“How do we bring people together to agree on facts when people don’t want to receive information that doesn’t fit with how they see the world?”
The impact of fake news, propaganda and misinformation has been widely scrutinized since the US election. Fake news actually outperformed real news on Facebook during the final weeks of the election campaign, according to an analysis by Buzzfeed, and even outgoing president Barack Obama has expressed his concerns.
But a growing cadre of technologists, academics and media experts are now beginning the quixotic process of trying to think up solutions to the problem, starting with a rambling 100+ page open Google document set up by Upworthy founder Eli Pariser.
The project has snowballed since Pariser started it on 17 November, with contributors putting forward myriad solutions, he said. “It’s a really wonderful thing to watch as it grows,” Pariser said. “We were talking about how design shapes how people interact. Kind of inadvertently this turned into this place where you had thousands of people collaborating together in this beautiful way.”
In Silicon Valley, meanwhile, some programmers have been batting solutions back and forth on Hacker News, a discussion board about computing run by the startup incubator Y Combinator. Some ideas are more realistic than others.
“The biggest challenge is who wants to be the arbiter of truth and what truth is,” said Claire Wardle, research director for the Tow Center for Digital Journalism at Columbia University. “The way that people receive information now is increasingly via social networks, so any solution that anybody comes up with, the social networks have to be on board.”
Journalists, the public or algorithms?
Most of the solutions fall into three general categories: the hiring of human editors; crowdsourcing, and technological or algorithmic solutions.
Human editing relies on a trained professional to assess a news article before it enters the news stream. Its proponents say that human judgment is more reliable than algorithms, which can be gamed by trolls and arguably less nuanced when faced with complex editorial decisions; Facebook’s algorithmic system famously botched the Vietnam photo debacle.
Yet hiring people – especially the number needed to deal with Facebook’s volume of content – is expensive, and it may be hard for them to act quickly. The social network ecosystem is enormous, and Wardle says that any human solution would be next to impossible to scale. Humans are also partial to subjectivity, and even an overarching “readers’ editor”, if Facebook appointed one, would be a disproportionately powerful position and open to abuse.
Crowdsourced vetting would open up the assessment process to the body politic, having people apply for a sort of “verified news checker” status and then allowing them to rank news as they see it. This isn’t dissimilar to the way Wikipedia works, and could be more democratic than a small team of paid staff. It would be less likely to be accused of bias or censorship because anyone could theoretically join, but could also be easier to game by people promoting fake or biased news, or using automated systems to promote clickbait for advertising revenue.
Algorithmic or machine learning vetting is the third approach, and the one currently favored by Facebook, who fired their human trending news team and replaced them with an algorithm earlier in 2016. But the current systems are failing to identify and downgrade hoax news or distinguish satire from real stories; Facebook’s algorithm started spitting out fake news almost immediately after its inception.
Technology companies like to claim that algorithms are free of personal bias, yet they inevitably reflect the subjective decisions of those who designed them, and journalistic integrity is not a priority for engineers.
Algorithms also happen to be cheaper and easier to manage than human beings, but an algorithmic solution, Wardle said, must be transparent. “We have to say: here’s the way the machine can make this easier for you.”
How to treat fake news, exaggeration and satire on Facebook
Facebook has been slow to admit it has a problem with misinformation on its news feed, which is seen by 1.18 billion people every day. It has had several false starts on systems, both automated and using human editors, that inform how news appears on its feed. Pariser’s project details a few ways to start:
Verified news media pages
Similar to Twitter’s “blue tick” system, verification would mean that a news organization would have to apply to be verified and be proved to be a credible news source so that stories would be published with a “verified” flag. Verification could also mean higher priority in newsfeed algorithms, while repeatedly posting fake news would mean losing verified status.
Pros: The system would be simple to impose, possibly through a browser plug-in, and is likely to appeal to most major publications.
Cons: It would require extra staff to assess applications and maintain the system, could be open to accusations of bias if not carefully managed and could discriminate against younger, less established news sites.
Separate news articles from shared personal information
“Social media sharing of news articles/opinion subtly shifts the ownership of the opinion from the author to the ‘sharer’,” Amanda Harris, a contributor to Pariser’s project, wrote. “By shifting the conversation about the article to the third person, it starts in a much better place: ‘the author is wrong’ is less aggressive than ‘you are wrong’.”
Pros: Easy and cheap to implement.
Cons: The effect may be too subtle and not actually solve the problem.
Add a ‘fake news’ flag
Labelling problematic articles in this way would show Facebook users that there is some question over the veracity of an article. It could be structured the same way as abuse reports currently are; users can “flag” a story as fake, and if enough users do so then readers would see a warning box that “multiple users have marked this story as fake” before they could click through.
Pros: Flagging is cheap, easy to do and requires very little change. It would make readers more questioning about the content they read and share, and also slightly raises the bar for sharing fake news by slowing the speed at which it can spread.
Cons: It’s unknown whether flagging would actually change people’s behavior. It is also vulnerable to trolling or gaming the system; users could spam real articles with fake tags, known as a “false flag” operation.
Add a time-delay on re-shares
Articles on Facebook and Twitter could be subject to a time-delay once they reach a certain threshold of shares, while “white-labeled” sites like the New York Times would be exempt from this.
Pros: Would slow the spread of fake news.
Cons: Could affect real news as much as fake, and “white-labelling” would be attacked as biased and unfair, especially on the right. Users could also be frustrated by the enforced delay: “I want to share when I want to share.”
