LinkedIn ran an experiment with more than 20 million users over five years that, while aimed at improving the platform’s functionality for members, impacted some people’s lives. A new study reveals that it is possible.
In an experiment conducted around the world from 2015 to 2019, Linkedin compared weak and strong contacts suggested by its “You May Know” algorithm (the company’s automated system for recommending new connections to users). The proportions were varied randomly. Researchers from LinkedIn, MIT, Stanford, and Harvard Business School then analyzed aggregate data from the tests in a study published this month in Science.
LinkedIn’s algorithm experiment may come as a surprise to millions of people because the company did not inform users that the test was underway.
Tech giants like LinkedIn, the world’s largest professional network, regularly conduct large-scale experiments in which different versions of their app’s features, web design, and algorithms are tested on different people. The long-standing practice known as A/B testing aims to improve the consumer experience and keep them interested, while allowing businesses to make money through premium membership fees and advertising. Users often don’t know that companies are running tests on them. (The New York Times uses tests like this to evaluate headline language and make decisions about the products and features companies release.)
But the changes LinkedIn has made show that such tweaks to widely used algorithms can be social engineering experiments that could change the lives of many people. Experts who study the social impact of computing say that conducting large-scale, long-term experiments on people whose job prospects can be affected in ways that are invisible to the industry is difficult. He said it raises questions about transparency and research oversight.
Michael Zimmer, associate professor of computer science and director of the Center for Data, Ethics, and Society at Marquette University, said, “Our findings indicate that some users had better access to job opportunities or less access to job opportunities.” This suggests that there were significant differences in access.” “When thinking about the ethics of engaging in this type of big data research, these need to consider the long-term effects.”
A study published in the journal Science explores an influential theory in sociology called “the strength of weak ties,” which says people are more likely to obtain employment and other opportunities through close acquaintances than through close friends. was verified.
Researchers analyzed how LinkedIn’s algorithm changes affected users’ job mobility. They found that relatively weak social connections on LinkedIn were twice as effective in securing employment compared to stronger social connections.
Linkedin said in a statement that it “acted consistently” with its user agreement, privacy policy and membership settings during the investigation. The privacy policy states that LinkedIn uses members’ personal data for research purposes. The statement added that the company uses the latest “non-invasive” social science techniques to answer important research questions “without any experimentation on members.”
Microsoft-owned LinkedIn did not directly respond to questions about how the company considered the experiment’s potential long-term effects on users’ employment and financial status. However, the company said the study did not unfairly benefit any users.
The goal of the study was to “help people at scale,” said Karthik Rajkumar, an applied research scientist at LinkedIn and one of the study’s co-authors. “No one was at a disadvantage in finding a job.”
The study’s lead author, Sinan Aral, a professor of management and data science at the Massachusetts Institute of Technology (MIT), said LinkedIn’s experiment was an effort to give users equal access to job opportunities.
“What they’re trying to do is run an experiment on 20 million people and use what they learn to develop better algorithms that will improve everyone’s chances of getting a job,” Professor Aral said. said. Some people are socially mobile, while others are not. ” (Professor Aral does data analysis for the New York Times and received a research fellowship grant from Microsoft in 2010.)
User experimentation by major internet companies has a checkered history. Explains how, eight years ago, social networks were secretly manipulating the posts that appeared in users’ news feeds to analyze the spread of negative and positive sentiment on their platforms. The results of a Facebook survey have been announced. The experiment, which was conducted over a week with 689,003 users, immediately sparked a backlash.
A Facebook study, authored by researchers at the company and professors at Cornell University, claimed that people were tacitly consenting to emotional manipulation experiments when they signed up for Facebook. The study states that “all users consent before creating an account on Facebook,” which “constitutes informed consent to this study.”
Critics disagreed, with some attacking Facebook for exploiting people’s moods and invading their privacy while causing emotional distress. Others argued that the project used academic co-authors to lend credibility to questionable corporate research practices.
