Meet the robot journalist creating specialist content

As Carmen gets to work on her 98th investment story (averaging 50 a week), there is no company meeting to attend, no team bonding to avoid and zero reasons for leaving early – although, admittedly, the last is a moot point currently.

Carmen is a robot, or rather an algorithmic journalist, and an extremely valued member of the editorial team.

By using artificial intelligence (AI), combined with human innovation, Carmen – or to give her full name: Content Automated by Robots for the Monitor E-publishing Network – is able to produce fact-based articles, freeing up human journalists to interpret, analyse and explain developments.

And facts and figures are at the heart of why AI is so extremely useful in specialised content.

In Carmen’s case, writing for Investment Monitor – a data-driven website examining the corporate dynamics shaping the foreign investment landscape – makes perfect sense. For companies seeking to invest in countries, regions and cities, reliable data will help inform their decisions.

So with Carmen spinning out guides to investing in Ecuador to Morocco, Jordan to France, replete with accurate financial and socioeconomic reporting, this resource becomes invaluable.

Where the story begun

The term ‘AI’ can evoke a range of human emotions, from the fear of robotic warriors running amok to the amazement of entering (and leaving) Amazon’s first Just Walk Out store in London without paying.

In real terms, AI in publishing is simply a machine’s capabilities to exhibit human-like performance at a defined task. It is also important to remember that the technology is only as good as the humans behind the algorithms.

That is not to say that when AI combines with processes such as machine learning – the ability of a machine to improve automatically through experience – it will be taken to another level.

After all, in 2016 an AI-written novel aptly named The Day A Computer Writes A Novel passed through the first round of the Nikkei Hoshi Shinichi Literary Awards, although further reading is required to understand how much human involvement allowed this to happen.

But for now, publishers since 2019 have been using AI to help create reports that would normally be extremely time-consuming for their human counterparts.

An AI journalist – named Cyborg – accounted for one-third of the specialised financial content on Bloomberg News, while similar programmes produced basic sports reports for The Associated Press and The Washington Post.

The key in AI’s usefulness is the sheer amount of data required to sift through to generate all this content, and the speed and accuracy at which it is constructed.

Taking Carmen as an example, to produce a country-based article, you need to research population numbers, gender-divides, employment figures, GDP trends, and education, business and healthcare figures.

Easy for an algorithmic journalist, frustrating for a human journalist.

The nuts and bolts

So how exactly did Carmen come to live and work? We spoke to Kamen Parushev, Data Product Manager at New Statesman Media Group and part of the data engineering team behind their push into AI-based journalism.

“Carmen is an experimental auto reporter who uses our powerful datasets to produce content which complements the work of its human colleagues.

“Аt this point, Carmen uses the data available in our databases to enrich text with information. It uses specific tags inside the text to determine which information to pull, transform and insert in our articles.”

So far, so good. By giving artificial intelligence fixed tasks, drawing from a vast array of available information, you can benefit from its speed and precision in filling in the gaps.

Remarkably – and perhaps tellingly for the future of AI in publishing – the proof of concept took just two weeks and, allowing for development, Carmen was deployed and writing within a month and a half.

The future

Carmen’s next assignment is already briefed: Reporting on the Covid-19 economic recovery in the G20 countries.

And according to Parushev, this is just the start.

“We are already working on Carmen 2.0. Our goal is to provide even more flexibility to our editorial teams.

“Carmen 2.0 will be able to automatically update content directly on our websites, apply advanced data transformation functions, calculate rankings, and operate with free text input.”

Currently, every article written by Carmen requires a human to hit publish, but such is the trust in the technology, the next incarnation will be able to refresh live content with the latest data – meaning the article is constantly timely.

Within a year, Carmen 2.0 will move onto imagery, video and audio.

“By using AI, natural language processing and machine learning techniques on a user friendly platform, our editorial teams get more time to focus on the vital niche articles being written.

“We want to create a web interface for automated content production based on modular building blocks. No matter what type of content you’re using – text, data, visualisations, video or audio – we can automatically generate that content at scale.”

For AI in journalism, humans do – and will continue to – dictate the scope, alongside the datasets available. We are not in a position for independent AI research or full creativity, and this suits publishers and platform owners.

Using AI to free up time for talented reporters and writers to do what they do best is a major breakthrough in itself.