Mark Zuckerberg just spent two days in marathon question-and-answer sessions on Capitol Hill in Washington, tackling complex topics like privacy, data protection and election security.
Facebook's CEO vowed to protect election integrity, use artificial intelligence to fight off fake news and hoaxes, and hire 10,000 more content moderators to handle the flood of disinformation.
"The most important thing that I care about right now is making sure that no one interferes in the various 2018 elections around the world," Zuckerberg said during a hearing before the Senate on Tuesday.
In that session and Wednesday's appearance before members of the House of Representatives, Zuckerberg answered a combined total of roughly 600 questions during nearly 10 hours of testimony.
Yet he never mentioned memes.
The omission is glaring because memes, which are often associated with jokes -- like a distracted boyfriend or the American Chopper guys yelling -- are also a significant driver of misinformation on Facebook. They're designed to go viral, giving them a far better chance of reaching your newsfeed thanks to a random like or share. That memes never came up, on either side, during the hearings speaks to the fact that everyone is missing the real problem.
Watch this: House members hit Zuckerberg hard on data collection, tracking
"If you look at what the Russian troll factory was doing, a lot of it was not paid posts," said Ben Nimmo, a defense and international security analyst with the Atlantic Council's Digital Forensic Research Lab. "If Facebook is only concentrating on that, then it's missing that substantial part of the equation."
Facebook, Reddit and Twitter are still struggling to fight off trolls and election interference that take advantage of the social networks. Posts from Russian troll factory the Internet Research Agency spread political chaos by stirring up emotions around divisive issues. Social networks have tried to stamp out foreign influence campaigns, but trolls have proved to be resilient, and many members of Congress referenced the threat fake news poses to this year's midterm elections.
As jokes, memes aren't meant to be believed. Nobody sees a post of pizza in the water captioned "Italian Navy" and actually takes it literally (we hope). Yet the medium has evolved into poorly Photoshopped pictures that do threaten to trip up people. One example was an image of Barack Obama seemingly giving the Presidential Medal of Freedom to various accused sex offenders, which the fact-finding website Snopes had to debunk.
"I don't know how we got here," said Kyle Stratis, a moderator of the Meme Economy subreddit, and co-creator of the Danq Exchange. "I think people see the power that these things have to be shared, and you go from satire to something that's way more sinister because it fools more people."
During earlier congressional hearings with Facebook, last November, House representatives showed several posts from Russian trolls that had appeared on the social network -- but none of them were articles. They were all memes. The most infamous post showed Hillary Clinton's head Photoshopped onto the body of Satan as the devil prepared to mix it up in a boxing match with Jesus Christ. I'm not making this up.
Though these fixes apply to news stories, they've never addressed disinformation coming from memes. It wasn't until March 29 that Facebook announced it was expanding its fact-checking program to include images and videos, which memes would fall under.
But that program is still in its infancy, compared with Facebook's other fact-checking efforts. The social network quietly ran a trial of the expansion during the special elections in Alabama, and blocked several accounts from Macedonia that were spamming Facebook with propaganda memes, said Samidh Chakrabarti, a product manager at the company.
Now Facebook is publicly rolling out a test of photo and video fact-checking, but only in France. It says it'll expand to more countries soon.
While the rest of the world waits, hoax memes were spreading on Facebook even as Zuckerberg testified on Capitol Hill.
"It is much harder for a machine learning algorithm to handle memes," said Jennifer Grygiel, a social media professor at Syracuse University.
As an example of how memes are slipping by, a post from last month claiming that Zuckerberg "invented the word BFF" and asking users to comment to see if they were hacked has garnered more than 6,000 comments. And the comments were still pouring in as Zuckerberg was answering questions in DC.
Countless memes on Hogg appeared if you searched Twitter for the hashtag #crisisactor, while a notorious edit of Gonzalez appearing to tear up the Constitution became widespread. It was a doctored image of her from the cover of Teen Vogue, ripping up a shooting target.
"Memes are easy to make, and you're not writing a 250-word article about something fake," Stratis said. "It's almost a perfect storm that makes these memes the best way to disseminate propaganda."
Facebook's meme enforcement challenge
The world's largest social network knows it has a problem with memes that spread misinformation.
Facebook has seen photos of NFL players altered to show the athletes burning the US flag, lies claiming that people could vote by sending a text, and misleading captions that turned harmless pictures into flash points for political discord.
When it comes to fact-checking images, Zuckerberg's company is relying on the same strategy it uses for news stories. If a picture is flagged as false, the company's algorithm buries it on your newsfeed, numbing its influence. If the image completely violates Facebook's community standards -- a post promoting violence, hate speech or terrorism -- moderators delete it.
Comments also help flag posts, said Facebook spokeswoman Lauren Svensson. Posts like "no way this is real!" help signal to moderators that a photo is fake.
"This initial rollout with the AFP [news service Agence France-Presse] in France is our first test of fact-checking in photos and videos," Svensson said in an email. "We know that photos and video are more nuanced, and we'll look to refine this with the AFP before we roll out this capability further."
Facebook's algorithm runs into trouble since the hoaxes aren't text-based. And the social network has slipped up in the past with mistakes from its photo patrol algorithm. The company has algorithms to check images already, which it uses to stop terrorism and pornography posts.
Memes are a different beast.
"Political memes are not something like a nude body part. It can be mashed up in a way that's creating a visual rhetoric. A machine isn't smart enough to pick up on that," Grygiel said.
That's where the 10,000 additional staffers could come in handy. But then Facebook runs into the issue of free speech on its platform.
The social network walks a fine line between censorship and fact-checking. During his testimony before Congress, Zuckerberg faced the ire of several senators and representatives reacting to accusations that Facebook had censored Diamond & Silk, pro-Trump supporters with more than 1.5 million followers.
The company wants to let people voice their opinions, but it also needs to prevent them from sharing misinformation. The dilemma gets especially tricky with memes.
"Anyone can create or post a meme. The fact that it's a meme doesn't make it malicious," Nimmo said. "The question then becomes, What criterion are you going to use to separate acceptable memes and acceptable posts from unacceptable ones?"