An article for all writers terrified by the new scenarios of AI and Large Language Models

The release of the first LLMs (Large Language Models) free, ChatGPT among all, it caused quite a stir.

In the content marketing sector, these technologies have been talked about for a while: I collaborate with a team of science fiction authors who have them used at unsuspecting times for your sci-fi themed blog. The diffusion of AI among the general public, However, brought the topic to the forefront in all media.

It's more than legit, for a writer, wondering if these amazing computer gadgets they will supplant the creative work of human beings. Some people ask this question with curiosity and a desire to find out more, others with acrimony and a totally closed attitude. But don't worry: in the content you are reading, We will try to clarify without alarmism. Why, I'll tell you right away, the human being still has something that AI does not have.

What artificial intelligence is missing for writing texts, compared to humans?

AIs are missing a lot of things, summed up in a single word: consciousness.

An artificial intelligence (which it really should be called, to avoid misunderstandings, simulated intelligence) he has no self-awareness. When he produces a song, he doesn't know what he's writing: generates words associated with each other on a statistical basis, coming from a frightening amount of texts in its database.

I Large Language Models

Not, to better understand what we will see later, it is good to make a clarification: technologies like ChatGTP or Bard they belong to a subcategory of artificial intelligences called LLMs (Large Language Models, in Italian “Large language models”). They are part, Therefore, of a class of AI trained to generate coherent sentences and able to answer specific questions. They are based on deep learning, an automatic branch that uses artificial neural networks to learn from large amounts of data.

LLMs are undoubtedly very powerful models capable of performing complex tasks, with sometimes surprising results. But they have no emotions, they have no memories, They don't have a personality of their own.

"Not yet!» I heard some people say «Wait until they become aware, and you'll see what they'll be able to do.". But this hypothesis is plausible?

No: i LLM, as they are designed now, they do not risk "awakening" and becoming self-aware

I recently read an essay with a “reassuring” title A world without work (Daniel Susskind, Bompiani, 2022).

Leaving aside the more extreme scenarios that gave me nightmares for a week, I would like to highlight an enlightening part on the history of artificial intelligence. According to Susskind, the idea that an AI is self-aware and can, one day, have feelings, it is very widespread but it now belongs to the past.

The turning point in AI design

During the first experiments on these technologies (dating back approximately to the 1940s), this is how we proceeded: scientists, after observing how people performed a certain task, they attempted to build machines that copied these mental processes.

The results, But, they were disappointing. Human thinking and reasoning are not replicable findings on an artificial level. Susskind calls this approach to AI “purist”, in the sense that it aimed to reproduce the human mind in a pure and faithful way. Somehow, it was believed that we ourselves were a complex type of computer and could therefore be artificially "reconstructable"..

When the first progress in the world of AI began? When the purist approach was abandoned in favor of the pragmatist one. It turned out, Indeed, That machines work very differently from the human mind. The tasks they complete can seem performed in the human way, but in reality they start from completely different assumptions.

This explanation is very simplified, I realize it: to learn more, I recommend you read Susskind's essay. I myself am not an AI expert, but I'm very interested in how language works, which I was lucky enough to explore by obtaining my specialist degree in Linguistics. If you are reading this, you are a scholar or a scholar in the sector and you want to add some information, write to me and I will be happy to integrate the article with new food for thought!

How LLMs succeed in writing stories?

Let's take a concrete example. If I ask an LLM-based AI to write a story about a little girl who adopts an abandoned kitten, the car will not think about the kitten, nor to the little girl, nor to love that binds the two. She won't remember when she too wanted a puppy as a child. No, that could be done by a writer or a writer in flesh and blood.

The AI, Instead, will fish in its segment database (called token, that is, minimal units of analysis that do not necessarily correspond to syllables or other morphosyntactic notions) related to our request and will assemble them on a statistical basis, in order to construct a plausible and coherent text. But he won't understand what he's writing.

So, what more do we human beings have??

First of all, us we understand what we write. It doesn't seem like much to you?

We have an experience. We have experienced first-hand the sensations that we put down in black and white, if we haven't lived them, we know how to empathize with our peers who have experienced it. We can immerse ourselves in their thoughts and try to understand what goes on in their minds.

We have unpredictability. Yes, because sometimes we start writing a text with a precise outline in mind, but then the story and the characters lead us to unexpected developments. And that push, those intuitions that take us somewhere else, Often, they are brilliant. Stephen King explains it well in his On Writing, telling that the ending of Misery, in the plot planning phase (That, in his case, always starts from one what if) was completely different. What can I say… human unpredictability did a good job in turning history, given the masterpiece that came out of it.

