How to Think About How AI will Change the Arts?

A lot of the stories about AI and art right now are about legal issues — artists and content companies suing Big AI for copyright infringement — or reporters prompting AI models to create essays, novels, poetry or images and then explaining why AI will never be able to truly compete with human artists. Both stories, in my opinion, are distractions. (current copyright law never anticipated how AI trains on existing creative work, and judging the potential of AI based on current models is looking backward rather than forward).

But yes, AI will change everything, because it will alter how we think about what creativity is, what ownership of creativity means, how and who can be creative, what an artist is, and how we share and interact with art and artists. Just as significantly, the business of creativity — the models and marketplaces that support creative production — will be disrupted.

But how to illustrate? Over the next few months, I hope to give some examples of how my thinking has evolved about it. At the moment, “how to think about it” may be the most important place to start. It’s easy to be distracted playing with ChatGPT and being dazzled by stories of potential but having no idea what it might do for, say, a symphony orchestra.

So an example. Last month at The New York Times‘ Dealbook conference, Jensen Huang, founder and CEO of Nvidia, the world’s leading maker of the graphic GPUs used in most AI computers, said that beginning several product generations ago, AI had replaced humans in designing the architectures used for Nvidia devices. Was this a difficult transition, he was asked? Yes. He had noticed that while human engineers made steady but incremental improvements from generation to generation, when the company tried using AI to design architecture, it broke all the design rules humans had played by and created shortcuts that far outperformed human designs. The difficulty for him, he said, was conceptual, in letting go of the things he knew about GPU design and trust the AI as it created and built on its own language which he didn’t necessarily understand himself. Today, he said, there’s not a human that could compete with AI in designing Nvidia architectures. The same is happening in software coding and in other industries.

So what might be such a conceptual shift for an orchestra? It’s not in having an AI write new symphonies. But for humans, composing a piece for an orchestra and getting a successful performance on a stage requires a dizzying array of technical skills. A composer needs to know how to translate the sounds she hears in her head into notation, has to know how every instrument works and what it’s capable of, how the sounds of instruments blend together and how to orchestrate them. Then you need dozens of musicians each skilled at their instruments. And a conductor who can not only read the composer’s score, but also translate it into an interpretation, effectively communicate that to players and then, in a performance, hope that everything comes together.

How many people are there in the world capable of marshalling the skills and opportunity necessary to go from head to stage to recording? A few hundred? Ten thousand? A hundred thousand? But how many people might be hearing music in their head but not have the traditional skills (or opportunity) to get them created? A million? Ten million? A hundred million?

Many popular musicians can’t read music and who cares? It’s the music they make that counts. But perhaps with AI, someone who hears a symphony, instead of having to learn how to write it down, describes it in words to the AI (as one would do now with ChatGPT for text or Dall-E or MidJourney for images). Someone who got very good at prompts could create not only a sophisticated symphony (and print scores for human musicians to play) but also create the composer’s ideal performance/recording of the piece. Maybe the new work might have instruments performing notes or passages that aren’t possible with human musicians on traditional instruments. Perhaps such “extended” instrumentation might evolve the musical language in ways physical orchestras might work to adapt. In the historical evolution of orchestras, major shifts in style and language were often driven by new technology.

So let a million symphonies bloom. If history holds, most of them will be crap. But bringing the ability to compose to vastly more people would inevitably stir things up. It would exponentially expand the number of people who have a creative relationship with orchestras. And it will influence what we value and what we listen for. It would inevitably bring more people from outside the Western European tradition into creating in the orchestra idiom, expanding and diversifying the language and revitalizing the artform.