New tool helps stand guard against fake news
A high-stakes federal election comes up in less than a week, and on the Internet the digital lies are flowing as freely as ever. I wrote recently about scientists developing software to detect fake news stories, but their programs still aren’t ready to go.
Steven Brill isn’t waiting.
Brill, cofounder of American Lawyer magazine, has joined forces with former Wall Street Journal publisher Gordon Crovitz to develop NewsGuard, a browser-based tool that tries to fend off fake news using human brains, rather than silicon chips.
“We are based on the proposition that sometimes human intelligence is better than the artificial kind,” Brill said.
There’s no way human reviewers can examine the thousands of news stories flushed into the cloud every hour of every day. So NewsGuard doesn’t even try. Instead, it vets the sources of news — the websites that serve up the stories.
“It was clear to us that a lot of the fake news stories come from a very limited number of websites,” said Crovitz, who figures only about 2,500 news sources provide nearly all of the American online news diet. NewsGuard has gone to school on about 1,500 so far.
It’s a smart approach, though not entirely original. In 2015, the independent media critic Dave Van Zandt launched Media Bias/Fact Check, which posts bias and accuracy rankings for roughly 2,000 news sites. Visitors can find out if a site offers reliable information and whether to expect a left- or a right-wing slant.
Media Bias has little money compared to NewsGuard, which raised $6 million from outfits such as Publicis Group and the John S. & James L. Knight Foundation.
And Media Bias’s focus on political bias seems like a sideshow. It’s good to know that Mother Jones is a left-wing magazine. But it’s better to know if it’s consistently accurate, or consistently wrong.
NewsGuard’s two dozen researchers do not test for political slant. Instead they’re probing the integrity of news stories, and of the organizations that produce them.
NewsGuard uses nine measures, including whether stories from a site are consistently accurate, and whether errors are corrected. Is there a clear distinction between news and opinion? Are all advertisements clearly labeled as such? What are the site’s sources of funding?
It’s old-school shoe-leather reporting, with NewsGuard researchers using prominent fact-checking sites such as PolitiFact and interviewing the operators of the news sites. Try doing that with an artificial intelligence program.
The results of each test are weighted and combined to make up a final score of up to 100. Any source with a ranking below 60 gets tabbed as questionable, and NewsGuard’s meter will flash red for readers.
Respected publications like the leftist Nation and the right-wing National Review both score high on the NewsGuard scorecard. But it finds that rabble-rousing liberal Daily Kos runs too many stories with deceptive headlines and posts content from anonymous writers, giving it an overall rating of 52. Then there’s the conservative Breitbart, which gets lousy grades (57) for allegedly running many inaccurate stories, and for refusing to provide NewsGuard with information about how the company is financed.
Since new sites pop up all the time, NewsGuard deploys a “SWAT” team to evaluate them, if their stories begin trending. Indeed, the company recently flagged several new sites that appeared to be in the United States and were full of biased and inaccurate stories. NewsGuard determined they were based overseas — apparently in Macedonia and Australia.
To get NewsGuard’s ratings of your favorite news sites, you need a free software extension compatible with the leading Internet browsers. It installs in seconds and adds a little shield-shaped icon to the upper right corner of your browser’s toolbar.
Now, say you go to The Boston Globe website. The shield will glow a friendly green. Click on the shield and you’ll get a message: “This website generally maintains basic standards of accuracy and accountability.” The “Nutrition Report” is a one-page backgrounder about the news provider, including its owner, how it makes money, its editorial positions — even embarrassing stories. The Globe, by the way, scored a perfect 100 on the NewsGuard report.
But visit, say, the National Enquirer, and the NewsGuard icon turns bright red, warning that the tabloid “generally fails to maintain basic standards of accuracy and accountability.” The Enquirer earns a 20.
The NewsGuard extension also works within Facebook, the world’s favorite place to share news stories — whether they’re true or not. The red or green icon shows up next to stories in your news feed, indicating whether they came from trustworthy sources.
The NewsGuard system is devoid of bewildering algorithms or AI jargon. It’s just an automated version of common sense. Brill believes that’s the way to defeat fake news, arguing that the computer scientists haven’t got a chance.
“They’ve been saying they were doing that for two years,” he said, “and its not doable.”
And even if it could be done, automated fake-news filters suffer from a lack of transparency. The software might decide that a story is phony, but it probably couldn’t explain the decision to a human, or if it could, its answer would probably be incomprehensible.
For instance, scientists have said stories with a large number of capitalized words and exclamation points are more likely to be fake. But to a human being this sounds silly. Why disbelieve a story just because it includes five exclamation points? Many people would discard the fake news filter, not the fake story.
Because it relies on human reasoning, NewsGuard has the huge advantage of making sense.
Brill and Crovitz plan to make a profit, as well, by licensing their system to Internet companies. Microsoft has already agreed to make NewsGuard a built-in feature in future products, and Brill said he’s in talks with other online titans. The goal is to have NewsGuard running by default on our computers and phones whenever we scan the Web for news.