshop on Creativity & Generative AI.

Alongside Jillian Arnold, Kelly McKernan, and Ted Chiang (talk notes here), I was invited to give a talk and share my perspective on generative AI with the machine learning research community.

I chose to speak about AlphaGo’s impact on professional Go, how it connects with what’s happening in creative industries around me, and some musings on life, art, and meaning.

I've posted the full article on my blog, but you can also read the essay below:

Background: Art & Competition

I’ve been a fashion and fine art photographer for nearly 20 years. My youth was shaped by both technical and artistic interests, and I was fascinated by things like micromouse and computer science at school, spent six years on Singapore’s national shooting team, and once founded an esports team in StarCraft 2.

I grew up both loving art and lived in pursuit of an Olympic dream, and because Go represented both artistic beauty and competitive spirit—it was uniquely special for having a bit of both worlds.

The World of Go

I played Go on and off over the years, spent some time living with professional players, and through photography, covered major world tournaments like the Ing Cup.

For those unfamiliar, Go is an ancient strategy board game played with black and white stones. The rules are seemingly simple with all stones being of equal value. But once you start playing, something complex with up to 200+ moves unfolds, each game like an individual story.

I loved the sound of a stone placed on a board, the elegance and simplicity of the game’s rules, and how just like a painting, a player’s style was an expression of who they were, their moves like brush strokes in a painting.

The path to mastery for becoming a pro wasn’t easy. Many left home between 6-7 to study at Go academies in big cities, living away from their families even as children.

I had a friend who at the age of 11 or 12, in order to play in the finals of an important tournament, was not told by the adults when his mother passed away. He never got a chance to say goodbye.

Go is beautiful to me, but it is not without sacrifice. People dedicate their whole lives to the art, and for many years, it was seen as the pinnacle of games to solve in AI. Players and fans alike loved it for its mystery, and how it remained unsolved by machine.

AlphaGo

In 2016, DeepMind created AlphaGo, an AI that defeated Lee Sedol—one of the world’s best Go players.

This was a landmark moment because up till then, most researchers believed that such a feat was still years away—even AlphaGo’s own team thought the same when interviewed in the AlphaGo documentary.

Lee Sedol’s loss was both a marvel of achievement in AI and a devastating loss for the Go world.

In the days after the event, I spent a lot of time staring into space and contemplating what this meant for the future of human creativity and our pursuit of art.

From Go to Art

A year before AlphaGo, I ran into Yann LeCun at a conference and asked him about the possibility of training an AI in the style of my work. I got the gist that it wasn’t a question of if, but when that style-replicating models would appear.

While many professional artists had experimented with GANs by then, most people, just like those in Go, thought that today’s generative AI models would still be a tech of the far future.

After AlphaGo, I began to look at AI development and research through a new lens. I was trying to build an artist platform at the time and ended up attending a business program, going to a bunch of tech and founder conferences, and connecting with people in machine learning to learn more about recommendation systems. They opened my eyes to AI developments happening in industry, and I began to think with some clarity that the timeline for AI in art would come much sooner than imagined.

I had thought perhaps it would take five years, and it took six, when DALL-E 2 arrived in 2022.

At the time I felt like the creator’s AlphaGo moment was here, and I said as much to friends. Because even though it didn’t arrive perfected, you could see that the technology was fundamentally different from what came before.

Photography & Copyright

People get into art for a variety of reasons, and a lot of it is for passion and self-expression. Many of us start off with little next to nothing. In photography, we would often shoot for magazines for very little pay, or sometimes even for free or out of pocket if we lived in competitive cities like New York.

As a freelancer, there were no salaries or benefits. Some years I would skip health insurance because I couldn’t afford it—until the government fined me for not buying insurance. It was a constant seesaw of juggling a million things alone.

But we accept these terms because in return, we owned the copyright to our work. Knowing that we could sell and license the work later is a big part of what made photography a viable career path for me. It’s how I earn my living. The framework of copyright provides me the protection I need for my income.

Two years ago, I found out that a man in Luxembourg took one of my photos, mirrored it for a painting, and entered it into an art biennale. It won an award, was offered for sale for thousands of euros, and when he was found out, he claimed it was his right to use my work because he found it online.

After hearing that he hired a lawyer first, I looked for legal counsel to protect myself. We eventually took things to court when the art biennale said they couldn’t tell if the work was copyright infringement or not. (I won the case)

It felt like we had reached a saturation point where many people thought just like this guy did, and I hoped that by fighting this, I could create an example for smaller artists, both to provide them with a reference, and a story to defend themselves from people who might try to argue the same points.

Most people understand that you can’t take a Game of Thrones poster or picture of Taylor Swift and make money off them. But somehow, selectively, people seem to forget this fact when it comes to names they don’t know. It’s been a constant battle for creatives online to get our rights respected, and as this case unfolded, generative AI began making its way into our world.

