The artificial author
How AI could change how we write fact and fiction

Artificial intelligence (AI) could completely change the way that we write. AI can generate huge amounts of comprehensive content on any topic, filter fact from fiction, and provide a creative spark to poets and screenwriters. But how should we think about this revolutionary tool? And could computers’ endless output possibly reveal something hidden about ourselves?
 

By Adam Smith

How is artificial intelligence going to change the way humans write? I think first of all we should consider what it is to write. What does writing do, and why? The purpose of writing is communication, although, I suppose that some people might be rarefied enough not to concede this point.

Writing is not just a way of recording information, but also a way of communicating it. Writing can communicate in many ways and at many levels.

For example, here is the first line of a well-known poem by Philip Larkin: “They fuck you up, your mum and dad”.

Although there is nothing in the sentence which would directly communicate that it is a poem - and although the sentence itself might not be considered a particularly poetic one - nonetheless, this oblique communication of poetry presents us with an important challenge to what we mean by poetry. If I were to say “They fuck you up” but don't explain that I'm quoting Larkin's poem, is it necessary that I still have to say “from Philip Larkin's well-known poem, “They Fuck You Up?” That would seem rather awkward…

So writing can communicate things indirectly. It is also a way of communicating with ourselves as much as others.

If I write something down, like a shopping list, then generally speaking I will remember it more easily. If the content of a piece of writing is relevant to me - and not just to others - then having written it down makes that relevance stronger in my mind.

Soothsayers and tealeaves

Those six paragraphs, bar the opening question, were written by Philosopher AI - an artificially intelligent algorithm built on a language model that uses deep learning to make human-like texts. The model, GPT-3, is the most sophisticated form of autocomplete that we have developed. Fed on the plethora of tweets, posts, articles, blogs, and other content that make up the internet, it has ‘learned’ the most likely sentences to follow a prompt. And, apparently, has a flair for poetry.

As therapists might recommend free association or as a beatnik writer might pen a stream of consciousness under the influence of hallucinogens, artificial intelligence is able to perform what AI author K Allando-McDowell has described as “automated graphomania”.

Not everything will be pertinent, or even comprehendible; it is up to humans to pick the meaning out of the patterns like soothsayers examining tealeaves. What does the robot philosopher’s answer say about us, and about itself?

‘Writing is a way to communicate’ is maybe the most straightforward response to the question, and the most obvious way of utilising Artificial Intelligence for writers of both fact and fiction. Over the last two decades, our ability to communicate with each other has exploded in the advent of social media. Yet, has also opened the gates to conspiracy theories and viral lies.

The sheer mass of posts has made it hard for humans to sort what is accurate from what is not. However, Artificial Intelligence may be able to turn the tide against this wave of misinformation. Companies like Logically in the United Kingdom use Artificial Intelligence alongside human researchers to stop the flow of online conspiracies, running potentially false claims through a database of previously fact-checked information to assign a credibility score and warn people against the uncredible before they share it.

Artificial Intelligence may also help journalists in other ways: writing short summaries of sporting events or quarterly earning reports, analysing databases and alerting humans to trends from big data, and even recommending content to readers. Rather than take journalists out of a job, this could save them valuable time to gather information, fact-check, and write the deep-dives that only a human can.

Writing, as Philosopher AI aptly pointed out, can also be used to communicate with ourselves. For creative writers, the AI team at Denmark’s Alexandra Instituttet partnered with publishing house Egolibris to create a digital editor. After being fed over 3000 bestsellers in a variety of genres, this algorithm called Edison, would provide feedback about readability, character interactions, and narrative arcs.

Some may fear the rise of cookie-cutter novels or scripts, but here Artificial Intelligence can simply scale up the insight already provided by works like Blake Snyder's Save The Cat or Joseph Campbell’s The Hero with a Thousand Faces – and perhaps reveal hidden depths behind our tastes. Art holds the mirror up to life, and it is possible that Artificial Intelligence can hold the mirror up to art.

No system is perfect, however, for good or ill. The former brings with it the kind of creative freedom utilised by data poet Ross Goodwin, who’s recent short Sunspring was created by an AI trained on dozens of science-fiction screenplays.
The result was glitchy, with scattershot dialogue that barely related to the previous sentence and paradoxical stage directions that were impossible for actors to adhere to, but the shape of the film is palpable – like a half-remembered dream – and remains entertaining to watch. Using an AI like this has a clear creative history. Over 200 years ago, the romantic poet John Keats proposed the notion of ‘negative capability’, whereby writers can pursue their vision unbound by a conventional framework of truth, logic, and science; the same framework a human might be limited by, but an Artificial Intelligence is freed from.

The biased and the banal

The ill, as is often said of Artificial Intelligence, is that it is not completely free from framework. Trained on human speech and perceptions, it therefore reflects human biases. The Coded Gaze exhibit at Amsterdam’s NxTMuseum shows how AI in facial recognition tools can reflect the racism and sexism within law enforcement and security agencies – unable to provide people of colour with the same individuality and privileges it provides white faces.

Moving forward – whether reading the essays in this dossier or thinking about artificial intelligence more generally – there is also the final part of Philosopher AI’s message to consider: relevance.

Although making a seemingly banal point about shopping lists, the algorithm in fact stumbles on the salient observation that, whatever has been written down, it is more likely to remember. At the bottom of the Philosopher AI website there is a prompt to share its output to Reddit. Doing so would then, eventually, become part of the training program for another Artificial Intelligence. The text made with GPT-3 in other programs are likely to be published on the internet too, all becoming part of the feedback loop.

While we do not know what tomorrow’s algorithms will tell us, their answers will be based on the questions we ask and actions we take today. It is our duty to ourselves, and to the future, to make sure they are good ones.