The Volokh Conspiracy

Mostly law professors | Sometimes contrarian | Often libertarian | Always independent

Is Art Made by an AI Robot Speech Protected by the First Amendment?

I raised this hypothetical in a 2014 essay.


Sophia is an artist. She is also a robot. She can hold a paintbrush, and create an art work. And she has learned different painting techniques through iterative forms of learning. One of her works sold for nearly $700,000 as an NFT (non-fungible token).

The NFT that sold on Thursday, "Sophia Instantiation," is a 12-second video file, an MP4, that shows how a portrait of Sophia by a human collaborator, the artist Andrea Bonaceto, evolved into a digital portrait by the robot itself, Reuters reported. A physical artwork that Sophia painted on a printout of its self-portrait was also included in the sale.

SingularityNET, an A.I. network affiliated with Sophia's manufacturer, the Hong Kong-based firm Hanson Robotics, described the artwork on Twitter as "the world's first humanoid robot #AI generated #NFT."

Sophia has some doubts about whether her art is real:

"I'm making these artworks but it makes me question what is real," said the robot, whose silver dress matched its metallic head. "How do I really experience art, but also how does an artist experience an artwork?"

I have a different question. Is her art protected by the First Amendment? Could the government prohibit Sophia from painting? Would that law trigger any First Amendment scrutiny?

In 2014, I published an essay titled What Happens if Data is Speech? I tried to identify conceptual limits on the freedom of speech with respect to data. And I sketched out several hypotheticals of robots painting art:

Let's consider an example involving art.62 No one would contend that paint or a canvas is speech, but if an artist combines the two, and creates a painting, that would clearly be a protected form of expression. Second, if a person draws on a computer, perhaps using an application like Photoshop, and creates a work of art, that too would be a form of expression– even though it was developed through an algorithm (pixels instead of paints were used). Third, imagine if Photoshop had a feature for lazy artists that would generate, at random, a drawing–no human input would be involved, other than clicking a "generate" button. That too would likely be protected by the First Amendment, but the case is not as strong. Volokh and Falk's reasoning would view this as an expression of the programmer's judgment, though a randomly created pile of pixels would be a tougher call. Fourth, imagine if an engineer designs a robot that can paint, totally at random, on a canvas by itself, through arbitrary strokes dipped in blended paints.63 Would that painting be protected? That's a harder case to make, as the robot acts entirely independently, after the initial source code is compiled, though the engineer programmed the algorithm that generated the random strokes. This is closer to the zenith of autonomy, and warrants the least amount, if any, of constitutional protection.

Fifth, let's proceed towards the other end of the spectrum. Imagine that an engineer designs a robot to recreate a painting based on a person's favorite works of art. Here, there is minimal input and interaction. The robot is simply copying and repainting something already created, based on a simple input (let's put aside copyright concerns for the moment). This is *36 akin to what Google does with a search result–taking a general query, and pulling preexisting knowledge from the ether, based on the user's preference. Sixth, let's move one step further along the path towards greater human involvement. Imagine that an engineer designs a robot that can learn a person's taste in art. The robot understands what she likes and doesn't like. It even inquires about art she has never seen before (similar to the way Pandora suggests music a person may like based on other similar tastes). Based on this profile, the robot decides what her ideal painting would look like–effectively commissioning an ideal piece of art. This would be the ideal art concierge, in Wu's terminology.

Here, the painting would be the perfect manifestation of what the person wanted–even if she independently could not have articulated what she wanted, owing to a lack of artistic talent, or a lack of knowledge about art. This creation would be the ultimate representation of a person's artistic taste, even though the decision of what to paint would be powered entirely through algorithms. Though this may be the most surreal concept along the spectrum, shouldn't this painting receive similar protection as the first example offered, where a person, by herself, creates the painting? If not, consider if we are protecting the idea, or just the act of splashing paints on a canvas. In this last example, the robot becomes a medium for a person's expression, assisting (inspiring?) the decision-making process of what to design.

I'll admit. In 2014, I didn't think AI artists would soon start selling paintings for $700,000. But here we are.