Everyone’s heard of Artificial Intelligence (AI) and how it’s changing the world, but what is it? How does it work? And, importantly, how does it affect authors?
What is Artificial Intelligence (AI)?
AI is basically a computer system that attempts to simulate human intelligence. The key is that it learns by itself by consuming massive amounts of data – reading everything on the internet – and looking for patterns in what it sees. By understanding the patterns it finds and applying statistics, it can then produce something of its own with a similar pattern.
This learning process is crucial. Simpler AI systems are trained by being provided with lots of data, with a human telling the system what it’s looking at. For instance, it might be shown many pictures of cats, and many pictures of dogs, and it is told what’s in each picture. It then looks for patterns and commonalities in the pictures, and works out how to differentiate a cat from a dog. More sophisticated artificial intelligence systems have the ability to learn by themselves, not requiring what it processes to have been pre-labelled by a human. Those systems are described as “Deep Learning”.
To see how an AI system can learn, consider the example of a help desk’s chatbot, which can be trained by reading transcripts of millions of real-life customer exchanges to learn how a human replies to different questions. Then, if someone asks it a similar sort of question, it knows the way in which to answer. The important point to note is that it doesn’t just copy a reply it’s seen a human use, but generates its own reply based on the question being asked, the context in which the question arose, how it’s being asked, and its own knowledge of how similar questions have been answered in the past that have been seen to satisfy the customer.
And, crucially, the system continues to learn: if it gives an answer but the customer clearly isn’t happy with its response, it learns how to do better in the future by seeing how the customer responded and what they asked next.
But the input to an Artificial Intelligence system doesn’t have to be text or speech. AI can also analyse millions of paintings to spot patterns in what it sees, and then produce its own work by using similar patterns itself. It “just” needs some prompts to get it started to tell it the kind of image to produce, and away it goes (prompts are similar to the questions a chatbot receives). A little later in this blog, I share my experience using artificial intelligence to create an image for a book’s front cover, and how an author used it to create a novel.
How does Artificial Intelligence (AI) Work?
So, how do these AI systems manage this?
Artificial Intelligence software is structured in a similar fashion to the human brain. Like the way the brain’s 86 million neurons interconnect in groups and can be triggered by various inputs, so the computer processes data by passing it to groups of “nodes” that interconnect in a giant mesh. Each node analyses the data in its own way, producing a value as the result of that analysis. If that value exceeds a certain threshold, that node triggers, and passes data to other nodes for them to analyse in their own way.
The nodes are grouped in layers, similar to the human brain as shown in this simplified example:
Nodes in the input layer process the incoming data and pass information on to the nodes in the middle, which analyse it further before sending it to the output layer. The output layer then collates all their analyses and produces an output.
These groups of layers of nodes are called “neural networks”, and provide a way to classify the input data at incredible speeds. Using this technique to analyse images that would take a human several hours, for example, would take a computerised neural network only a few minutes. A deep-learning network may have millions of interconnected nodes, grouped in up to 50 different layers.
The most common AI tool currently in use is ChatGPT. The suffix tells us a lot about how it works: The “G” tells us that it’s generative, i.e. it can generate output text. The “P” tells us that it has been pre-trained and has used deep learning to analyse lots of data by itself to find its own patterns in what it has consumed. Perhaps the most interesting suffix is the “T”, which tells us it uses what is called a “transformative architecture”. This is key to the speed and accuracy of the system, and was a method first introduced in 2017. The input text is not read from left to right as we would, but the text is broken into chunks, which are all examined simultaneously. The system then decides which words are the most important, and the AI algorithm then gives more attention to those than to the other words in the text.
From its analysis and training, it then decides on the most appropriate response.
When it comes to using AI to create something (e.g. a picture or some text), one of the most important aspects of using AI to understand is the importance of the prompts the system is given as the input to its analysis. The computer analyses the prompt, seeks to understand it, and then produces a response that it thinks is the most appropriate, given the content, structure, and context of the prompt. There are some examples of prompts later, when we look at using AI to create a novel and a suitable front cover image.
What’s Artificial Intelligence (AI) Used For?
Artificial intelligence has more applications that I could name here. The most common are probably speech recognition, predictive text, chatbots, grammar checking, and image analysis. I read an article recently that described how AI was being used to improve safety on the Devon railway network near Dawlish: cameras monitor the cliffs and track, with AI analysing the images to detect hazards such as rock falls, and alerting a central control room if anything is found.
Artificial Intelligence is a very powerful tool for image processing, and is now more accurate (and much faster) than humans. It can learn what something normally looks like, and then identify any anomalies. Medical science is benefitting greatly from its ability to analyse MRI and X-ray images to spot cancers.
Similar techniques provide a key component for self-drive cars, by analysing the environment around it using cameras and other sensors, and checking what it “sees” for hazards, road signs, other vehicles, etc.
Work on Artificial Intelligence started in the 1940s, when scientists were studying how our brains use interactions between neurons to make decisions, but it wasn’t until 1989 that a neural network was first used to train a computer. Its first application was to recognise hand-written postcodes on envelopes to help automate postal sorting in the US.
These days, the place we’re most likely to come into contact with AI is the chatbot, which is increasingly used on websites to provide a first-level of “Contact Us”. The customer types in a question, and artificial intelligence analyses the text to work out what is being asked. It can then provide an answer from its accumulated knowledge in a way that sounds like a human answering your question. Although “connect me to a human” can sometimes be the best text to enter!
