AP set for artificial intelligence breakthrough in verifying user-generated video

Its ubiquity in news media – informing thousands of clients, including newspapers, broadcasters and government agencies – should be a source of reassurance when media trust is everywhere under threat. Yet the AP, despite a 173-year history dating to the Mexican-American War, has a low public profile. “People don’t always know who the AP is, even though we provide so much information,” concedes Anna Johnson, newly-appointed news director for Europe and Africa.

The AP’s role in fortifying public faith in news is especially vital in the area of user-generated mobile phone video, which so often provides the most visceral – and potentially controversial – elements of the modern breaking news report.

Johnson reveals that the AP is set to introduce to its newsroom a new cloud-based tool that uses artificial intelligence to instantly verify – or expose as fake – the thousands of UGC videos examined by AP journalists every week. The AP Verify programme combines visual recognition and machine learning technologies to make snap assessments of UGC. The tool has been in testing phase since 2017 when it was approved for research funding as part of the Google-supported Digital News Initiative.

“We hope later this year to be able to start using it,” says Johnson, who took up post late last month at AP’s large London hub, where the AP Verify team is based. “It does those basic things like making sure things match up… that a street sign would [actually] be in that place.”

The value of UGC was highlighted during the making of ‘The Missing’, an extraordinary AP project that has logged 61,135 migrants who have died or disappeared since 2014, simply in the course of travelling from one place to another. In one of the most disturbing pieces in this global series, AP reported on the expulsion by Algeria of 13,000 sub-Saharan African migrants who were cast into the burning desert and left to “walk or die”. The story’s veracity was supported by UGC footage taken by Liberian migrant, Ju Dennis, who had hidden a mobile phone on his body and filmed his fellow sufferers as they went on their journey.

“It’s hard to watch but it authenticates something that we couldn’t physically see ourselves,” says Johnson. “We couldn’t be on the Algeria border to witness this, it’s not like the government was welcoming the media to see them do this. So, we talked to [Dennis] and interviewed him and the UGC added something to our story which was already well-reported and compelling.”

Being able to apply AP Verify’s AI to UGC clips can speed up the reporting process by rescuing journalists from sometimes overwhelming amounts of UGC, Johnson believes. “It will cut out stuff that is just noise, that’s not worth looking into. We are looking for innovative ways to be faster, more streamlined and more efficient so our journalists can really focus on reporting and say this clearly is manipulated video.”

While accuracy is vital to AP’s reputation, speed is also of crucial importance. Partly because social media doesn’t wait and can create dangerous bubbles where UGC and rumour spreads, unqualified by reliable interpretation. But also because AP’s clients expect the agency to have the story first when it comes to breaking news.

Going “really big on the big stories” is Johnson’s number one priority. “It’s about making sure that we are first to confirm something, first to have the live shot, first to get UGC,” she says. “It’s keeping to our accuracy statements but also being quick and looking for new ways to do things to be faster and more efficient.”

Johnson arrives in London from Arizona, where she was AP’s news director for the western United States. She recently helped oversee the agency’s coverage of the US midterm elections and wants to bring AP’s latest expertise in detecting political misinformation to its coverage of the European elections in May.

AP last month investigated the impact of fake news on Nigeria’s elections and how false stories had circulated to the degree that voters were told that President Muhammad Buhari had died and been replaced by a Sudanese body double.

Whereas many big US news organisations caught a crab over their 2016 presidential election coverage by failing to predict the rise of Donald Trump or to detect rising dissatisfaction in middle America, AP largely escaped such criticisms. The agency has staffers covering every one of the 50 US state legislatures and produced a pre-election series ‘Divided America’, highlighting the polarisation of voters over issues such as the economy and immigration.

Johnson says that lack of political bias is fundamental to the agency’s credibility. “I always say that when you are an AP reporter and you walk into a room, it should be a mystery what you believe.”

AP journalists must keep their opinions to themselves on social media and only deal in facts. “It’s really important what you say on social media,” says Johnson. “It might never come back to you but it might hurt one of your staff members who is trying to cover that issue – someone could say the AP has got an opinion. We don’t have commentators and we don’t do editorials.”

That doesn’t help to raise AP’s profile but the agency is more public-facing than it once was. It has its own mobile app and its correspondents, such as Vatican specialist Nicole Winfield, for example, are encouraged to give interviews to other news organisations. “We have a lot of journalists with true expertise and we want them to share that knowledge,” says Johnson.

Anonymity is also resisted within the AP’s content, where the use of unnamed sources is discouraged. Johnson says there is “a lot of scepticism” from the public around the use of anonymous sourcing in the news media. “The AP has really high standards on that: [sources] have to have first hand knowledge and we have to know how they know. We have to describe why the person wouldn’t speak – that they could potentially lose their job or that there was some danger to them.”

In her role overseeing AP staff in more than 60 European and African cities, she wants to increase the amount of transparency in reporting. This ranges from being explicit in telling clients that the AP is investigating an incident but unable to confirm what has happened, through to providing customers with raw source material on a story. “Unless there are legal reasons, we will give you the documents,” she says.

When AP uses UGC film it must track down the person who actually shot it and obtain permission to republish it. The AP Verify project, which is being led by Paul Shanley, AP director of international development and partnerships, is a first stage in this authentication process and cuts out time-consuming tasks without removing the need for editorial judgment.

“We are excited about the potential,” says Johnson of AP Verify. “It uses the latest machine learning technologies to search social media platforms for the original source of the UGC. It also can use image recognition technology such as signs and text to pull out important pieces of information from a video to help us understand whether the video is accurate.”

With the growing potential for ‘deep fake’ videos (combining artificial intelligence with actual shots of a person to create something bogus), Johnson admits it is “getting easier for manipulated content to look more real”.

But AI technology can equally be used for good – by building public trust in AP content, and thus in the news media as a whole.

Read more from Ian Burrell's column, The News Business, and follow him on Twitter @iburrell.