Following the crowd: Users vulnerable to sharing low quality news with high likes and shares

Editor’s note: is a technology company, though we also like to look at conversations happening around the use of algorithms to highlight information online. This is one of the occasional posts we do about discussions happening on topics that we find interesting.

By Christopher Brennan

To share or not to share?

While a lot of talk around tech, such as my piece last week on “The Social Dilemma,” looks at the big macro-level forces at work, this week I wanted to focus on something smaller, the decision behind someone clicking “share.”

The mechanism behind the share has actually been a place where tech companies have experimented to combat misinformation, such as with WhatsApp limiting the number of times content can be forwarded by users.

Last week Twitter also announced that it was going to bring the “read before you retweet” feature to the world after putting it on trial this summer. It swoops into action when someone is going to retweet something with an article they have not actually clicked on and reminds them that “headlines don’t tell the whole story” before they reshare it. Twitter said that users open articles 40% more after seeing the prompt. 

This got me thinking about why people share what they share. Luckily I had the chance to speak with Nicholas Micallef, a researcher affiliated with NYU Abu Dhabi, who with colleagues at Indiana University put out a paper about a specific angle on sharing: the impact of the numbers that we see beforehand. The upshot is that users of their misinformation-focused game called Fakey, a sort of social media emulator, were more vulnerable to sharing low quality information when they were exposed to the numbers of likes and shares. 

Does misinformation benefit from social media users joining in to viral crowds? Below is our conversation, edited lightly for length and clarity.

This whole experiment was based off of a tool, or a game, called Fakey. Can you explain a little bit about what it is and what it is designed to do?

The idea here is to teach news literacy and get people better at organizing news. That is the reason why we don’t show the source. We show the title and a short description. And we want people to learn how to recognize news content. They can take five actions. So they are going to either share or like the article, so that they can show that they feel that it’s a mainstream article. But then when they have doubts, when they feel that it’s a low credibility article, they can fact check it. And then they can always go ask for a hint, which comes in the form of the source. And finally, they can skip the article, because generally, you browse social media, and sometimes you just don’t look at an article … And so [Fakey’s] being provided as a game so that you get points based on your performance and your skill. We try to make it engaging from that perspective as well. This [paper] is basically just a small preview, but the game does improve people’s skills, we found.

An image that explains the game Fakey

Something that I wanted to mention is that the particular article [you’re referring to] is focused on low credibility sources. But what we found for low credibility sources,  that people like and share with them and engage when there are higher engagement metrics, that applies also for mainstream news. Which is also a problem. People are liking and sharing mainstream news because there’s a lot of likes, not because they are looking at the quality of the article. So that ties up with your work [at] as well. It is something that we found. So people are not looking at the quality, they are looking at the likes.

So the main finding of the paper was that exposure to social metrics, what we may think of in our own lives as knowing whether something is going viral or not, influenced the likelihood of users clicking “share” for themselves. Was this surprising?

The results did confirm the initial hypothesis of the project. I do not think that there were any results that surprised us. One thing is that the correlation with sharing and metrics also applied to mainstream sources. This means that social metrics results hold regardless of source credibility. And in our view this is problematic for both mainstream and low-credibility information.  Displaying high levels of engagement makes it less likely that people will be careful about fact-checking questionable content, at the same time makes it more likely that they will like or share it. For mainstream content, these findings indicate that high social metrics may drive people to share articles simply because a large number of others have done so, rather than on the merits of the article content and source. This means that they are doing the correct action for the wrong reason.

The people liking articles that have already have a lot of likes, or sharing articles that already have a lot of shares, is that something you think is kind of inherent in social media, that will always be part of what “social” media is? Looking for the viral thing?

Yes, and we show that. That’s why we are arguing that this is when it comes to low credibility sourcing, when it comes to third-tier articles, maybe you should hide the metrics, or maybe you should show them less. So something needs to happen, because as long as you are showing likes and shares, people will continue to be influenced by that.

You mentioned the danger of metrics and people sharing things based on the presence of high engagement. In the paper you also mention that inauthentic campaigns, influence operations as they are often called, as being related to this. I was wondering if you could talk about why that is a particular danger.

So there are these campaigns which have bots. They like or share items so that they increase the number. Because once they increase the number, you have me see it and I see how many people liked it, but those likes and shares won’t be authentic. So that’s affecting us, the belief that that’s all organic. That’s why it’s dangerous, and this danger is why there is so much importance to the engagement metrics.

In terms of how to fight this, it’s interesting, because there are different approaches. One of which is looking at individual behavior, teaching people to make sure they are fact-checking things. And then the other is from the platforms themselves, whether it is priming, or whether it’s changes in algorithms, etc. And I was wondering which one you think is most effective?

I think that so I think that education has to be one of the factors definitely. But the platforms need to investigate how to reduce the impact. And so I think that, in my opinion, there needs to be multiple solutions …

[On platforms] they can get ads, and the ads are seen more and from the ads they can get money. So these [viral] articles, the firms want that. And this is quite a difficult problem to solve, because the platforms need to make money as well. So there needs to be some kind of balance. That’s where I think research needs to go, to find that balance between keeping this popularity for mainstream content, but not for low credibility, fake news content. So there needs to be a common solution, which does not take away the benefits of one to help the other. 

And when you talk about mainstream versus low credibility, you think that filtering the source might be the answer, or differentiated between sources just for the purposes of your paper?

So first of all, filtering by source, it has an 80% accuracy, from what previous studies have done. So 80% accuracy, meaning that 80% of low credibility sources may contain misinformation, and the other five percent are questionable and 15% they are not, I think . I think source can be a factor in this. But again, it’s not the only solution. It could be one of the measurements that social platforms can use.

One thing I was interested in is your take on on Twitter, the function that came out where Twitter will remind users to read the article if they try and retweet something without reading the article first. And I was wondering what you think about that as a method to counteract engagement metrics.

That’s an interesting method. But I don’t think that would always work. So it’s interesting, and it’s good that some that platforms have bee  have been doing stuff to improve the problem, so to speak, to make things better. So even Facebook, it is fact-checking items based on who reports them. It’s giving them to fact-checkers and marking them as false. So it’s good that these things started. There is still a long way, but it’s good that there is a start. There is a beginning.