LinkedIn ran a test on more than 20 million users over five years, which while aimed at improving how the platform works for members, could affect some people’s livelihoods, according to a new study. .
In tests conducted around the world between 2015 and 2019, Linkedin randomly changed the ratio of weak and strong contacts suggested by the “People You May Know” algorithm – its automated system. company to suggest new connections to users. The tests have been detailed in a study published this month in the journal Science and co-authored by researchers at LinkedIn, MIT, Stanford and Harvard Business School.
LinkedIn’s algorithmic tests may surprise millions of people because the company didn’t notify users that the tests were underway.
Tech giants like LinkedIn, the world’s largest professional network, regularly run large-scale experiments in which they try different versions of app features, web design, and algorithms. on different people. The age-old practice, known as A/B testing, aims to improve the consumer experience and keep them engaged, which helps companies make money through premium membership fees or advertising. Users are often unaware that companies are running tests on them.
But the changes made by LinkedIn are an indication that such tweaks to widely used algorithms could turn into social engineering experiments with potentially life-changing consequences for people. with many people. Experts who study the social impact of computers say conducting lengthy, large-scale experiments on people can affect their job prospects, in ways invisible to them. , raising questions about industry transparency and research oversight.
“Findings suggest that some users have better access to job opportunities or a meaningful difference in access to job opportunities,” says Michael Zimmer, an associate professor of computer science and director of the Center for Data, Ethics, and Society at Marquette University. “This is the kind of long-term consequence that needs to be pondered when we think about the ethics of engaging in this kind of big data research.”
Research in the journal Science examined an influential theory in sociology called “The power of weak ties“This suggests that people are more likely to get jobs and other opportunities through broad acquaintances than through close friends.
Researchers analyzed how changes in LinkedIn’s algorithm affected users’ ability to migrate. Surname found that social relations were relatively weak LinkedIn has proven twice as effective in securing jobs than stronger social relationships.
In a statement, Linkedin said that during its research, it had “acted consistently with” the company’s user agreement, privacy policy, and membership settings. The Privacy Policy Note that LinkedIn uses members’ personal data for research purposes. The statement added that the company used the latest “non-invasive” social science techniques to answer key research questions “without any testing on members.”
LinkedIn, which is owned by Microsoft, did not directly respond to a question about how the company considered the potential long-term consequences of the tests on users’ employment and economic well-being. But the company said the study did not benefit some users.
The goal of the study is to “help people at scale,” said Karthik Rajkumar, an applied research scientist at LinkedIn, is one of the study’s co-authors. “No one is at a disadvantage when it comes to finding work.”
Sinan Aral, a professor of data science and management at MIT and lead author of the study, says LinkedIn’s tests are an attempt to ensure that users have equal access to job opportunities. do.
“Do a test on 20 million people and then come up with a better algorithm for everyone’s job prospects since the knowledge you learn from that is what they’re trying to do” Professor Aral said, “instead of anointing some people for social mobility and others not.” (Professor Aral conducted data analysis for The New York Times, and he received PhD student grant from Microsoft in 2010.)
Tests on users by major internet companies have had a clear history. Eight years ago, a study on Facebook describes how the social network has quietly manipulated the posts that appear in users’ News Feeds to analyze the spread of negative and positive emotions on its platform has been published. The week-long test, conducted on 689,003 users, quickly generated backlash.
The Facebook study, whose authors include a researcher at the company and a professor at Cornell, suggested that people tacitly consented to an emotional manipulation test when they signed up for Facebook. “All users consent before creating an account on Facebook,” the study said, “constitutes informed consent for this study.”
Critics disagreed, with some attacking Facebook for invading people’s privacy while exploiting their moods and causing them emotional distress. Others argue that the project used an academic co-author to give credibility to questionable corporate research practices.
Cornell later said its internal ethics committee was not asked to review the project because Facebook conducted the study independently and the professor, who helped design the study, had not directly involved in experiments on human subjects.
LinkedIn’s professional networking experiments vary in purpose, scope, and scale. They were designed by Linkedin as part of the company’s ongoing efforts to improve the relevance of its “People You May Know” algorithm, which recommends new connections to members.
The algorithm analyzes data such as members’ employment history, job titles, and relationships with other users. It then tries to evaluate the likelihood that a LinkedIn member will send a friend request to a proposed new connection as well as the likelihood that the new connection will accept the invitation.
For the tests, LinkedIn adjusted its algorithm to randomly vary the prevalence of strong and weak relationships the system suggested. The first trial, conducted in 2015, “has more than four million test subjects,” the study reports. The second trial, conducted in 2019, involved more than 16 million people.
During testing, people clicked on the “People You May Know” tool and viewed recommendations assigned to different algorithmic paths. Certain “treatment variations,” as the study calls them, have caused LinkedIn users to form more connections with people with whom they have only weak social ties. Other adjustments caused people to form fewer connections with weak relationships.
Whether most LinkedIn members understand that they could be the subject of tests could affect their job opportunities.
By LinkedIn Privacy Policy says the company may “use personal data available to us” to study “trends in the workplace, such as job availability and the skills needed for these jobs.” “. Its policy for external researchers seeking to analyze the company’s data explicitly states that those researchers will not be able to “test or perform tests on our members”.
But there is no policy that explicitly informs consumers that LinkedIn itself can test or perform tests on its members.
In a statement, LinkedIn said, “We are transparent with our members through our user agreement research section.”
In an editorial, Science said, “We and the reviewers both understood that the tests conducted by LinkedIn operated according to the principles of their user agreement.”
After the first wave of algorithmic experiments, researchers at LinkedIn and MIT came up with the idea to analyze the results from those experiments to test the theory of the strength of weak relationships. Although the decades-old theory has become a cornerstone of the social sciences, it has not been rigorously proven in a large-scale prospective trial that randomly divides people into relationships. social systems with different strengths.
External researchers analyzed aggregated data from LinkedIn. The study reports that people who receive more offers for weak-average relationships are more likely to apply for and accept more jobs – a result that is consistent with the weak-constraint theory.
In fact, relatively weak contacts – those with whom Linkedin members share only 10 connections with each other – proved to be much more effective for job search than stronger contacts that people had. Users shared more than 20 connections with each other, the study said.
One year after connecting on LinkedIn, people who received more recommendations for contacts with a weak average relationship were twice as likely to land a job at companies where those acquaintances worked. than other users get more recommendations for strong relationships.
“We found that these moderately weak relationships are the best choice for helping people find new jobs and more importantly, stronger relationships,” said Linkedin researcher Rajkumar. .
The 20 million users who participated in LinkedIn’s experiments created more than 2 billion new social connections and completed more than 70 million job applications resulting in 600,000 new jobs, the study said. Weak connections prove most helpful for job seekers in digital fields like artificial intelligence, while strong connections prove more helpful for jobs in industries that are less dependent on hardware. soft, the study said.
LinkedIn says it has applied its weak relationship findings to a number of features including a new tool Notice to members when a first or second level connection is recruiting. But the company hasn’t made research-related changes to the “People You May Know” feature.
MIT’s Professor Aral says the deeper implication of the study is that it shows the importance of powerful social networking algorithms – not only in amplifying problems like misinformation but also in basic indicators of economic conditions such as employment and unemployment.
Catherine Flick, a senior researcher in computing and social responsibility at De Montfort University in Leicester, England, describes the study as a corporate marketing exercise.
Dr Flick said: “The study has an inherent bias. “It shows that, if you want to get more work, you should be on LinkedIn more.”
