Add Gmail to the list of reasons why Big Tech must open its black boxes – NYPost
In yet another sign of Big Tech bias, it turns out Google’s Gmail program leans strongly Democratic — automatically sending Republican fund-raising mail to users’ spam folders far more often than the Democratic ones in the runup to the 2020 elections.
The news comes courtesy of North Carolina State computer-science researchers, who examined 318,000 emails from May through November 2020 and evaluated Outlook and Yahoo accounts as well as Gmail.
As the Republican National Committee sums it up in its complaint to the Federal Election Commission, the Gmail algorithm sent GOP fundraising e-mails to spam at a rate 820% higher than Dem ones. The study puts it less shockingly: Nearly seven in 10 Republican messages auto-spammed, vs. less than one in 10 Democratic ones.
The other programs’ bias favored GOP e-mails, but by much narrower margins. And vastly more Americans use Gmail for their personal correspondence.
The RNC estimates it lost $1.5 billion in donations as a result, a loss of resources that certainly could have shifted, say, the Georgia runoffs that handed control of the Senate to the Democrats.
Was it intentional? Only Google’s engineers know for sure. The same issue is why Elon Musk wants to make Twitter’s “censorship” algorithms open-source, so that any qualified outsider can flag issues.
As we noted last week, it’s long past time that the tech firms who wield such vast power over our discourse “open up their black boxes so we can all see what is really going on inside.”
The nation needs transparency from Big Tech.
A Peek into the Political Biases in Email Spam Filtering
Algorithms During US Election 2020 –
Department of Computer Science, North Carolina State University
Raleigh, North Carolina, USA
“Research Questions: In this paper, we attempt to assess the fairness of the SFAs of three dominant email services, Gmail, Outlook, and Yahoo, in the context of the 2020 US election. Specifically, we study the following four research questions: This is a pre-print. The final version is published in The ACM Web Conference 2022 (WWW ’22).
• Q1: Do SFAs of email services exhibit aggregate political biases? How do these biases compare across email services?
• Q2: Do SFAs treat similar emails from senders with different political affiliations in the same way?
• Q3: Do the interactions of the users with their email accounts, such as reading emails, impact the political biases of SFAs?
• Q4: Do SFAs exhibit different political biases for recipients belonging to different demographic?
To the best of our knowledge, there is no prior published work that has examined biases in SFAs towards political campaigns.
To answer these questions, we conducted an extensive study during the 2020 US election over a period of 5 months from July 1, 2020 to November 30, 2020 on Gmail, Outlook, and Yahoo. We created 102 email accounts and subscribed to 2 Presidential, 78 Senate, and 156 House candidates. To accurately estimate the political biases and mitigate any potential effects of demographics (ethnicity, age, and gender), we created multiple email accounts with different combinations of demographic factors and designed two experiments. The first experiment studies the general trends of biases in SFAs across the email services for the Presidential, Senate and House candidates. The second experiment studies the impact of different email interactions such as reading the emails, marking them as spam, or vice versa on the biases in SFAs.
The full report is available here: