Future News 22: The New York Times' early AI explorations

Can you imagine life without electricity? The device you're reading this article on, for starters, would be made obsolete and your kitchen would quickly become a graveyard of dead appliances. Work, too, would be profoundly impacted as would most people's method of commuting since all forms of modern transportation use electricity to some degree or another.

Don't worry, they're not going to turn off the grid yet. But it is worth running this thought experiment through your head because artificial intelligence (AI) and, more specifically, machine learning advocates and educators, most notably Stanford University's Andrew Ng (@AndrewYNG), have described AI as the 'new electricity'. The analogy, although imperfect, is helpful in that it counters some of the hype and fear  killer robots and mass job losses, for example –  surrounding the technology.

A recent study co-authored by Princeton University's Matt Salganik (@msalganik) and 111 other academics also poured cold water over the machine learning excitement. They used 160 research teams on high-quality data sets to predict six life outcomes for children, parents, and households only to discover that even the best AI predictive models weren't very accurate. "These results show us that machine learning isn't magic; there are clearly other factors at play when it comes to predicting the life course," Salganik said.

Machine learning, then, could dramatically enhance and compliment newsgathering and the consumer experience, rather than making journalists obsolete. Some media outlets (see below) have already used the technology, with supervised learning algorithms, where developers provide input and output patterns to the system, proving most popular.

AI journalism case studies 
The New York Times, meanwhile, is adopting a cautious, measured and rigorous approach to AI. There are many newsgathering-related tasks that could be easily automated, including transcription, fact-checking and even story selection (it is a matter of time before machine learning is combined with ChartBeat, for instance), but the publication's R&D strategist Amelia Pisapia (@ameliapisapia) stressed to me that her team's mantra was to "figure out how to apply emerging technologies in the service of journalism". They are not, in other words, looking to replace a hard-nosed hack with an AI.

It's early-days yet since the R&D team typically first identifies an area of interest, researches it to see if it has a use case, and then launches a prototype. If this shows "potential", further resources may be dedicated to exploring the idea. An example would be The Times' collaboration with Verizon's 5G journalism lab or the publication's partnership with New York University's Media Lab to explore the opportunities of spatial computing for journalism. 

Machine learning is still in its research phase at the outlet, although some areas of interest, including natural language processing (think of Amazon's Alexa) and computer vision (driverless car technology, for example), have been identified by developer Or Fleisher (@JuniorXSound). However, he has found that many algorithms are generic and he has had to "zoom in" on their potential capabilities for The Times and re-train or fine-tune them. 

"We are trying to think about how we can build capability...instead of replacing people who are already doing the work on the ground," he said. "How can machine learning be used in situations where the human is in the loop and AI model is providing interesting insights to editors and reporters, instead of doing what they are already very good at." 

There is, of course, still some ethical quandaries around machine learning even if it does not become a job killer. Removing biases from data sets, which effectively provide the fuel to run the algorithms, is something that Fleisher is clearly conscious of. 

“There is a big responsibility question to these questions and we don’t take it lightly," he explained. "The reason why a lot of these projects are experiments and under research is because we try to be very rigorous with ourselves and to hold ourselves accountable on the decisions that are implemented in these solutions.” 

As to when AI would be up and running in The Times' newsroom, Pisapia and Fleisher were non-committal on timelines. I also asked Alexa, who was useless (again) on this matter.

Credit: RawPixel


The latest research from the World Economic Forum points towards a worrying phenomenon: information inequality. The study, among other things, found that low-income groups are less likely to pay for journalism. As pay-walls are erected, this could leave those in the working classes and lower middle classes at a considerable disadvantage compared to those in other socio-economic groups. The report, however, did disclose some positives in that young people (18 to 35-year-olds) in Germany, the UK and the US were found to be twice as likely as over-55s to buy news.

Staying on the subject of paywalls, a Canadian fast food restaurant has partnered with Post Media (National Post and others) so that the group's publications are free-to-read until the end of April.

As misinformation continues to spread about the Covid-19 outbreak, Google has committed $6.5m for fact-checking outlets, including FirstDraft and UK-based Full Fact. The crisis and subsequent lockdowns, meanwhile, are putting immense stress on local and regional print publications across the West, potentially leading to hundreds, if not thousands, of news deserts. In Australia, News Corp has suspended print editions of 60 local publications.

News deserts and information inequality could have an impact on the 2020 White House election, which, let's not forget, is still pencilled in for November. Pew Research's Elizabeth Grieco (@Grieco001) has published new analysis on the consumption of political news in the US, outlining a deep partisan divide.

For high praise, tips and complaints, please email the editor at