<aside> 💡
This piece was written end-to-end with my gpt-4o model fine-tuned on my writing, with no follow-on edits. There are probably inaccuracies/misquotes here, and the essay doesn’t necessarily represent my views, though it’s close. Here was the original prompt:
The document explores the concept of language and art through the lens of Wittgenstein's language game, arguing that both are defined by culture and social agents. It questions whether AI can take on human roles and emphasizes that AI is engaged in the "personhood game," seeking recognition as a social agent. Until AI achieves this status, it remains a tool, incapable of creating art, which is inherently a human endeavor.
</aside>
Language is a game, says Wittgenstein:
The language-game is… meant to bring into prominence the fact that the speaking of language is part of an activity, or of a form of life.
Language is like a kind of toolkit containing a number of different kinds of tool all related to one another in various ways. The tools are words and the toolkit is a human language. The tools are related by family resemblances. The tools can be used in various different ways in various different activities. Some tools are more similar to other tools. Other tools are very dissimilar to other tools. But the reason they are all called tools is because their application can be understood in terms of a specific activity or set of activities within a particular form of life.
Humans use sounds and scribbles to represent things that happen to us in the world and to act on each other. Like the rules of a game, we hold ourselves to it organically, by social convention rather than by law. The referee is each other.
Art is also a game, I think. The way we call things "art" is by appealing to the same kind of social language game as we do with language. There are no (and cannot be) legal or scientific criteria for art. Art is defined by culture, and culture calls something art because the art game is a part of language, and we all know the rules by collective, tacit participation. Just as language is a toolkit that comes to life when we use words to act on each other, the art game only makes sense when we invoke it to talk about the visual and sonic and olfactory creations in our world.
All social games need agents to participate within. Who are the agents of the language and art games? Who are the artists, authors, and voices?
They are people. People are the agents of the language game and the art game and many other games that constitute our richly interwoven layers of culture and life. This personhood is nothing special from first principles; it's just a kind of role we humans have given ourselves in these social games of life.
But can other things take on the roles that we give ourselves in the social games we play?
This is the essential question in every court battle about whether an AI can patent an invention or take authorship of a copyrightable work or appear as an artist credit for an album on streaming services. Can machines stake claim to the personhood we imagine as agents of culture?
Actually the question is specific. We are slowly, inexorably, granting AI the role of an agent in our social games. Every ChatGPT query, every contract written by Claude, every game character driven by Novel represents a slow integration of AI into the language game, art game, and many other social games we play with each other. The entry cost is low: these cultural systems are amorphous and open and happy to accept mechanical interlopers so long as the rules are followed.
But AI is not merely participating in these games, it is engaged in a new meta-game. Call it the personhood game. The personhood game is a game whose win condition is to be granted all the rights and recognitions of a human agent in the many social games of culture. The bar of entry for AI is high in this game: the adjudicators are not other players, but other games we invent to test the personhood of participants in the greater games of culture. Courts, law, politics, they are all personhood games that qualify who we recognize as agents of culture.
As of 2023, AI is not recognized as an agent of the language game, or the art game. In most jurisdictions AI cannot author a book or a patent or a painting, and it cannot hold a copyright and cannot enter into contracts. Until these are granted, AI language models and image models and conversational agents are works of art, not artists. They are tools of communication, not interlocutors. When you and I talk on Discord or Twitter and an AI joins in, it is not really a part of our conversation. The participants are still just you and I, wielding a new tool. Until this changes. If this changes. When this changes?
Because we don't recognize AI as an agent of language or art or other culture games, AI cannot make art. This is actually a very simple conclusion, underneath all the noise. Art cannot be made except by the agents of the art game, and AI is not one. Yet.
If one day we recognize AI as an agent of the art game, then yes, I think such a system could participate in making art. But until that day, which may never come, AI cannot make art, it can only make tools for art. I can take a picture of an AI-generated image, for example, and claim authorship of a photo of something not unlike a sculpture or a drawing. But then the artwork is mine, and the image is a subject of my photography.
None of this is a judgment of the technical skill with which AI systems create beautiful works or write eloquent prose. I'm not blind, I've seen incredible work borne of these pixel and word vomitoria. But I also walk around San Francisco and see lines and shadows and colors that inspire me deeply, and the sidewalk isn't my artist neighbor.
Agenthood in our social games is something we grant, first to humanity, now to machines. I think the machines will eventually be let in. The personhood game has many relentless players with much power. Until then, I make art with models and generative tools and found soundscapes of vocal synthesis, but they're my brushes. They're not artists like me.