Can Artificial Intelligence (AI) Write a Novel?
The big question for authors, though, is whether AI will ever be able to create a readable novel. Currently, it can’t do this by itself – or, at least, it can’t produce work of any quality by itself. Some publishers have recently been forced to close their doors to unsolicited submissions because they’ve been inundated with articles and short stories generated by artificial intelligence. The quality is so bad that they can’t be used, but each submission needs to be read; publishers have been unable to cope with the sudden surge in these submissions. And this is bad news for writers because many authors rely on being able to submit freely to such companies for their livelihood.
One author, Stephen Marche, decided to see if he could use AI to write a quality novel. It turned out it took loads of work, and was heavily reliant on breaking it down into small chunks and being very clever with the prompts that were used, including the style of writing you ask it to mimic.
The Society of Authors reported on his experimentation recently, and noted that Marche explained that, to write a novel in the style of Raymond Chandler, asking AI to write in that style simply resulted in “a very pale photocopy of Raymond Chandler.” Instead, he said it worked out to be better to ask AI to “write something about a murder scene in the style of Chinese nature poetry, then make it active, then make it conversational, then Select All and put it in the style of Ernest Hemmingway.” Hmmm… it sounds like he had to do a lot of experimentation to get that right!
How to use Artificial Intelligence (AI) to Create Cover Artwork
Using Artificial Intelligence to create an image for the book’s cover, though, seems somewhat simpler. To see how easy it is to use AI to create the artwork for a book cover, I went to the FOTOR website and selected their “Text to Image” option. I was then able to enter a text prompt to start their artificial intelligence working for me. I decided to try to create an image that could be used for the front cover of a thriller, so entered the prompt “scared woman running through trees”. The AI system analysed my text and produce a couple of suggested pictures, although the result wasn’t really what I was looking for:
They looked too much like cartoon images, and I didn’t want snow, so I then added a “negative prompt” of “snow cartoon” to tell it the type of picture to avoid. I also wanted just one figure, not several, and I wanted it to look more like a photo than a painting, so I modified my initial prompt to be “one scared woman running through trees photo”. This was immediately a lot better:
However, I wanted the woman to be running away from the imaginary camera into the woods rather than towards me, so I added an additional prompt of “running away”. I also decided I wanted it to be more misty (so I added a prompt of “foggy”), and that she should be wearing a red coat rather than a jogging outfit to give the impression she was running away from something rather than just out for exercise. With new prompts added, I asked it to generate the new image:
It was now much better, but she had been left wearing what looked like trainers or running shoes, which I didn’t like. Also, I wanted more trees, and for the image to have a green tinge. And what would it look like if she was running with a holdall, I wondered? With additional prompts added, it gave something close to what I was looking for:
It still wasn’t perfect, but it was clear that, with a little more finessing, I would be able to get a suitable image that could then be imported into something like Canva to create a thriller’s front cover. And it took only a few minutes.
One thing was clear: the skill in using the tool came from entering the best positive and negative prompts. It seemed to me that, yes, it would give self-published authors an easy way to generate their own front covers, but it would also provide professional artists with a powerful tool they could use in their work. At the moment, at least, it wouldn’t be making those professionals redundant – not until AI could read a complete novel, analyse it, understand it, pick out the key points, and then make a decision on the artwork by itself.
The concern with this, though, is what was used to train the artificial intelligence. It has viewed trillions of images to be able to understand what I was asking for. Where did those images come from? And here’s the moral dilemma – if an artist spends time and effort to create the artwork for a book’s front cover, who’s allowed to use it? The publisher is given the right to put it on the front cover of a book, but is a third party (i.e. the company developing the AI system) allowed to use that artist’s work for free in order to teach its computer what a typical cover looks like?
What is the Impact of Artificial Intelligence (AI) on Authors?
And this is where one of the problems arises for authors. The system of copyright is crucial for an author’s income, but there are many pirate sites on the web that have managed to get hold of books and illegally provide them for free. I know of three sites, for example, that have pirate copies of Eavesdrop, my own thriller.
When AI is trained, it reads and analyses everything it can find and learns from it. That means it has read my book for free and benefitted from it. As it has from millions of other books illegally available. The knowledge it has gained from reading them is then used to generate its own works, from which it can gain revenue. My book – and that of probably every other author – has been used to generate income for someone else, and we poor authors – without which it couldn’t work – received no recompense for that. As a result, there are currently many legal cases in progress, with groups of authors taking the AI companies to court over these copyright abuses.
The Society of Authors is working hard on this issue on behalf of authors, to try to find a way to make the system fairer. Time will tell whether a solution can be found.
Ian Coates is the author of a thriller, Eavesdrop, first published by Bad Day Books, the suspense and thriller imprint of Assent Publishing. He worked in the high tech electronics industry for 30 years, where he specialised in the design of radio communication equipment. His intimate knowledge of that environment always triggered his imagination to think about the mysterious world of spies, and allowed him to bring a unique authenticity to his thriller. Ian is proud to support the British Science Association and donates a proportion of his book proceeds to that charity. He lives and writes in Worcestershire, England, and is a member of the Society of Authors and the International Thriller Writers Association.
His second thriller, Backlash, will be released later his year.
IanCoatesThrillers 