Partnership with fact-checking sites, such as Snopes
Fake news could automatically be tagged with a link to an article debugging it on Snopes, though inevitably that will leave Facebook open to criticism if the debunking site is attacked as having a political bias.
Pros: Would allow for easy flagging of fake news, and also raise awareness of fact-checking sources and processes.
Cons: Could be open to accusations of political bias, and the mission might also creep: would it extend to statements on politicians’ pages?
Headline and content analysis
An algorithm could analyze the content and headline of news to flag signs that it contains fake news. The content of the article could be checked for legitimate sourcing – hyperlinks to the Associated Press or other whitelisted media organizations.
Pros: Cheap, and easily amalgamated into existing algorithms.
Cons: An automated system could allow real news to fall through the cracks.
This system would algorithmically promote non-partisan news, by checking stories against a heat-map of political opinion or sharing nodes, and then promoting those stories that are shared more widely than by just one part of the political spectrum. It could be augmented with a keyword search against a database of language most likely to be used by people on the left or the right.
Pros: Cheap, and easily combined with existing algorithms. Can be used in partnership with other measures. It’s also a gentler system that could be used to “nudge” readers away from fake news without censoring.
Cons: Doesn’t completely remove fake news.
Sharer reputation ranking
This would promote or hide articles based on the reputation of the sharer. Each person on a social network would have a score (public or private) based on feedback from the news they share.
Pros: Easy to populate a system quickly using user feedback.
Cons: User feedback systems are easy to game, so fake news could easily be upvoted as true by people who want it to be true, messing up the algorithm.
Visible design cues for fake news
Fake news would come up in the news feed as red, real news as green, satire as orange.
Pros: Gives immediate visual shorthand to distinguish real from fake news. Could also be a browser plug-in.
Cons: Still requires a way to distinguish one from the other, whether labor-intensively or algorithmically. Any mistake with an algorithm, say one that puts Breitbart articles in red, would open Facebook up to accusations of bias.
Punish accounts that post fake news
If publishing fake news was punishable with bans on Facebook then it would disincentivise organizations from doing so.
Pros: Attacks the problem at its root and could get rid of the worst offenders.
Cons: The system would be open to accusations of bias. And what about satire, or news that’s not outright fake but controversial?
Tackling fake news on the web outside Facebook
News is shared across hundreds of other sites and services, from SMS and messaging apps such as WhatsApp and Snapchat, to distribution through Google’s search engine and aggregations sites like Flipboard. How can fake news, inaccurate stories and unacknowledged satire be identified in so many different contexts?
A central fact-checking service could publish an API, a constantly updated feed of information, which any browser could query news articles against. A combination of human editing and algorithms would return information about the news story and its URL, including whether it is likely to be fake (if it came from a known click-farm site) or genuine. Stories would be “fingerprinted” in the same way as advertising software.
People could choose their fact-checking system – Snopes or Politifact or similar – and then install it as either a browser plug-in or a Facebook or Twitter plug-in that would colour-code news sources on the fly as either fake, real or various gradations in between.
Pros: Human editors would become less necessary as the algorithm learns, and wouldn’t have to check each story individually. Being asked to choose a fact-checker might encourage critical thinking.
Cons: Will be labor-intensive and expensive, especially at first. It could be open to accusations of bias, especially once the algorithm takes over from the human input. Arguably only those already awake to the problem would choose to opt in, unless a platform like Facebook or Google assimilates it as standard.
Page ranking system
Much like Google’s original PageRank algorithm, a system could be developed to assess the authority of a story by its domain and URL history, suggested Mike Sukmanowsky of Parse.ly.
This would effectively be, Sukmanowsky wrote, a source reliability algorithm that calculated a “basic decency score” for online content that pages like Facebook could use to inform their trending topic algorithms. There could also be “ratings agencies” for media; too many Stephen Glass-style falsified reporting scandals, for example, and the New York Times could risk losing its triple-A rating.
Pros: Relatively easy to construct using open-sourcing, and could be incorporated into existing structures. Domains that serially propagate fake information could be punished by being downgraded in rank, effectively hiding them.
Cons: Little recourse for sites to appeal against their ranking, and could make it unfairly difficult for less established sites to break through.
Connect fake news to fact-checking sites
Under this system, fake news would be inter-linked (possibly through a browser plug-in) to a story by a trusted fact-checking organization like Snopes or Politifact. (Rbutr already does this, though on a modest scale.)
Pros: Connects readers with corrections that already exist. Facebook or Google could use a database like Snopes in its algorithm.
Cons: Unless this kind of system gets hardwired into Facebook or Google, people have to want to know if what they’re reading is fake.
On current evidence, many people feel comfortable when presented by news which doesn’t challenge their own prejudices and preferences – even if that news is inaccurate, misleading or false.
What many of these solutions don’t address is the more complex, nuanced and long-term challenge of educating the public about the importance of informed debate – and why properly considering an accurate, rational and compelling viewpoint from the other side of the fence is an essential part of the democratic process.
“There’s a feeling that in trying to come up with solutions we risk a boomerang effect that the more we’re debunking, the more people will disbelieve it,” said Claire Wardle. “How do we bring people together to agree on facts when people don’t want to receive information that doesn’t fit with how they see the world?”
• Jasper Jackson contributed to this report
guardian.co.uk © Guardian News & Media Limited 2010