This is a question which is increasingly gaining salience within the public discourse. One would be naive to think that the notion of machines replacing wordsmiths derives only from the current epoch.
As far back as 1953, Roald Dahl in a short story from the book Someone Like You, expressed an Orwellian scenario where a machine known as the Great Automatic Grammatizator was able to write prize-winning novels based on the works of living authors in only 15 minutes.
Since those days AI has emerged from being a field of study to forming an integral part of technology within modern society. From streaming shows on demand to ordering your next delivery meal, AI underpins the relationship between technology and society.
When we dial this into writing, it is indeed the subset of AI known as Machine Learning that has taken a quantum leap on the last 20 years. This has been made possible by the gargantuan digital landscape that now exists via the internet allowing machines to learn from big data. And enrich their learning to levels not seen before.
With all these advancements, this begs a greater question - when rather than will AI come to replace human writers? In the next 5 years, next 25 years or within the next 100 years? One way to attempt to try and put a timeline on this paradigm shift within content writing is to provide a snapshot of the existing landscape.
When it comes to natural language generation Open AI are at the vanguard. They’re currently on their 3rd iteration of a language model known as GPT which stands for Generative Pretrained Transformer.
They recently achieved more mainstream acclaim after UC Berkeley student, Liam Porr, published an article on his blog created from GPT-3. It eventually even got to the top of Hacker News.
Although GPT-3 is currently in a closed beta, Porr managed to convince a friend doing a PhD to give him access to the Open AI API. He wrote a script that gave GPT-3 an article headline and intro. It then generated a few versions of the article and Porr chose one for his blog which he copied and pasted from GPT-3’s version with very little editing.
Although the article reads very well, several comments did suggest that a bot was behind it. Moreover, Porr acknowledges that some editing was done. To what extent we will never know. See article excerpt below.
For those unable to wait around for GPT-3 to emerge from beta, there are other automated writing services out there such as Kafkai and Contentyze who have featured on the IndieHackers Podcast. They work in slightly different ways. Kafkai automatically produces content for selected niches without the need to provide an article title.
Contentyze enables you to input an article title and automatically create a post from there. They are also building in an SEO feature allowing for the articles to be SEO-optimized. Having assessed content from both of these companies it’s clear that although the content is produced almost instantaneously, a reasonable amount of editing is still required to make it suitable for publishing.
Here are a few areas where bot generated content has room for improvement:
Grammar - In some cases AI generated content will position two words alongside one another that don’t make any sense. In other cases certain words will be repeated in close succession. Other times the writing is simply grammatically incorrect. See example below.
Logical reasoning - Currently AI content generation lacks the ability to consistently practice logical reasoning within its writing. Example: The grass is outside and so when it rains it gets wet. However, if the grass is covered by a roof it will remain dry.
Understanding context - The nuances of context can often change the meaning of the words. Turing tests that try to distinguish humans from chatbots often include a challenge to decode so-called Winograd sentences. For example, the meaning of “they” changes in the following sentence when you swap the words “feared” and “advocated”: The city councilmen refused the demonstrators a permit because they [feared/advocated] violence.
In spite of the challenges AI generated content currently faces, it’s suffice to say that it’s only going to continue gaining more traction within content creation due to the self perpetuating nature of its learning.
Although we may not quite be at the point where we can hand over entire content writing tasks to bots, we can certainly start to consider using them for more menial/upstream aspects of content creation such as writing introductions for articles. The media industry have adopted this approach by embracing automated journalism where breaking news stories are produced by bots using real time raw data inputs.
We should start to consider AI content to be supplementary to our current content writing efforts in a way which creates a symbiosis between human and machine. We save ourselves time on certain outputs and in the process we help AI to become more proficient at writing exemplary content through a continuous feedback loop.