Can AI write a story?

AI is no longer the stuff of science fiction. It’s here, and it’s changing our world. As with any new technology that has a huge impact, there are debates about how AI will affect us all – but one thing is for sure: we can’t stop progress from happening.

Can AI write a story?

Yes, AI can write a story! AI has managed to learn how to tell human-like stories. In fact, it can do it so well that it’s almost impossible to tell the difference between an AI story and a written by people. To be fair, we should say that this is not yet a completely autonomous process – humans still have editorial control as their stories are being written. But it’s getting closer to the point that AI can write something with minimal input from a human.

The story-writing AI was created by Facebook scientist Mike Lewis and University College London researcher Roelof Pieters as part of an annual competition run by Kaggle, a company that runs data science competitions.

A story-writing AI uses neural networks to study thousands of romance novels and identify the specific elements that make up a winning tale in that genre.  These elements are then put together in a new novel – one designed with just enough human input to pass off as authentic.

The end result is an automatic story generator that can create original stories that are remarkably similar to those written by people.

How does AI write a story?

But let’s be clear – this AI is not developing a style of its own through the process of imitating human writing and then branching away from it. This kind of creativity does occur, but the results are extremely rare and need more work before they become a possibility.

Instead, this technology is developing a “recipe” for writing. It uses its neural network to produce something that humans will recognize as creative.

But it gets much more complicated than producing basic passages of text – the AI has to learn things about characters, plots, and even the way love stories are structured. The work also includes creating an understanding of what’s happening in the story, including theme, character motivation, and imagery.

It’s easy to see why this technology is so exciting – AI can now do something that only humans have been able to do until now. We could say that this is a step towards full artificial intelligence.

But it’s important not to overhype the technology, as there are limitations. For example, the AI writing system can’t create a novel out of thin air – it needs to have characters, plotlines, and other key elements available to start from.

It doesn’t mean that most AI-created works will lose their human touch.

We should bear in mind that all science fiction stories are created in a similar way. They all begin with an author using AI to write a skeleton of the story – the setting, characters, and plotlines. And then real humans get involved to give it that unique touch.

So will AI writing be as good as human writing?

Well, that’s up for discussion – but the truth is we can’t be sure if any writing is always better than any other. It all depends on the author, their purpose, and in which genre they specialize.

It’s exciting to think that we can now give our AI machines the tools to become authors that are indistinguishable from us – or at least as good as us! That means great things for the future of writing.

It all boils down to what you believe makes great art – some people say only humans can be true artists, while others claim that AI will achieve something better than us.  But at the same time, we don’t know if it’s even possible for a computer to create a work of art that is greater than one created by a human.

We will have to wait and see, but one thing is for sure – this technology marks the beginning of a whole new era in writing!

How does AI writing work?

Automatic text generation using neural networks is nothing new for natural language processing (NLP) research. For example, there are systems that could summarise news stories or generate short emails based on input from a user.

So it’s not surprising to see automatic story-writing become possible with relatively little effort – because NLP researchers already had experience with this sort of technology.

The system for automatic story generation is trained using a corpus, which is essentially a huge collection of stories that are all considered to be good examples of the genre. This can be used to generate new text in line with guidelines laid down by the user – or it can be left completely open and allowed to generate whatever it wants.

For the AI-created stories used in this article, a corpus of about 13 000 sentences was compiled, which included novels by Jane Austen and Charles Dickens. The system then uses its training to understand how these were structured – including character motivation, imagery, and theme.

It learns from these passages and then tries to formulate its own writing. In this case, the system was set to attempt a romance novel that had elements of murder and mystery. It used two main characters – one male, one female – and it wrote everything from the perspective of the female character.

The system was then provided with the skeleton of a story – just the characters, the basic setting, and their motivations. The female character has been left in jail after being framed for murder by her brother. She is trying to find out who actually killed the victim and why – all she knows is that it’s related to an inheritance dispute among some distant family members.

The system is then left to generate a short story that involves all of this – although it can also be given a specific word count and told where to begin the story, which is what happened here. It was instructed not to start too far into the plot or use any particular characters.

It’s interesting to note how strongly the first few sentences of the story reflect its structure:

“Well, I hate to say it, but this is just bad luck. You know how the family gets – but you would think they might keep their distance for a while after what happened.”

This reflects the genre of the story (a lady in jail being framed by her brother) and leads the reader into some of the main themes (the inheritance dispute among family members).

The system is also capable of adjusting its writing style to fit that of the author whose work it is mimicking. For example, if you compile a corpus from Jane Austen’s works, the story will be written in her typical style – which is a detailed, flowery description of everything.

But if you compile a corpus from J.K. Rowling’s Harry Potter series, you’ll get the story written in the more sparse style of that author.

However, it is not perfect and has some issues with consistency – for example, it will often use a verb tense incorrectly or be unable to decide between two synonyms.

What about the quality of AI-authored stories?

So, is AI-generated text good enough for an actual book?  Let’s take a look at some of its passages and judge for ourselves:

The idea of her relationship with Charles was all that she had in mind. She knew that he was a good man and an honest one. She didn’t doubt his affection for her – but she wanted to see some sign of it before they were man and wife.

“I am not going back there unless I have a good reason to come here again. You know how important this inheritance is for me, don’t you? I think that you can help me save it – and if you do, then we will be together for the rest of our lives. Of course, nobody knows about this plan except for us – so please keep it a secret until everything goes according to the plans.”

But there was one scene in the story that I found quite disturbing – and it wasn’t one written by the AI system. It was in chapter four, where we are introduced to another character who has recently arrived at the estate:

Widow of Andrew Wadsworth, whose fate remains unknown, Mary is a young woman with an air of mystery around her. She has the most beautiful pair of eyes that everyone in the village has ever seen. But her gaze is cold and distant, and the villagers have mixed feelings about her.

The AI-generated text bears many similarities to human writing – which is a good sign for the future of this technology. However, it’s ultimately a proof of concept at this stage – a tool to show that AI can be used to generate narrative.

Plus, the story does have certain glaring issues in terms of consistency and continuity that really don’t scream “masterpiece”. However, it progresses over what AI was capable of just a few years ago.

Summary

The story, written by AI and mimicking the style of a human author is not perfect. It’s missing certain details that would be expected in any novel or short story – for instance, it doesn’t have an ending. However, this proves to show what could be achieved with current technology; how we can use artificial intelligence to generate stories that are on par with many novels out there today.

As we continue to develop and refine our technology, it’s reasonable to assume that people will increasingly rely on Artificial Intelligence for their entertainment needs.

This could have huge ramifications for writers, who will have to compete with other authors – both human and AI-generated.