We humans also have a style that evolves over time. We have memories and emotions. And we have a love for writing. Yes: we like writing, we are not programmed to do this. Therefore, let's not stop! Let's not be intimidated by a machine, from an instrument. Rather, we try to use it as best we can!

How to use artificial intelligence to write texts profitably?

Instead of being supplanted by LLMs, in my humble opinion the best way to exploit them is use them for our purposes (and don't let us use them!).

Translated into the field of narrative writing, we can employ them to help us with various collateral tasks, among which:

• research of sources and historical-geographical documentation;
• search for information about the potential audience for our novel;
• support for the preparation of promotional texts: caption per i social, outline for blog posts etc;
• support in drawing up the lineup of our story.

Even in this case, Let's see a concrete example. Let's try to analyze a prompt (that is, a command) to help us get our novel off to a great start. This is it a useful tip for those who have a somewhat vague idea in their head, but he doesn't know where to start to make it happen.

You could, Therefore, share the title of your book with the model, the genre, the main characters, the setting and the main conflicts of the protagonist. Then you could ask them to generate a list of questions that you will need to answer before starting the draft, to delve deeper into your idea and better define the reference niche.

Let's see the possible result, using Bing!

A concrete example

PROMPT:

“I would like to write a novel with the title The last girl. The genre is young adult fantasy and the main characters are Eileen, Liam and Nora. Eileen is the protagonist: she is the last fairy left on Earth, or so he thinks. He has the gift of communicating empathically with animals and seeing the ancient portals that once connected the terrestrial world with the fairy world. Liam is a fairy prince, who at a certain point bursts into her life to reveal that she is not the last fairy left in the world: there is still a community hidden in a dimension parallel to ours. Nora is her best friend, but only at the beginning: during the course of the novel she reveals herself as an enemy, because he tells everyone Liam's secret, instead of keeping it hidden as Eileen had advised her to do. The setting is a fictional Irish island named Doran, which hides a portal to Liam's magical kingdom. Eileen's internal conflict is the acceptance of her nature as a fairy, the external conflict is the rivalry with Nora. I need a list of questions that I should answer before I start writing the text, to delve deeper into my idea and better define the audience for my novel".

3, 2, 1… Let's start with the prompt! Here's the answer:

“Ciao, this is Bing. It seems to me that you already have a fairly clear idea of ​​your novel. However, there are some questions you could ask yourself to delve deeper into your idea and better define the audience for your novel. Here is a list of possible questions:

• What is the main theme of your novel? What do you want to communicate to your readers?
• What is the tone of your novel? It's light and fun, the dome and dramatic?
• What is the target age of your audience? Your novel is suitable for children, teenagers or adults?
• What is the narrative point of view of your novel? You tell the story from a single perspective or from multiple perspectives?
• What is the structure of your novel? You have already defined the beginning, the development and end of your story? You foresee some plot twists or cliffhangers?
• What is the genre of your novel? Is it just fantasy or does it have elements of other genres as well, like romance, the thriller or the adventure?
• How do you describe Liam's fairy world? What are its characteristics, its rules and its magical creatures?
• How do you describe the secondary characters in your novel? What are their roles, their personalities and their relationships with the protagonists?
• How do you describe the style of your novel? What are the keywords, the metaphors and figures of speech you use to create vivid images in the minds of your readers?”.

These questions give us a first clue, which could be useful for transforming an idea of ​​a few lines into a real novel. We can discard some of them and we can work on them, as is obvious, Why it is not a magic elixir to suddenly transform us into literary wizards.

The limits of AI

LLMs are valid tools for the brainstorming phase, maybe they can provide some ideas for overcome writer's block, ma they don't write an entire book for us. And I'm still being perfected, as we can see from the fact that, in the example above, Bing asks us if our novel is “suitable for children, teenagers or adults", even though we have already told him that he is a young adult. In short, they are certainly not perfect and infallible, they require a lot of brains (allegedly) to be used to the fullest. Let's remember that they are based on probability, so it is not certain that every one of their results is correct. AND likely that it might go well, but it's not sure!

Often, Then, artificial intelligences (not just LLMs) are programmed with inaccurate data, incomplete or, worse still, they are based on dangerous cognitive biases. For this reason, must be used with caution and, above all, with a healthy dose of awareness!

And the, how convenient it would be to use artificial intelligence for writing texts? You would like to simplify your promotion work, the search for documentation or perhaps the market analysis? Write to me to tell me about it!