Head of Content at Let's Enhance
This year brought us two Nobel prize nominees in literature. Feels like there’s a “buy one get two” promotion going on, and that’s good! More awarded authors - better stories.
There are two absolutely opposite thoughts:
“Holy Molly, AI is gonna destroy us. So, writing texts is just a first and easiest step on its grand plan in conquering humanity and taking all we’ve got!”
“Oh Jeez, keep calm, the machine simply can’t do the creative job. The art area will always belong to a human being”
So, where is the truth?
As far as texts can serve not only recreation but informative purpose, is there a place for AI technologies in journalism?
In reality, AI became a really good helper in writing texts.
No, he is not going to be next Nabokov or to generate a sold-out masterpiece for you either.
Not spending an awful lot of time on a monotonous and mechanical writing job is a precious gift. Journalists would sell their souls to not transcribe audio from interviews, transform data or make copies based on the articles, etc.
And to have the machine slave writer sounds more than seductive.
Trained AI will allow journalists to spend more time on investigation and collecting facts.
Frankly speaking, not all content should be highly creative or witty after all.
And that’s where AI can help us a lot.
AI is brilliant for reports based on data. That’s a type of job business reporters find snoozy, meanwhile the AI “journalists” do not complain and just do their job.
This year, Australian’s Guardian published a report of political donations mostly written by AI.
But Washington Post were the pioneers who implemented the new technologies directly into their job.
They created a homegrown AI technology called Heliograph.
With a bit of fear it was launched right before the Summer Olympics in Rio. During this event, it dropped over 300 short reports.
After that, the technology proved its usefulness one more time during the elections in 2016. Since then Heliograf was used to generate not very complicated articles and tweets like these:
But that was 3 years ago. And now it looks amazing but not really striking.
Earlier editors work with outcomes and craft several versions because AI may not really understand the context and due to this may give the wrong results.
However, AI reporter is a zero-typo writer it’s still not an absolute honey cake.
But for now times have changed.
Have you heard about OpenAI?
That’s a company backed by Elon Musk and based in San-Francisco.
The system called GPT-2 is actually the text generator. It takes one word and then supposes what’s going to be next in the context. The point is - it’s self-trained.
GPT-2 is the result of the “unsupervised” approach.
How does neural network usually work?
You teach it to produce outputs feeding them with tones of carefully picked inputs. Like a whale eats plankton. And afterward research team teach it to generate needed output.
With the “unsupervised” learning method the researchers also prepare the dataset for it. But everything else is done by the neural network.
Just check it.
And it does not create gibberish. The style of language and sentences are absolutely matched with the input sentence you write.
And there are huge achievements in natural language processing. It doesn’t sound like a robot anymore.
It’s rough, atmospheric and coherent.
The Guardian exposed an experiment based on the GPT-2.
They had written an opening line of George Orwell’s Nineteen Eighty-Four:
“It was a bright cold day in April, and the clocks were striking thirteen”
...and waited 5 seconds till the system designed the text in consonance with.
And it did!
“I was in my car on my way to a new job in Seattle. I put the gas in, put the key in, and then I let it run. I just imagined what the day would be like. A hundred years from now. In 2045, I was a teacher in some school in a poor part of rural China. I started with Chinese history and history of science.”
The system got a slightly futuristic tone and crafted a story based on it. And.. it catches the mindflow!
And that’s where the point of concern is dig. The researchers don’t provide open access to this technology, because they are concerned about the misuse of it.
First of all, in the wrong hands, it can be used as a fake-news turbine.
You may think “Come on, it’s 21st century, yet there has never been a time the information was so easy-to-get and easy-to-check”.
And still, there are people who may believe news they don’t know are fake. And will spread it. And cause panic.
But it can be as well a perfect tool for detecting fake news. Fire with fire!
Because it knows its own mannerism and can distinguish human-written from machine-written news.
But still, AI is not a threat to human writers. And will never be because journalists and storytelling are about critical thinking and judgment, first and foremost. Humans write for other humans and that’s it.
But it’s a big helper for true journalism. While AI will help with data work, human journalists will do a high-value job based on curiosity and digging into facts, personas, and world.
And how sensitive AI is to mistakes, the imperfections are what make us real.