Cornell University later said its internal ethics committee did not need to review the project because Facebook conducted its own research and the professor, who helped design the study, was not directly involved in the human experiments.
LinkedIn’s professional networking experiments varied in purpose, scope, and scale. These were designed by Linkedin as part of the company’s ongoing efforts to improve the relevancy of its “You May Know” algorithm, which suggests new connections with members.
The algorithm analyzes data such as a member’s work history, job title, and connections with other users. We then try to measure the likelihood that a LinkedIn member will send a friend invite to a proposed new connection and the likelihood that the new connection will accept the invitation.
In the experiment, LinkedIn tweaked the algorithm to randomly vary the proportion of strong and weak ties that the system recommended. The first wave of experiments, conducted in 2015, “involved more than 4 million participants,” according to the research report. More than 16 million people took part in the second wave of testing in 2019.
During testing, people who clicked on the “People You May Know” tool to view recommendations were assigned different algorithmic paths. Some of what the study called “treatment variations” caused LinkedIn users to form more connections with people with weaker social connections. Other adjustments made people less likely to form connections with weak ties.
It is unclear whether most LinkedIn members understand that they may be subject to experimentation that could affect their job opportunities.
LinkedIn’s privacy policy states that the company may “use available personal data” to research “workplace trends, such as job openings and the skills needed for the job.” . The company’s policy for outside researchers seeking to analyze company data states that these researchers cannot “conduct experiments or tests on our members.”
However, neither policy explicitly informs consumers that LinkedIn itself may conduct experiments or tests on its members.
“We are transparent with our members through the research section of our user agreement,” LinkedIn said in a statement.
“It was our and the reviewer’s understanding that the experiment conducted by LinkedIn operated under the guidelines of its user agreement,” Science said in an editorial statement.
After the first wave of algorithmic tests, researchers at LinkedIn and MIT came up with the idea of testing the strength of weak ties theory by analyzing the results of those experiments. Although this theory has been a cornerstone of social science for decades, it has not been rigorously proven in large prospective trials that randomly assign people to social ties of different strengths.
Outside researchers analyzed aggregated data from LinkedIn. This study reported that people who received more recommendations for moderately weak contacts generally applied for and were accepted for more jobs, which is consistent with weak tie theory. This is the result.
In fact, relatively weak ties, those who share only 10 or more mutual connections among LinkedIn members, are far more successful in job hunting than strong ties who share 20 or more mutual connections. It has been proven to be highly productive, the study says.
One year after connecting on LinkedIn, people who received a moderately weak connection recommendation were twice as likely to get a job at the company their acquaintance worked for compared to other users who received a strong connection recommendation. It has become expensive.
“We found that these moderately weak ties are the best option for helping people find new jobs, rather than stronger ties,” said LinkedIn researcher Rajkumar. Ta.
The 20 million users who participated in the LinkedIn experiment created more than 2 billion new social connections, completed more than 70 million job applications, and led to 600,000 new jobs, the study said. I am. Research has found that weak ties are most useful for job seekers in digital fields such as artificial intelligence, while strong ties are more useful for jobs in industries that rely less on software. It has been revealed that
LinkedIn said it applied its findings about weak ties to several features, including a new tool that notifies members when first- or second-level ties are hiring. However, the company has not made any research-related changes to the People You May Know feature.
Aral, a professor at the Massachusetts Institute of Technology, said the deeper significance of this study is that powerful social networking algorithms not only amplify problems such as misinformation, but also serve as basic indicators of economic conditions such as employment and unemployment. He said it showed the importance of this.
Catherine Frick, a senior research fellow in computing and social responsibility at De Montfort University in Leicester, UK, said the research was more of a marketing exercise for the company.
“There are inherent biases in this study,” Dr. Frick said. “This shows that if you want to get more jobs, you need to use LinkedIn more.”
audio creator kate winslet.