Guardian Australia has just published its first automated article but reporters fear not – you’re still needed for a while yet
On Friday morning Guardian Australia published its first news story written by an automated system called ReporterMate. It’s a program that takes a dataset and a story template file then turns it into a news story without much human intervention.
So why are we doing this? And, as my colleagues keep joking – won’t it make reporters redundant?
These days news organisations have fewer staff but still need to cover a large amount of news. And, despite having less time, journalists often have to write formulaic stories. Such as covering the weather – how hot is it? Is it a new record? Or recurring stories like politician’s expenses and political donations. Who spent the most? Who are the top five? Were donations more or less than last year?
These types of stories also often require some mathematical analysis, which can tie up other resources in the newsroom. Journalists would prefer to be spending time on deeper, more meaningful investigations, such as why the climate is changing, and how does it affect people? Or which politicians are abusing the expenses system?
So, with a grant from the Walkley Foundation, I built a system that can automate the analysis and writing of these formulaic stories. Data in, news story out. It joins an older project of mine, DisclosureBot, in the small field of automation in journalism.
This isn’t an original idea (a great example is the LA Times’ Quakebot) and other companies have been doing it for several years now. But these are either proprietary commercial services or built for specific types of data. Two of the more prominent companies in this area are Narrative Science and Automated Insights.
If the news media want to control how technology like this is used and how it affects their industry, it’s far better to construct an open-source system that can be used by any media organisation. It’s not going to make anyone redundant any time soon – ReporterMate can only provide coverage for certain types of stories, and the coverage it provides is limited to templated summaries.
It’s possible that more sophisticated techniques will soon be available, such as taking the template out of the system and relying on machine learning to determine which data and analyses are likely to be most interesting. But even then I think it’ll be a fair while before we have a system with artificial intelligence that can write an article from scratch – reporters will still be needed for a while yet!