Computational Literary Theory

[レオタルドのフェティッシュ]

As the LLM technology continues to improve, I've been thinking about how this will revolutionise literature. Literature, as we know it, is a complex feat, a consequence of the ever-increasing capacity of the human mind. Fiction, in particular, is known to be able of carrying deep meaning and implications, or as critics would like to call it, multi-dimensionality. As we fed a gargantuan database of knowledge, especially of literary tropes and styles into their training data, one may become able to create even more sophisticated and nuanced works of fiction however they wish.

You can dictate the model to generate intricate plots, develop complex characters, and build immersive settings, crafting compelling storey arcs and character development in a way we might have never figured out before. AI algorithms can analyse vast amounts of data and identify patterns that may not be immediately apparent to human writers. By analysing successful works of literature, LLMs can identify the key elements that make a storey memorable and impactful, and provide guidance on how to incorporate these elements to taste.

However as we know that LLMs are ultimately some glorified autocomplete mechanism not so far removed from the feature present on your phone keyboard, they obviously just replicate a coherent and superficially convincing walls of text. It is probably unlikely that they can genuinely attain the emotional depth and nuance that only a human writer can bring to a work of literature (as for simply replicating, we are not quite there yet). The best works of literature are often those that capture the complexity and messiness of human experience, and it's unlikely that an AI algorithm will be able to fully replicate this anytime soon. Discuss.

P.S. don't ask why I am using british spelling

 

[Collision]

The best works of literature are often those that capture the complexity and messiness of human experience, and it's unlikely that an AI algorithm will be able to fully replicate this anytime soon.

I think that this is too general to be a meaningful statement. The basic idea might sound true but you can't define, "the complexity and messiness of human experience," to a computer. Unless you can create a formal definition of this then the statement is just a platitude. There's really no reason, that I know of, to seriously believe a sufficiently powerful machine won't be able to do this. I think the more important question about a system like that is what it will cost to build and operate.

Generally, when discussing whether or not a machine could be intelligent there's a lack of rigorous and formal definitions. I think it's important to have formal definitions for words like "analyze", "intelligence", and "algorithm". Until this type of language can be formally defined it's difficult to discuss what properties a machine can and cannot have. Relying too much on colloquial definitions just muddies the waters. It's essential that we don't ascribe too much humanity to machines due to our own biases.

 

[Regal]

It is probably unlikely that they can genuinely attain the emotional depth and nuance that only a human writer can bring to a work of literature (as for simply replicating, we are not quite there yet). The best works of literature are often those that capture the complexity and messiness of human experience, and it's unlikely that an AI algorithm will be able to fully replicate this anytime soon. Discuss.

The problem with literature in the AI world is that it is easy to generate. Part of that is likely due to how formulaic lit is in general (ie The Hero's Journey formula, etc). From what I can tell, in 10 years I think you will be quite surprised what LLMs can put out there and be entertaining at best and cliche at worst.

Their weakness obviously is generating new ideas. They won't invent the next Harry Potter or Crime and Punishment. However if the goal is to be unoriginal (which is what the market is filled with pre-AI anyway) then AI can be quite successful.

 

[Fairykang]

It probably has a potential to be used as a tool for augmenting writing literature. That greentext story about the pit supervisor comes to mind. The issue is it being used to churn out SEO books for a quick buck which is already happening.

 

[lain is here]

It depends under what metrics you value literature. If you're the type of person who values literature as nothing more than a means for entertainment, then yes, theoretically AI could get to the point where people can have it generate stories that they find engaging enough to be entertained by.

But if you're the type of person who values literature as a gateway into the soul that the author has channeled and expressed in their writing, then no, AI by definition will never be able to create literature because it's not human nor an individual. Even if AI could theoretically generate a book that does have the feel of such a work, it's still meaningless because it's just a simulation. Under this definition, part of the value of literature is that you are getting to know somebody (the author) on a deeper level. In an abstract way literature under this definition is about getting to know somebody, but with an AI generated story, there is no 'somebody' to get to know, it's just simulation of a theoretical 'somebody.'

Let's take the example of a personal favorite book of mine, No Longer Human by Osamu Dazai. The reason why I love this book, is because Dazai is able to express abstract feelings that I myself share, and in that process I feel less alone for having them, knowing that somebody else (a real person) has had the same thoughts/feelings as well. If an AI generated a book like No Longer Human, it would be meaningless because even if it's able to convey feelings that I have had, I wouldn't feel less alone because it wasn't another human being validating them, but rather a robot simulating human thought.

It's the same difference between talking to a flesh and blood person and an AI. Yes, the content of the conversations may be the same, but the underlying value and context is completely different. With a person, it's not just the content of the conversation that matters, but non-verbal cues and relating to the shared experience of being human, not to mention the forming of an emotional relationship between two individuals. With an AI, the entire substance ends with the literal content of the conversation, there is no real subtext to be had, and no genuine two-sided emotional relationship to be had.

 

[LostintheCycle]

I think the dealbreaker for people will be the loneliness.
Do you think most people would still enjoy anime if they couldn't talk about One Piece with their friends at school? Same with books, people love to talk about them, read what other people say, and participate in book clubs. If our media is generated, that implies it would be individual, after all that's the whole idea right? If people have a private media bubble which isn't shared with others, then who will they talk to about it? What even is the point of all this media then?