“Collaborating with algorithms” – notes from our panel session at #ijf17

I’m at the International Journalism Festival in Perugia, and appeared on a panel entitled “Collaborating with algorithms” alongside Greg Barber, Nicholas Diakopoulos, Christina Elmer and Jacqui Maher. You should be able to watch a reply of it here, and these are the notes I made at the time…

“More than zero proposals of marriage” – Jacqui Maher, Condé Nast

Jacqui Maher was talking about her experience of making a chatbot for British Vogue to support their coverage of fashion weeks in February. One of the things they learned is the extent to which humans will chat to the bot as if it is a fellow person. They hadn’t anticipated that people would say ‘thank you’ or just a ‘hi!’ to the bot, which initially responded with ‘Sorry, I don’t understand that’. She said it was essential that you can adapt any bot you build based on what people are saying, and suggested setting up an alert system that lets you gather all the things the bot didn’t understand so that you can refine the response.

Jacqui expressed a common frustration when trying to build services on third-party platforms. Just after the fashion weeks finished, new functionality was released into the Facebook Messenger API, which would have been ideal to use. Also, Jacqui revealed, the bot had received “more than zero proposals of marriage” from users.

Condé Nast have also, Jacqui said, been experimenting with using algorithms to make the most of their archive and to provide ways of content discovery. It’s difficult, she said, with a magazine published across different editions and different continents, for each of the Vogues to know what the other are doing. They are also using machine processing to go through their picture archive, to get a handle on over 100 years worth of fashion photography, using computers to work out what colours are in a photo, what objects etc.

“They feel the project belongs to them” – Christina Elmer, Spiegel Online

Christina Elmer is head of data journalism Spiegel Online, and she showed some examples of how they have integrated data automatically into their sports coverage. They have a system which analyses data-points from the Bundesliga, and works out a player score for each game, based on about 50 different factors during the match, and weighted according to where that player performs on the pitch. She said it provided a much more objective way of scoring players, and also was one of those great projects for the data-team to work on – because once it was up and running they never have to touch it again, it just publishes out the ratings each Monday.

They also have an interactive tool that can embed a tactical analysis of a match into their articles. It measures the interactions between players on the same team, so you can see at a glance what the dominant partnerships are in any team. Christina says it was important to have the sports teams involved directly in the design and building of the tool, so that they can really feel the project belongs to them. It’s important, too, with internal tool design, to make sure that the people who are going to be using the tool get a say in how it is set up. It helps them feel really bought into the idea.

On the broader theme of computer-assisted reporting and automated reporting, Christina had some words of caution. Firstly, she said, we must remember what the end users actually want. It might be possible, for example, to automate match reports, but what the reader really values is the expert insight a sports journalist can bring that pure number-crunching can’t. She also said that it’s important to think carefully about the sources of data that you use. For weather, for example, you can almost certainly automate it and it’s fine, but for complex issues like science, where there is a lot of data available, your choice of source will have a strong impact on the output.

“How do algorithms control the most privileged screen real estate?” – Nicholas Diakopoulos, College of Journalism Univ. of Maryland

Nicholas Diakopoulos is from the College of Journalism at the University of Maryland. He was talking about how the algorithms used by companies like Google and Facebook impact on content discovery. They studied Google search results during the course of the 2016 US election campaign, most notably what was appearing in the news box when you searched for candidates. They noted that the New York Times and CNN absolutely dominated this, taking something like 50% of the spaces between them. That raised, he said, the question: were these organisations just better at SEO than the rest, or was there something else about Google’s algorithm that was giving them that privileged spot so frequently. At present, there’s no way of knowing, since the Google algorithm is the equivalent of the Colonel Sanders’ original KFC recipe – top secret.

Of course, Google is a private company, providing a service, and doesn’t have a regulatory obligation to be impartial in its display of news sources. But Nicholas feels that from a media diversity point of view this is a concern for our industry. Should search engines and social networks be forced into more transparency around the algorithms they use, in the name of fairness? Should we be systematically tracking any biases that emerge in these systems.

And Nicholas was concerned about the unintended consequences of algorithm changes. He cited, for example, Facebook saying that they were going to use the volume of conversation under a post as a positive signal that would widen organic reach. Nicholas pointed out that there are, though, some topics which news organisations cover and which serve an informational need, which by their very nature won’t lead to conversations, because they deal with subject that are taboo or that people don’t want to talk about their own experience of in public. Is it right that this information might become harder for people to discover?

We also talked about many other topics, including how computers can help in the newsroom because they are very, very good at doing very, very boring tasks very, very quickly, and touched on some of the themes I wrote about yesterday in 3 ways we’ve been looking at working with algorithms at the Guardian. You can view the panel session online.

Find all my blog posts from the 2017 International Journalism Festival.