This year’s Global Editor’s Network (GEN) Summit, in Athens, Greece, had a big focus on the use of verification and automation. BBC News data scientist and PGCert Data Journalism student Alison Benjamin went along to see what was being said about artificial intelligence (AI), data and technology in the news industry. Here are her highlights…
1. John Micklethwait: Three categories of fake
Like many speakers, editor-in-chief of Bloomberg John Micklethwait identified fake news as a clear danger of AI. A particular highlight was how Micklethwait’s discussion of the term, popularised by BuzzFeed’s Craig Silverman, divided fakes into three categories:
- State-sponsored fake news;
- Moronic news that is plainly false; and
- ‘Fake news’ as a pejorative label for commentary that people disagree with.
Watch the talk in full in the video above.
2. Katharine Viner: How date stamps enable transparency
Kath Viner, editor-in-chief of The Guardian, described the organisation’s recent effort to foil misinformation by including the year of publication of articles when they are shared on social media.
She said partisan groups had been sharing old news stories published by The Guardian — as though they’d just happened — in order to push their agendas on social media.
Viner praised her digital team for implementing a technical fix so quickly in the context of The Guardian’s slim financial resources, noting that the paper has made its first operating profit in the last 10 years.
She called on tech companies to take similar steps to combat fake news.
3. Krishna Bharat: Authenticity in online comments
Bots, manipulators and marketers all pose a threat to informed discussion of news, according to Google News founder Krishna Bharat, who told the conference that fake and misleading comments were harming civil discourse and democracy itself.
How can a reader know who’s “behind the mask” of a comment online? What credibility should a reader afford to a point of view? Bharat asked how AI research, originating in tech companies, could address these problems.
Bharat also argued that the data that tech companies hold on audience members could be used to “empower” audience members, by giving them the chance to authenticate who they really were:
“I carry around a phone with me all the time. The accounts I have with Amazon, with Google, with Facebook have been there for decades…
“So surely, they know me. In fact, they do. And I should be able to assert that I’m a real person. Right? They know my name. They know my location. They know my gender. They know my interests. And beyond that they know a lot more that they could compute using AI models [e.g. through inferring my location, friends, work, religion] if they wanted to.”
I asked Bharat how consent would work in such a system.
Bharat’s answer focused on how data could be used to triangulate and infer whether a person was likely to be real.
And he described a partnership between tech companies and publishers, where the former enable disclosure, and the user is asked to make a choice over what to reveal. He added:
“And there are number of ways to implement this. One way is to have the platform already offer this, by saying ‘this is what they bring to the comment’, ‘hey, we worked with Facebook, Google, Amazon, and between them, they have authenticated you as a real person, as somebody who’s a woman, who’s pro-choice, or whatever, right?’.
“[We can then ask] Would you like to take these badges and make them apart of your comment?
“And you can say, no I don’t want to, or yes I do…
“From a publisher perspective, the idea would be that they would partner with one or more tech companies to produce a capability that allows the audience to assert that they’re true and so forth.
“Now this could also be automatically considered in [comment] rank and that raises a host of challenges. If you want to do this comprehensively, then you’re actually discriminating against some users who just didn’t happen to have the badges, and you’ve got to weigh that.”
Many speakers also touched on AI’s potential for enhancing the journalism process.
One technology that came up again and again was speech recognition and translation software: it provides transcription that speeds the process a writer takes in parsing what was discussed in an interview.
Krishna Bharat envisaged world where journalists get superhuman powers through “Reporter’s Notebook 2.0” technology – which might include real time transcription, fact-checking, and optical character recognition of written text.
Bharat is describing a journalism context where intelligent agents augment some of the drudge work of reporting, but also enable content and concept discovery when data are linked.
4. Monique White (and others): Content mining and automated coverage of repeated events
Another common theme from speakers was how repetitive events or tasks lend themselves to automation (I was reminded of my colleagues in BBC News Labs, who have been experimenting with how generating “semi-automated journalism” could increase local coverage, using data from the NHS. This is an example of how automation can broaden the scope of content created about an issue, for large and local audiences).
For Bloomberg’s John Micklethwait and Monique White, automation has a role to place in verifying and producing coverage events that occur regularly – like corporate earnings reports.
The extraction of structured information from unstructured input sources, like regulatory filings, as White explained, can then be used to create alerts for journalists.
At GEN, tech wasn’t really discussed in dualistic, good/evil terms. In general, I heard AI heralded as a means to an end; or as augmentation to enhance journalism.
Those fearful that robots will steal their jobs will take comfort in the analysis Micklethwait left his GEN audience with:
“The value of what we do is often now trying to explain things, trying to predict things, trying to research and find things, and that’s where the value of “shoe leather” has reporters going out and digging things up comes through.
“And it’s true with commentary and analysis. You cannot automate Martin Wolf of The Financial Times. You can’t automate the journalists of The Economist or The Atlantic, and editors also matter.”