by Jim Craner and Lori Ayre
Do you know the difference between misinformation and disinformation? Misinformation is information that is mistaken -- but it's an honest mistake. Disinformation, on the other hand, is information that is purposely crafted to deceive -- think propaganda or false political rumors.
Libraries have been battling misinformation since their inception, of course. But with the rapid rise of disinformation in the digital era, libraries will need to fight this war as well -- and the first battlefield may very well be YouTube.
As a board member and volunteer for my local library, I [ed: JC] visit our library a few times a week, whether it's for board business, checking out a book, or just to have a quiet place to work outside of the house. And every time I walk past our library's public access computers, most of the tweens or teens there are invariably watching YouTube. I'm sure this situation is familiar to you librarians and library staff -- as well as many parents -- reading this.
There's nothing wrong with YouTube, of course -- as an adult well past my teenage years, I watch videos on YouTube almost daily, whether it's last night's "Late Show" monologue over breakfast or a technical database instructional video at work. But YouTube has been in the news a lot over the past couple of weeks, and there are several related issues that Internet-savvy librarians need to be discussing.
Let's recap what's happened over the past few years as well as the past few weeks:
- Elsagate - in late 2017, commentators began reporting a trend in YouTube videos for kids: thousands of algorithmically-generated videos mashing up popular trademarked characters (e.g., Elsa from Frozen which gave the "scandal" its name) with "creepy" or inappropriate content, such as Minnie Mouse being graphically dismembered. Reporters at Gizmodo, Wired, and the New York Times, among many others, investigated and YouTube suspended many channels as a result.
- A number of commentators, journalists, and media analysts have reported about the "radicalization" effects of certain YouTube algorithms, particularly in the "alt-right" and ethnonationalist communities. Essentially, YouTube's algorithms recommend increasingly extreme videos as part of the video company's drive to increase viewership and ad sales. Viewers who watch an innocuous political video can be led through the "Watch Next" feature to videos on the same topic that might feature disinformation, and in turn to outright hate speech videos.
- In a similar but even more disturbing "recommending increasingly extreme videos" story, the New York Times published an article documenting how YouTube's "Watch Next" video algorithms will recommend younger models to viewers who watch erotic videos. The recommended videos progressively include younger and younger subjects until they include children in bathing suits. The article documents that pedophiles were exchanging links and recommendations to other otherwise-innocent videos within the YouTube comments until the company disabled all comments on any videos featuring children.
- More recently, a video journalist for Vox publicly complained to YouTube about the significant harassment he faces from the fans of a particular right-wing "shock jock" commentator on YouTube [CONTENT WARNING: this sentence links to an NPR article that cites examples of racist and homophobic content]. The journalist, Carlos Maza, who is gay and Latino, compiled a video of the commentator's numerous homophobic and racist slurs, all featured on the man's YouTube channel and viewed by millions of subscribers. YouTube bungled the public relations response, declaring that the videos were legitimate debate instead of targeted harassment. After a severe public backlash, they eventually removed the commentator from their paid advertisement revenue-sharing service, but left the videos up.
- Another Internet controversy occurred last month across all the social media services when a digitally altered video of House Speaker Rep. Nancy Pelosi was released, slowed to make the Speaker seem drunk or drugged. The video spread virally, despite being quickly debunked, and was even spread by top government officials of the opposing party. YouTube, in particular, removed the video when the controversy began; Facebook, on the other hand, still displays the video (albeit with a warning pop-up questioning the veracity).
Free Speech Issues
Obviously, the promise of a global video platform where everyone can express themselves has been tempered with the reality of hate, mob tactics, media manipulation, and general creepiness. Given these issues, and the US government's recent interest in investigating YouTube's parent company Google for antitrust violations, there have even been calls for government regulation of YouTube and other social media platforms.
Contrary to a widely-held misconception, social media platforms aren't restricted by the First Amendment. They are private companies that are not obliged to carry hate speech on their services, and their legal terms usually specifically prohibit hate speech that American government entities can't ban. While public libraries have First Amendment responsibilities, and a duty to champion the free flow of information, social media companies are investigating what role their own platforms and policies play in curbing the spread of disinformation. The stakes for society can be high, especially when potentially dangerous topics are discussed, such as anti-vaccination misinformation or targeted harassment of minorities.
As damaging as these situations are for public discourse, there is also the danger of corporate technology firms silencing particular speakers. Where exactly to draw the line between outright hate speech and political discussion can be a tough decision for social media executives. Libraries have an important voice in the public conversation about information, accuracy, sources, freedom of speech, and related issues.
The doctored Speaker Pelosi video caused a huge amount of outrage before being removed -- but some new technology developments could make the librarian's job of helping people find accurate information MUCH harder: deepfakes and babblers.
"Doctored" images have been around forever -- while Stalin was known for removing purged political figures from historical photos, people have been altering photographic images almost since their invention. But what's new is the widespread availability of "deepfakes" -- computer-generated video of a real person saying or doing things that never really happened. Imagine a video of a politician making a statement she never actually said, completely artificial but still published by the opposition as real. In fact, you don't have to imagine: check out this example of computer-generated footage of President Obama created by researchers at the University of Washington. Deepfakes are now able to fool almost every human observer, whether it's a still image, audio, or video.
"Babblers" are a little harder to explain. You've probably heard of "bots," automated scripts that post to the Internet, such as posting spam in an online forum. And you may have heard of organized propaganda or "click-farm" operations, where a few people in a single building may masquerade as hundreds or thousands of Internet users, to influence online discussions or inflate the statistics or reviews of an app. These are sometimes easy to spot because the automated programs are simple compared to a human, and the "real people" operations are relatively expensive.
But AI developers recently made a breakthrough: an AI algorithm (called a "babbler") that can write entire pages of coherent text that can fool almost any human. Let's say you wanted to flood a local online forum with posts about how awful your mayor is. If you took the time to write fifty different posts, each one noticeably different, it might take you quite a while. But a babbler could produce fifty forum posts, each different in style and content, in just seconds. Now imagine scaling that up to flood an entire social media platform with seemingly-sincere fake content -- the disinformation possibilities are scary.
The team that developed this breakthrough was so worried about the possible disinformation consequences that they won't release the algorithms to the public. But a competing researcher announced this week that he will be releasing his version of the algorithm to the public to spur awareness of this issue and help others work on information "defenses," so to speak.
The stakes are high: assuming the foreign disinformation efforts on social media affected the outcome of the 2016 US election, then the potential for deepfake and other new types of disinformation to wreak havoc is immense. Bots and spammers have plagued websites and email systems for years, and the problem has ebbed and flowed -- but that is nothing compared to the potential effects of malicious artificially-generated personas, organizations, and "video" evidence in the public discourse.
When information and disinformation battle in the public square or the digital public square, how will librarians and libraries be involved? What can you do to help inform your patrons about a world with overwhelming amounts of disinformation? While there are no silver bullets on the horizon, you can be reassured that many of the skeptical information literacy tactics developed in the first decades of the Web can still provide useful protection now: carefully consider the article's facts, verify the facts where possible, evaluate multiple sources/providers of information, and don't propagate unverified information. The hard-won reputation of libraries as knowledge-sharing institutions for the public good is needed now more than ever.
At the very least, we hope you’ll be armed with a heightened level of skepticism and will have some news for evaluating video content - not matter how authentic it may appear.