In recent years, artificial intelligence (AI) has made incredible strides in its ability to generate human-like text. As a result, AI writing is becoming increasingly commonplace, with businesses and organisations using it to create everything from marketing copy to financial reports.
While AI writing is still in its early stages and far from perfect, it’s clear that it poses a threat to the livelihood of professional writers. After all, if a machine can produce text that is indistinguishable from that of a human writer, why would anyone need to hire a real person to do the job?
It’s not just low-skilled jobs like content writing that are at risk of being automated by AI. Even highly skilled jobs like journalism and novel-writing could eventually be replaced by machines. In fact, one Japanese company has already developed an AI system that can write novels better than humans.
Of course, it will be some time before AI writing becomes good enough to completely replace human writers across all genres and formats. But as the technology continues to improve, the day when machines can do our jobs better than we can is fast approaching.
The end for human writers?
The four paragraphs above were generated by OpenAI’s deep-learning AI writing model, called the Generative Pre-trained Transformer 3 (GPT-3).
Indistinguishable from the words of a human writer, the programme can respond to any prompt entered by a user, and amongst many other forms of writing it can construct a short story, hold a conversation, or write a news piece.
This begs the question: is this the end for human writers?
According to Professor Mike Sharples, who has decades of experience in researching writing and AI, the answer is “not yet”.
“You can either take a sort of apocalyptic view of, AI is going to put professional writers out of a job, it’s all doom and gloom and AI is going to take over,” he told Euronews Next.
“Or you can take the glass-half-full approach, which is that there are some amazing tools that are coming and as writers we can make good use of them and as teachers, we can make good use of them”.
Sharples, a professor of Educational Technology at the Open University in the UK, has written extensively on AI writing and its development, and he sees attempts to resist it – whether in professional writing, business, or academia – as futile.
Luckily for journalists or others who write on topical issues, the way the AI systems are trained means they are not fully up to speed with the latest developments going on in the world.
Even so, while writers don’t need to panic just yet, he does say they “should be worried”.
AI writing is being increasingly used for churning out content on the web, especially for the likes of marketing or blog posts as companies compete for SEO supremacy.
Freelance gig work sites like UpWork are seeing more and more job posts looking for writers to specifically use AI writing tools such as Jasper in order to generate content more quickly.
“You could either see this as a huge boon, a huge tool to help you write more quickly to get your words out, to get published. Or you could see it as a threat because anyone else could be doing that,” said Sharples.
Standard plagiarism detection methods don’t work
Social media sites like Reddit are awash with users telling their stories of using AI writing tools to successfully get good marks at school or university, or asking for advice on which are the best tools to use to avoid being found out.
One user, who said they were a university biochemistry student, told Vice’s Motherboard: “For biology, we would learn about biotech and write five good and bad things about biotech. I would send a prompt to the AI like, ‘what are five good and bad things about biotech?’ and it would generate an answer that would get me an A”.
Work that would have taken them two hours now only took 20 minutes.
The problem for those setting and assessing academic writing tasks, is that even the most up-to-date plagiarism checkers cannot keep up. According to Sharples, attempting to do so would be “a futile computational arms race”.
“The AI isn’t just copying bits from the web, it’s genuinely creating new text,” he says. “It’s inventing new ways of expressing. So the standard methods are not going to detect it”.
It is not infallible. Sharples used AI to write some essays, which on the surface looked entirely plausible, and wouldn’t be picked up by a plagiarism check. However, there were some flaws.
The AI knew to include references like any good academic essay, but on closer inspection it turned out some of the references were made up. Other citations the AI included were actually taken from studies that had argued the exact opposite of the point the AI was making.
The other more obvious way assessors could detect if students are using AI writing tools, he added, is if the quality of their writing suddenly improved.
Instead of trying to fight it, Sharples believes that just as professional writers need to accept AI writing is here to stay, educational institutions need to do the same.
He argues that educators and policymakers need to rethink how to assess students, and that AI systems could help students to learn to be better writers.
They can for example be used to quickly show students different ways to express an idea, or as a creativity exercise, where students could write a story in tandem with an AI tool.
How does AI writing work?
Sharples describes the AI writing model used by OpenAI as a “highly souped-up text completer”.
Just as your phone does if it has predictive text turned on, AI writing models look at what has been written before and predict what comes next. But while a phone’s predictive text looks at the last few characters typed, OpenAI’s model can look back at around the last 700 words, and generate hundreds of words of suggested text.
And what it writes makes sense, because it has been trained on almost all of the written text available on the Internet. It knows the context it is writing in, and so is usually indistinguishable from text written by a human,
“It’s been trained on Wikipedia. It’s been trained on blog articles. It’s been trained on e-books, online books, and the world’s literature. So it uses that vast database,” said Sharples, explaining it creates an internal “mental store” of how the language is processed.
And it is not just mindlessly regurgitating text either – it can come up with new ideas.
“This is what still even the developers of these systems don’t quite understand – that it’s not just parroting text,” he said.
“It’s not just taking previous words and reusing them, but it’s creating an internal representation, not just at the surface text, but of the ideas and the concepts behind it”.
“It’s creating this neural network, this multilayered network. And we know that some of those layers are about the words, the style, but some other layers are about how the text is structured and about the content, the underlying ideas”.
This all has major implications for more complex, time-consuming writing endeavours, such as writing books.
One recent post on OpenAI’s community forum details a member’s use of GPT-3 to write an entire 38,000-word book about proverbs, platitudes, and truisms.
“I used GPT-3 to generate many lists of proverbs and quotations from around the world and then used it again to write a brief description for each one,” user daveshapautomator wrote.
“This book has over 600 proverbs and quotations. I sent it out to a friend for beta reading, and will format it for printing while working on proofreading with Grammarly. All in all, it should take only a few weeks to go from first draft to printed”.
The future of AI writing
The flourishing of AI writing tools has happened alongside the release of a number of other AI creation tools.
Facebook parent company Meta recently unveiled an AI tool that creates GIF-like videos from text prompts.
Another of OpenAI’s tools which anyone can access, DALL-E 2, creates still images from text prompts.
Sharples, the professor, sees the future of AI content creation going multimedia, with more sophisticated options for creating text, images and videos as a package.
He also expects further refinement of pure AI writing, with the “pre-training” of the AI systems including more recent content, making them capable of writing about more topical items.
Furthermore, he believes they will begin to blend with models of how the world actually works, so the writing becomes even “more coherent and more plausible”.
And soon, he predicts, we will begin to see sophisticated AI writing assistants in our day-to-day word processors.
He cites Microsoft as an investor in OpenAI: “So Microsoft is going to want to get its investment back by integrating these into Microsoft Word and into other tools. So you’re just going to see them used routinely, and it’s just going to be part of the writer’s repertoire”.