Special Issue: Artificial Creativity

Happy Monday! Today’s newsletter is a little different.

A few weeks ago, I stumbled across a topic that touched all my favourite genres: law, tech, philosophy, the creator economy. After reading and thinking and reading some more, I decided I couldn’t not write about it.

Are machines ‘creators’? How far have we blurred the line between natural and artificial intelligence? Should machine-produced art be protected by copyright?

These questions are messy, profound - and totally beyond my (lack of) expertise. But I’m excited to try and answer them with you.

Earlier this year, one of my favourite internet writers, Packy McCormick, wrote a piece called Power to the Person. In it, Packy explains we’re living in an age of individual influence: the upsurge in tools that were built to capitalize on creativity - TikTok, Twitch, Substack, NFTs, BitClout - has empowered individuals to cut the umbilical cord between income and companies.

The more individuals embrace this ever-expanding creator toolkit, the more they’re able to profit from their passions. In short, “it’s a great time to be a person.”

Hisorically, most technologies that tried to mirror human behaviour fell into the Uncanny Valley - a term used to describe the universal discomfort machines that are kind of human but not quite human cause. According to roboticist Masahiro Mori, machines like Sophia the humanoid creep us out because they have a strange familiarity - or indeed a familiar strangeness.

But practice makes perfect. Now, increasingly sophisticated iterations of AI are climbing out of the Uncanny Valley to create compelling performances that actually pass as human.

In a project called You Can’t Know My Mind, an AI software painted portraits of people in styles that depended on the software’s ‘mood’. This mood hinged on whether reports in the newspapers were positive or negative that day. If the software was fed too many downbeat articles, it told the sitter to go away; it was not in the mood to paint them.

This kind of artificial creativity exists across every field:

To steal a poignant line from Professor Andres Guadamuz: “the rise of the machines is here, but they do not come as conquerers, they come as creators.”

The world, as it currently stands, subscribes to a mostly human-centric copyright theory that protects original work authored by a person.

This anthropocentric approach is why US courts dismissed a copyright claim by a monkey who took a selfie. It’s also why European jurisdictions like Germany and France romanticize copyrighted work as a “personal spiritual creation.”

Where does this leave our brave new class of creators? Between a bot and a hard place.

Machine-generated art stems from such an independent process that it can’t be attributed to any one human being; as algorithms iterate and refine themselves, the humans behind the machine move further and further into the background. And, to put it simply: no human, no copyright.

Professor Andres Guadamuz makes a pithy case for why this is a little dicey: imagine investing millions of dollars into a system that generates music, only to find that the music isn’t protected by law and can be used by anyone in the world without payment. Annoyed is an understatement.

If developers doubt whether artificial creativity qualifies for protection, they would just, simply, not develop. There’s no incentive for them to invest in machines that can write, paint, create games, produce music, and - without this financial incentive - a large segment of the AI field would shrivel up.

At first blush, the solution seems simple. Tear down the distinction between people and things and give copyright protection to the creator-machines. Power to the Robots!

But extending copyright protection to machine-generated work raises even thornier issues than the ones it tries to solve.

Creativity is the basis of copyright. Can a machine ever be as creative as a Rembrandt or a Picasso?

For Harold Cohen - painter and professor at the University of California - the answer is no. Creativity involves a sense of self and, until machines develop that, “they will never be creative in the same sense that humans are.”

I currently work at a startup that uses machine learning to extract data from contracts and, although data extraction isn’t archetypically creative, I’ve learnt the same thing: machines just can’t perform like humans can.

They’re not torturted artists who feel a burning desire to create, or trained lawyers who have the ability to interpret clauses. They’re entirely dependent on human instruction and leadership. The upshot? Machine-generated creations lack a crucial originality.

In the You Can’t Know My Mind project, the software’s ‘mood’ was a mock mood influenced by newspaper data. Sony CSL’s music-making software remixes melodies from an already existing database of songs. Bas Korsten, who had the idea for the Rembrandt project, admitted the project wasn’t intended to create a new Rembrandt.

Without orginiality, these projects are just illusions of creativity. I’m not saying this to take the shine off these machines and make them seem less exciting - they’re incredibly powerful, thrilling examples of AI. But without the ability to replicate a deep emotional and psychological sense of self, it’s hard to posit - financial incentives aside - that artificial creativity has any true creativity worthy of copyright protection.

Say, one day, machines develop a sense of self and are able to be ‘truly’ creative. Cool. Slightly terrifying, but cool. Can we give them copyright protection now?

Not quite. It’s not as simple as banging a gavel and lovingly opening up the world of copyright law to robots everywhere. As Casey and Lemley say in this excellent paper, to regulate technologies, we first need to know what they are.

And here we come to a problem: sorting the bots from the nots is surprisingly difficult.

Laws have done a terrible job of defining robots so far. The ever-improving ability of machines to act intelligently is causing them to blur the boundaries of what, exactly, it means to be human.

As Casey and Lemley say, this convergence isn’t just about the steady march of machines towards more humanlike capabiltiies. Biohacking tech, like Elon Musk’s ‘Neuralink’ or MITs ‘mind-reading’ headsets, smudge this line too. As we slide into the future, “our machines will only continue to become more human, and we humans will continue to become more machine.”

So, if we ever erased our current copyright laws and replaced them with robot-friendly alternatives, what would those laws look like? How would they even begin to define robots, let alone robot-generated work? Questions like this are thorny, prickly speedbumps on the road to copyrighting artificial creativity.

Then there’s the question of who, exactly, the copyright should be attributed to. When it comes to AI, creation is a social, not an individual, phenomenon. There are code authors, multiple developers, end-users - all of whom have some role to play in generating machine-produced works of art.

When I first started researching artifical creativity, my gut instinct was to bestow copyright protection exclusively upon developers. They created the machine, so they should own the work the machine spits out. Right?

Wrong. The best way to explain this, per Professor Guadamuz, is to look at Microsot Word:

So should copyright be bestowed upon the end-user? Wrong again. When it comes to AI, sometimes a user’s contribution to the creative process is as menial as pushing a button.

The point is: there’s no one answer. Who is ‘most’ involved in the process of creating machine-generated work should get the benefit of copyright protection, but uniformly attributing to this to a sole person is impossible.

~

Put everything I’ve said aside. Let’s forget about what it means to be ‘creative’ or how to define robots, and bring it back to basics. Other than giving developers financial incentives, is there really a reason to robotically revolutionize copyright law?

This blog says it better than I ever could:

If not, maybe we’re barking up the wrong tree. If inventivising developers and stimulating innovation is the real concern here, we can do that without bending the entire framework of copyright law. We could use other avenues - like unfair competition and patent law - to achieve the same effect.

Leaving artificial creativity beyond of the scope of copyright law doesn’t mean machines don’t have a part to play in the creative process. Maybe we just have to reframe our perspective on what their role is: that of a creative collaborator, rather than a creator itself.