Some of you may recall that, back in July of 2020, there was one of those interminably long “tech CEO” hearings in front of Congress that appear to serve only to make politicians look foolish, out of touch, and technically illiterate. Among the most laughable moments was when Rep. Greg Steube wasted everyone’s time by whining to Sundar Pichai of Google that some of his campaign emails were being flagged by Gmail as spam. As tons of people pointed out in response (1) perhaps the problem was that Steube’s campaign emails were spammy, and (2) spam filters mess up all the time. Amusingly, I’ve had multiple emails sent by people employed by Google get lost in my spam filter. Content moderation at scale is impossible to do well, and the first and best example we have of that is the spam filter.
Anyway, a few weeks ago some computer science researchers at North Carolina State University released a preprint of an article looking at “the political bias in email spam filtering algorithms during the US 2020 election.” The researchers signed up for a variety of campaign emails from federal candidates running for both houses of Congress and the White House, and then had emails go to the three top free email systems: Gmail, Microsoft’s Outlook, and Yahoo Mail, and then looked at the spam filter results. It’s an interesting, but not particularly enlightening study.
The study found that Gmail’s spam filter filtered more Republican candidates emails as spam than Democratic candidates. But Yahoo and Outlook both had the opposite, filtering more Democratic candidate emails than Republicans:
We observed that the SFAs of the email services indeed exhibit political biases: they treat the left and the right campaign emails differently. Gmail leans towards the left as it marks a higher percentage of the right emails as spam. Outlook and Yahoo, on the other hand, lean towards the right
There may be some different takeaways from this, but generally, they should be that spam filters have always been somewhat random and not very good and that they are trained off of a massive corpus of data, not based on any political leanings of programmers. That’s just not how any of this works. The fact that Google’s system had one result, while Yahoo and Microsoft’s had others should cement that there’s no secret bias going on here. Just algorithms being algorithms.
But, of course, you can predict where this is going in today’s silly political world, in which Republican political insiders have to play victim every chance they can get. All last week I saw a few GOP political operatives continually linking to the study and screaming bloody murder about how Google was censoring conservatives and “interfering in the election” (while conveniently saying nothing about Microsoft or Yahoo). Eventually, you had to know that Fox News would pick up the story:
If you can’t see that image, it’s a blaring headline:
Google’s Gmail favors left-wing candidates, sends far more emails from conservatives to spam: study
Hmm. No mention of Yahoo or Outlook. While the article does mention the Yahoo and Outlook results later on in the article, it really plays up the idea that Google is somehow deliberately targeting Republican politicians, which is just laughable if you know anything at all about how spam filters work.
The fact is that spam filters suck and often make mistakes… but also that many, many people treat political mail as spammy, often because it is spammy. I’ve gotten on a few political lists over the last few years and I tend to mark them as spam as I’ve never, ever signed up to be on any such list. To turn this into “evidence” of political bias is utter nonsense.
Similarly, the demands from some that these companies need to expose their spam filtering algorithms is similarly disingenuous. The only one who benefits from those algorithms being revealed are… spammers. They’d love to know how the filtering works to get around it. So, basically, the GOP is now demanding that Google helps spammers spam more people. Incredible.