Spool Five

The Undeath of the Author

Here lies One Whose Name was writ in Water

In 1967, Roland Barthes published his influential essay “The Death of the Author”. The essay was a key reference point in the structuralist and post-structuralist movements and called for the abandoning of the idea of the ‘author’ as the focal point for meaning when interpreting a text. According to structuralist theories of language, meaning arises through a variety of factors (grammar, culture, history, etc.). Seeing the meaning of the work as the product of a single mind (the ‘author’) misses this diversity of perspectives at the heart of language.

Structuralism was relatively short-lived, and the ‘death of the author’ never came to pass. We still rush to bookstores to buy the newest “J.K. Rowling” book, eager to explore the universe that only she can weave. We watch Youtube videos of interviews with authors, actors, and directors, hoping to gain a deeper insight into our favourite cultural works. Rarely, outside of academia, do readers or viewers conduct a structural analysis of Harry Potter books to trace the influences of Greek mythology, or contemporary post-Colonial politics on the work. These kinds of critical analyses do take place, but the public sphere debate is more likely to focus on J.K. Rowling’s pronouncement that this or that character is gay or not. Today, the author still often has the final say when it comes to revealing meaning in a text.

With the recent prominence of Large Language Models (LLMs), the question of the author is once again front-and-centre in the discussion of the value of a work. Only this time the question of ‘value’ is not some kind of academic or theoretical notion of ‘meaning’ or ‘semantics’. Instead, the focus is on the cold, hard value of money. Namely, if an author/artist produces a distinctive work, are they entitled to ownership and authority over this work, including over royalties?

This question is the subject of a number of cases taken against companies with products like Stable Diffusion, whose models have been trained on the work of real artists, writers, and creators. The models are so good that you can often simply request, for example, an image in the style of a particular artist, and the result will be quite close to what the artist themselves would have produced. There is a threat that we no longer need an actual author to produce the kinds of works we expect from authors and creators.

Was Barthes correct then that the meaning of the text is not located in the ‘authors’ mind, but rather is woven into the text, through language, available for anyone to mine and dissect? It certainly seems that LLMs were able to bypass any old notions that there was something quasi-mystical about the act of creation and the mind of the artist. Since the raw material for these models was simply text (and images), perhaps Derrida was right in that there is no ‘outside text’, no sacred authority who sets the meaning of a work. All that was needed could be found within the text itself, and with enough computational power it could be extracted and replicated.

The moment when written text could first be mass produced (the invention of the printing press) is often celebrated as a key moment of emancipation. Prior to people owning their own copies of the bible and learning to read it they had to depend on the interpretation of their priest. Once they owned their own copy, they now had their own authority over the meaning of the work. With digital computers and the internet, it also became possible for people to easily record and publish their own thoughts. Everyone became their own authority, their own creator, their own author.

Looked at in a certain way, this is a progressive movement from a situation where authority and power is centralised with those who were literate and had access to materials for writing (royalty, the wealthy , priests, etc.) to a situation where the masses have their own ability to read, interpret, and create.

However, as we know, it is not as simple as this. A key ambiguity of the situation is exposed with the lawsuits against LLMs. Yes, a visual artist today has access to a range of powerful tools for creation, and for disseminating and profiting off this creation. But, what happens when this very network of dissemination and democratisation is co-opted by a more powerful tool, which can take this data and use it to ‘imitate’ the work of the artists?

On the one hand, digital technologies have delivered on the ideals of structuralism by demystifying the notion of the ‘author’. Now, anyone can be an ‘author’ (you don’t have to be ‘born into it’), and furthermore, the thing that capitalists and markets have profited off for years (the ’name’ of the author, their style, etc.) has also been demystified through LLMs and their power to easily reproduce style. ‘Authorship’ is no longer solely located in the mind of author themselves, but is embedded in the semantics and patterns of the work they produce. On the other hand, real, living authors and creators are suffering because of this, especially those who are independent and live off their ’name’.

Arguably, the most ‘interesting’ thing (for now) about LLMs is precisely their ability to reproduce style and to appear human. This is both a source of amusement and wonderment for people, but also a source of fear and terror for those who depend on the uniqueness of their style for a living. It is ironic because, even as LLMs erode away at the notion of ‘authorship’, we are endlessly fascinated by it - “Write my email in the style of a Shakespearean sonnet…”

We continue to cling to notions of individuality and style, even as they are deconstructed before our eyes. In this sense, LLMs do not represent a ‘death of the author’, but rather a kind of ‘undeath of the author’, where the ‘author’ has, in a sense, been demystified and killed, but continues to live on in a kind of zombie state as we revel in our computer’s ability to imitate and replicate the things we previously took as all too ‘human’. The clearest representation of this is perhaps in science fiction where we see AIs ‘reanimate’ someone after they have died, by analysing their social media posts, etc. The person who has been ‘brought back to life’ is usually represented as something strange and foreign, more of a zombie than a human.

Style is not an end in itself

We could think of the opposite of ‘undeath’ as ‘rebirth’. How is it that we can be reborn, reinvent our identity in an age when our identity can easily be encoded by a language model?

I think an important lesson from Barthes is that the ‘voice’ or ‘style’ of the creator, as revealed through a work, is not equivalent to the creator themselves. Our own voices and tastes are far more diverse and difficult to pin down than what is revealed in the work. As the philosopher Paul Ricoeur says, the act of writing (of creation) is one of ‘distanciation’. From that perspective it is absurd to think that a person could ever be fully represented by an analysis of their writing, paitings or social media posts.

One of my favourite artists to think about when thinking about the role of style in the work of art is Philip Guston. Guston first became known for helping develop a highly influential style of art - abstract expressionism.

At a certain point, however, Guston abandoned this and returned to representational works, painting ‘cartoonish’ pictures which sometimes contained - notoriously - clansman-like figures. At the time, many people in the artworld couldn’t understand the shift. As Michael Auping, a curator who organized a Guston retrospective at the Museum of Modern Art Fort Worth in Texas, puts it, “People whispered behind his back: He’s out of his mind, and this isn’t art. He could have ruined his reputation, and some people said he did.”1

Yet, these new works, while not as easily ‘marketable’ as his experiments with the abstract expressionist style, would become his masterpieces, and are among the most puzzling and thought-provoking experiments of 20th centaury modern art.

This story is not unique. Across all types of art, you can find countless examples of artists who, no sooner than happening upon a defining style, continue to work to experiment with it and even break it all together. Art, by its nature, is both creative and destructive, and the same if often true for the artist’s own identity. The artists that clings too tightly to their own public persona often fails to grow and develop.

The point here is that capitalism (and an LLM) thrives on crude repetition. To fully be ‘human’, in a way that also resists technological determinism, is to embrace the creative adaptivity of the human spirit. Barthes’ lesson, as Derrida would later show, was not really about the author per se. It was about the error of placing any one thing at the “centre” of the text, the error of thinking that there is only ever one ’true’ way of understanding a work. Language (and text) is a dynamic, evolving medium. The same is true of our identities (if we accept the idea that much of our self is constructed through language). There is no ‘centre’ of our identity, no single voice or style that could be copyrighted or fully represented by any kind of static model, be it a LLM or otherwise. There is always an excess, a remainder that seeks real novelty and creativity.

Sun Oct 8, 2023 - 1614 Words

Tags: philosophy LLM