After AlphaGo—Life, Art, and Meaning

Go has continued on after AlphaGo. People now use it for learning, practice, and do game reviews with AI tutors. But while there are practical upsides, it is not without cost.

In 2020, a professional Go player took his own life due to depression. He was known to have disliked using AI for practice as it became the norm, and was said to have hated the idea of following moves determined as most optimal by AI.

This loss was not singular. In the entertainment art industry, we lost an artist to similar struggles as AI began taking the stage at various studios before the layoffs started hitting the news.

In a tribute interview with teammates and top players, Fan Tingyu, a 9dan pro said:

“There are times when it’s inevitable to feel like Go has become boring, like it no longer held the colors it once did. … Now, when you spend a lot of time coming up with a new move, even if it looks interesting, all it takes is for AI’s evaluation of your move value being low, and showing you exactly how many ways it can counter you—and you will feel like the value of creativity itself has diminished.”

Lee Sedol’s reflections paint a similar story. In 2019, he announced his retirement because he couldn’t defeat AI, and in an interview with Google this year, he talked about how looking at AI was like reading the answer key:

“The Go I learned was a form of art, where you think by yourself, and through thinking and interaction between two people… But at some point, looking at AI, learning is more like reading the answer key.”

When asked if he would be a professional Go player again if he could go back in time—he said no.

This was someone who gave his life to Go. Someone so impactful that the world of Go shaped itself to his games. For him to say that he wouldn’t be a pro again—it was heartbreaking.

AI didn’t just defeat Lee Sedol in technical ability, it defeated his spirit. It changed his relationship with what he saw as a form of art, and made him regret pursuing the path he had chosen in life.

The fact that humanity has been fully surpassed in this arena—as much as we can learn from it—has not come without cost.

Across the board, people both praise the convenience of being able to play against strong AIs and lament the loss of value in human creativity. Some might argue that perhaps this is just a pain we need to suffer in order to gain a more optimized world. But the impact goes beyond the voluntary retirement from a competitive game in other professions, and I think that perhaps, we should care about the pains currently experienced by other people.

As the race towards more capable AI continues, we need to contend with what it means for humanity as we automate away more and more of art creation and meaning.

When you replace expression and purpose so that we are more optimized by following the best paths determined by AIs—what becomes of the human experiences we cherish?

Art vs Go

Professional Go continues to exist, remaining a human-only spectator sport. But this is not the same case with art.

For artists, not only do we have to compete with AI models trained on our work without consent, not only is it no longer just a technical demo like AlphaGo was—these are entire replacement machines, designed to replace the life and work that artists have built throughout our lives.

So while I can appreciate the technical achievement of how far generative AI has come today—it is painfully, painfully difficult when artists learn that not only will these tools be used to replace jobs en masse, both within and outside our own industries—but also that the data comes from the very people these tools are replacing.

For many of us who chose to do art, it’s not just a simple picture—our work is our identity, a life we have dedicated to a craft and towards our dreams—and it feels like it’s now been fed through a grinder, without any regard for what we feel and what will happen to us.

Taking Action

I created Cara in 2022 so there would exist a platform where people could easily find human artists.

Despite being a volunteer-built and community-supported project, we’ve grown to over a million users in our beta, validating that people feel a real need for human connection in art and digital spaces.

At art conventions, we constantly hear people share that Cara had a significant, meaningful impact on their lives, and they wanted to know if there were more people building tools that could help protect their work. I thought that NeurIPS would be a perfect place to share this question, and that’s why I chose to mention this today.

Before coming to the conference, I wrote a recommendation letter for a PhD applicant. He wrote something I really loved in his statement of purpose:

"I am fortunate that my two-year exploration in AI research gave me the privilege to witness the entire process of how the rise of Generative AI divided the world into two groups: advantaged groups that gain a lot from AI, and disadvantaged groups that undergo a loss out of AI.

Attention comes to the former who masters, manipulates, and mobilizes AI, but always neglects the latter as the superseded and forced recipient of AI. Having witnessed this division, I am determined to invest myself in aligning and auditing Generative AI to prevent its misuse and help disadvantaged groups in this new AI era."

— Chumeng Liang, developer of Mist

I understand the pursuit of capabilities research is exhilarating, but research to support creative communities affected by these developments remain vastly underexplored and that should change.

I’m sharing these stories not to make a case against progress, but to ask for more thoughtful development that considers the human impact of what you are creating. Thank you.

-

Related posts:

• Ted Chiang's Talk at NeurIPS
• Lee Sedol, Art, & Go
• Maybe We Should Care About AI Safety
• Luxembourg Copyright Case Against Jeff Dieschburg
• The Future of AI Art and Automation in Creative Industries