Feb 23, 2021 - Technology

How memes became a major vehicle for misinformation

Illustration: Aïda Amer/Axios

Wall Street's populist uprising, the Capitol siege and a strong U.S. anti-vaccination movement show the power of memes in spreading misinformation and influencing communities online.

Why it matters: For years, there's been growing concern that deepfakes (doctored pictures and videos) would become truth's greatest threat. Instead, memes have proven to be a more effective tool in spreading misinformation because they're easier to produce and harder to moderate using artificial intelligence.

  • "When we talk abut deepfakes, there are already companies and technologies that can help you understand their origin," says Shane Creevy, head of editorial for Kinzen, a disinformation tracking firm. "But I'm not aware of any tech that really helps you understand the origin of memes."

Catch up quick: A meme is a piece of mixed media, usually text laid over a photo or video, that is designed to go viral, often through humor.

  • Some memes can be lighthearted, like the viral Bernie Sanders mittens meme from Inauguration day. But many memes are meant to be deceptive, or prey upon fears and biases.

Driving the news: New research from media intelligence firm Zignal Labs shows how memes became a powerful agent for spreading misinformation online around the COVID vaccine.

  • The data shows that a single meme, first circulated late last December, has helped to drive thousands of new mentions of a conspiracy tying the COVID-19 vaccine to 5G.
  • The tweet, which continues to go viral today, features an electric circuit of a guitar pedal, claiming it's diagram of a 5G chip. Text overlaid onto the meme suggests the architecture of the chip is the same as that of the COVID-19 vaccine.
  • Exposure to vaccine misinformation was tied to a roughly six-percentage point reduction in people's intention to get vaccinated, according to a study published earlier this month.
  • Most of the misinformation researchers encounter on social platforms features media that manipulates context, like memes — not deepfakes.

The big picture: Memes have become more popular in recent years as photo-editing and sharing software becomes more ubiquitous.

  • "We're seeing a long-term, multi-year shift to richer media," Facebook CTO Mike Schroepfer recently said on a call with reporters.
  • "10-12 years ago we predominantly saw text on the platform," he said. "Now, imagery plus text and video is on rise."

Yes, but: While artificial intelligence has been successful in identifying misleading phrases and images separately, it isn't yet equipped to understand how context changes when text and images or videos are overlaid in a meme, Creevy says.

  • It is partly an image recognition problem. "Memes can be made arbitrarily complex," says Dileep George, an AI researcher and founder of Vicarious AI. For example, the form, size and placement of letters can be varied, and a scene can be put inside a letter and another in the background.
  • A human can detect those manipulations, but they can be "extremely hard for deep learning systems to parse," George says, adding that AI systems may get better at detecting them as the data used to train them expands.

But AI meets another challenge in memes: understanding the cultural context. Memes are often rooted in satire.

  • "Common sense needs to be solved to crack that problem," George says, adding it will require fundamental breakthroughs in building AI systems that draw more on principles from how the human brain understand cultural context.

What to watch: What could help to address meme misinformation at scale is decentralized human content moderation, Creevy says, or tasking many people to help identify potentially misleading memes.

  • In speaking with reporters on this issue, Schroepfer noted that Facebook has organized a "hateful memes challenge" to try to further decentralize Facebook's efforts in policing hate speech through memes.
  • Twitter recently rolled out a crowdsourced misinformation effort called Birdwatch, which allows people to identify information in Tweets they believe is incorrect or misleading and provide notes that offer informative context.

The bottom line: "I would love to think there will be tech developed that can solve the meme problem," Creevy says. "But tech hasn't gotten so far."

Go deeper: Experts say vaccine misinformation crackdown is coming too late

Go deeper