This article was written by an AI- or was it?
It was the year 2018, 2 years after Harry Potter fans were delighted with The cursed child and A journey through history and magic, when they were surprised with something they did not anticipate: Happy Potter and the Portrait of What Looked Like a Large Pile of Ash, a three- page chapter. The author this time however, was not the brilliant J.K. Rowling but instead a fully automated AI (Liao, 2017). Was this that big of a surprise? Maybe not. By now AI has been established in many areas of our life, including the creative field. Especially in the field of creative writing there are hundreds of different AI’s attempting to copy the human’s skill that has been formed for centuries. Such as the AI that has created the aforementioned example. This is now leading to the question: Will AI’s ever be able to replace authors in the field of creative writing?
What is Creative Writing?
In order to examine that, we need to define what creative writing is about. According to Oxford Dictionary, creative writing which is usually fiction or poetry, displays imagination or invention, and is often compared to academic or journalistic writing (Oxford Dictionary, n.d.). If you have every attempted to write creatively, you know about the many struggles it comes with:
What header will the readers draw into reading my story, what emotions should I depict? Do they care about my deepest worries and concerns or am I oversharing? What makes this unique, how does this differ from other stories? It would be nice to have a creative assistant for that, am I right?
AI as an Assistive Tool
While there are AI’s that are able to create pieces all by themselves, there are many other examples of AI being used to support the author, not taking over their role but rather augmenting the skills of the author. I want to introduce you to their foundation named GPT-2. GPT-2, or Generative Pre- Training 2, is the second iteration of an Open Source language model developed by Open AI. The model was trained “simply to predict the next word in 40GB of Internet text” (OpenAI, 2019).
This ‘simple’ strategy of training allows GPT-2 to be more flexible than many of its cohorts, because its dataset is so diverse GPT-2 can outperform its rivals that were trained on specific datasets such as news articles or Wikipedia entries, GPT-2 outperforms these models even within their specific datasets. GPT-2 presents “state of the art results on 7 out of 8 tested language modeling datasets” (Radford, et al., n.d.). GPT-2’s full functionality was never made public due to serious concerns posited by its creators that the model may be used to create content that can mislead readers, it can generate fake news or phishing content, both of which require a reader to deem this content as ‘true’ and believable.
Using AI as a tool to augment the author’s capabilities is a potentially bright future for authors. Shortly is an application that does just that, when a user inputs a piece of text. Shortly builds upon that text, adding new information and color to the scene to help a writer break through writer’s block. It doesn’t just spit out some random amount of text, though, it builds on the parameters that you set through the ‘Story Background’, the AI instructions in which a user inputs what they want the AI to write about next, and the output length slider, where a user can choose how long of a text they want. This has serious implications for authors, consider being able to get a fresh, in-your-own-words, perspective on your writing when you encounter writers block. Even if the text created by Shortly is not used, it is a refreshing way to approach a passage that is blocking the author. It’s effectively a road block destroyer, creating new avenues and way for the author to continue their piece.
AI as an author
While there are many AI’s that can support the author during their writing process, there are also AI’s that can write completely by themselves. One example is the GPT-2 based AI “Talk to transformer”. What is so special about this AI is that it is working remarkably well. This is why the producers did not release the full version at first, instead only a stripped- down version. Reason being their concerns about possible malicious uses of this technology (Kremp, 2019).
The future of creative writing
Artificial Intelligence has been slowly creeping into our lives for years now, it is in the factories that produce your clothes, it is in the algorithms that show you these clothes on your preferred social media platform, it’s in the cars that pass you on the road. It is here. And one of its most audacious endeavors yet is to creep into the creative world.
We see AI as a creative writing assistant mostly, an augment to the writer, oftentimes as a niche tool that is out of our grasp. But AI such as GPT-2 is already so sophisticated that its creators are too worried to release it in full because it may be used to create fake news, to write convincing phishing schemes or worse, the AI can write people! So, we are left with that conclusion, and now we must determine how we feel about it, if we want a world in which computers are ‘creative’ or if we find it problematic. Some who see it as a problem point to the inherent bias that AI and the algorithms AI produce hold.
To that I’d say, AI will be biased always, this much is true, but people will be biased always as well. The moral issue is that AI is just a reflection of our own collective conscious, and it is always playing catch up, and could very well reflect the ugly parts of our not so distant past and bring them right back into the present. AI must be taught to learn the culture of today if it is to be welcomed into society as a creative voice.
And this is not just a moral problem, it is a technological one, how can we train an AI fast enough to be consistent with the ever-evolving cultural norms of a society?
Beyond the moral and technological, we must consider the philosophical, how can something write creatively if it has no original perspective, how can an AI truly ‘create’ from an authentic point of view if it just recreates based on inputs. This is a consideration we all must tackle for ourselves, if we think creative writing written by a computer has merit. But there is a future in which AI can write creatively in a symbiotic relationship with the human.
Maybe one day authors will become more like architects, building the framework of stories, and AI will be the construction workers, realizing the plans of the author. An author may begin to set out characters, dispositions, main events, and allow an AI to fill in the rest. This type of automation is revolutionizing all types of industries. We see it in the industrial sector, factories are no longer filled with humans but with whirring robots. We are no longer the makers but the managers of the makers, this fate could certainly befall authors as well. When an AI can so convincingly write, the real task will be the one building the framework in which it writes, shifting our power from creation to management of creation. Which begs the question — are we really in control if we must shift our focus?
