For nearly a decade, researchers have amassed evidence that the social media platform Facebook unfairly amplifies low-quality content and misinformation.
So it came as a bit of a surprise when Science published a study in 2023 that found Facebook’s algorithms were not a major driver of misinformation during the 2020 US election.
The study was funded by Meta, Facebook’s parent company. Several Meta employees were also part of the author team. It attracted extensive media coverage. Nick Clegg, Meta’s president of international affairs, also congratulated the company, saying the company’s algorithm had shown “no detectable impact on polarization, political attitudes or beliefs”.
However, this finding was recently called into question by a team of researchers led by Chandak Bagh of the University of Massachusetts Amherst. In an e-letter also published in the journal Science, the researchers argue that the results are likely due to Facebook tinkering with the algorithm while the study was being conducted.
In a reply email, the authors of the original study acknowledged that the results “might have been different” if Facebook had changed its algorithm in a different way. But they insist their results are still true.
This whole debacle highlights the problems caused by funding big tech companies and promoting research into their products. It also highlights the critical importance of strengthening independent oversight of social media platforms.
suspicious merchant
Big technology companies have begun investing heavily in academic research for their products. We are also making large investments in universities in general. For example, Meta and its chief, Mark Zuckerberg, have donated a total of hundreds of millions of dollars to more than 100 universities across the United States.
This is similar to what Big Tobacco did in the past.
In the mid-1950s, cigarette companies launched a systematic campaign to discredit a growing body of evidence linking smoking to many serious health problems, including cancer. It was not about explicitly falsifying or manipulating research, but rather selectively funding research and drawing attention to inconclusive results.
This helped fuel the theory that there is no conclusive evidence that smoking causes cancer. This allowed tobacco companies to maintain a public image of responsibility and “goodwill” until the 1990s.
Major tobacco companies have launched campaigns to stir up doubts about the health effects of smoking. Ralph Liebhold/Shutterstock
positive spin
A Mehta-funded study published in the journal Science in 2023 claimed that Facebook’s News Feed algorithm reduced users’ exposure to untrustworthy news content. The authors said that “Meta was not entitled to pre-publication approval,” but acknowledged that the Facebook Open Research and Transparency team “provided significant support in the execution of the entire project.”
This study used an experimental design in which participants (Facebook users) were randomly assigned to a control or treatment group.
The control group continued to use Facebook’s algorithmic newsfeed, while the treatment group was given a newsfeed that displayed content in reverse chronological order. In this study, we sought to compare the impact of these two types of news feeds on users’ exposure to false and potentially misleading information from unreliable news sources.
The experiment was robust and well designed. However, in the short time it was in place, Meta changed the news feed algorithm to power more authoritative news content. In doing so, the control conditions of the experiment were changed.
The reduction in exposure to misinformation reported in the original study is likely due to the algorithm change. However, these changes were temporary. A few months later, in March 2021, Meta rolled back the news feed algorithm.
In a statement to Science about the controversy, Mehta said he had made the changes clear to researchers at the time and stood by Clegg’s statements about the paper’s findings.
Nick Clegg, Meta’s Global President. Will Oliver/EPA
unprecedented power
The study downplays the role of algorithmic content curation on issues such as misinformation and political polarization, and served as an indicator of widespread doubt and uncertainty about the harmful effects of social media algorithms. .
To be clear, I am not saying that the researchers who conducted the first study in 2023 misled the public. The real problem is that social media companies not only control access to researchers’ data, but they can manipulate the system in ways that affect the outcomes of the research they fund.
Additionally, social media companies have the power to promote specific research on the platforms they study. This in turn helps shape public opinion. That could create a scenario where skepticism and suspicion about the impact of algorithms becomes the norm, or where people simply start ignoring them.
Such power is unprecedented. Even Big Tobacco has never had such direct control over the public’s perception of itself.
All of this highlights why platforms are mandated to provide both large-scale data access and real-time updates on changes to algorithmic systems.
When a platform controls access to a “product,” it also controls the science behind its impact. Ultimately, these self-funding models allow platforms to prioritize profits over people and distract from the need for greater transparency and independent oversight.
