AI writing and the Death of the Author

What implications does generative AI have for the way we think about literature? Roland Barthes argues that text creation is an impersonal process and further argues that the author's person should not at all matter for the interpretation of their text. If that is true, does the author need to be a person at all? Studies show that people are put off by the idea of reading AI-generated creative texts. What does this tell us about why we read, and what opportunities for literature could generative AI bring?

Photo by Andrea De Santis on Unsplash

In his 1967 essay "The Death of the Author", Barthes claims that literary texts should be read separately from authorial authority, person, and intention. To him, the meaning of a text emerges primarily in its interaction with the reader.

This leads me to the question of today’s post. If Barthes claims that the author's person does not matter for interpreting a literary text, then does the author need to be a person at all? Barthes claims that “language knows a ‘subject’, not a ‘person’”. This could be read in the light of AI writing. As long as there is a writing subject in a text, that is enough. That subject does not need to be a person – so could a text-producing algorithm be subject enough?

If Barthes claims that the author's person does not matter for a literary text, then does the author need to be a person at all?

Indeed, Barthes makes further points that could be applied to LLM writing. He states that a “text is a tissue of quotations” drawn from manifold sources. The author's only job is to string them together. LLMs also work by stringing together statistically likely combinations of words, drawing on previously existing texts. It seems that to Barthes, the human writing process is computational. He suggests the emergence of a new conceptual entity of literary production, the 'scriptor', to succeed the author, which is removed from human emotions and draws from a vast set of language materials to produce a text. This description of the scriptor, with his access to a large set of language data and removal from human emotions, could just as well describe an LLM.

A screenshot of my conversation with ChatGPT, www.chat.openai.com

If the way LLMs write is not so far away from an author’s writing process (as Barthes sees it), then there should be no problem with accepting machine-written texts as being of similar value as human-authored texts, right? 

Well, apparently not. In a recent survey conducted by the Future of Fiction, 92% of respondents were opposed to the idea of reading stories written exclusively by algorithms. Besides copyright concerns and the fear of risking writers' jobs, why are we opposed to AI writing? Is it because we immediately recognise these texts as machine-written and find them lacking in quality? No again. Interim results of a 2023 survey by Mark Lawrence found that people were pretty bad at telling apart AI writing and human writing (5 out of 8 sample texts were wrongly or uncertainly identified). Two AI-written texts made it into participants’ overall top three favourite texts from the survey sheet. This shows that we are not only bad at identifying AI writing - when faced with it, we enjoy it. 

So how do the results of these two surveys go together? It seems that we only feel negatively about a piece of AI writing when we are told that an LLM wrote it.

We are not only bad at identifying AI writing - when faced with it, we enjoy it.

If they hold the potential to be enjoyable, why do we not want to read stories written by machines? Theorist Rita Felski argues that literary works offer us identification with characters across perceived “divisions of gender, race, sexuality, class, or even […] species”. Stories are exercises in interhuman connection, and this connection to a character “may bleed into an attachment to [the] author.” Alice on Medium asserts that reading makes us feel connected to the author: “In the modern world where genuine connections seem so hard to find, reading and writing is a way to feel a deep connection with someone”.

A screenshot of my conversation with ChatGPT, www.chat.openai.com

If we know that a text was written by an AI, there is no one fellow human on the other side of the text to which we can connect on an intimate, individual level – only the thousands of authors of its data set that form an unindividual, obscure mass, and that seems to bother us. Perhaps some survey participants voted unfavourably towards AI writing because of this perceived lack of interhuman connection. 

I think the emergence of AI-generated writing is a unique opportunity for us to examine why we read. Is reading, as Barthes and the like claim, a rational, analytical act, or do we need to find a new conception of how we view literature? AI writing is great at conveying and summarising information. But how radically creative can it be if it only draws from previously existing texts? 

When the camera was invented, people claimed that painting as mimesis of reality was at an end. The result was not the death of painting (and its replacement with photography), but a surge in abstractive art, which radically revolutionized our perception of painting and ultimately freed painting from the obligation to be mimetic.

Could AI-generated writing be the camera of literature?