Last year, a tech writer in the US launched an experiment to see if Google was racially profiling users of its popular email service, Gmail. The experiment involved using totally clean email addresses which had racially charged names, sending particular emails with common titles (e.g. “I Need Cash”) and then looking closely at the ads which each of different ethnicities were served.
The experiment unearthed compelling evidence that Google’s email service would give radically different advertising to different people, based on them having an ethnic name. In the last couple of weeks, I’ve repeated the experiment in the UK, to see if the results came out the same.
The results were stark, and similar to the original experiment – for example, an email sent by “Robert Howe” saying “Need Cash” gets foreign exchange solutions for business advertised to him; the same email sent by “Segun Akinkube” gets offered Payday Loans. Neither of the ads repeated in each other’s preferences; Segun & Robert got completely different ads served to them, when all other factors were the same. Why? That’s a question for Google to answer.
At the time, of the original experiment, Google disputed the conclusions, issuing a statement saying the experiment “relies on flawed methodology to draw a wildly inaccurate conclusion. If Mr Newman had contacted us before publishing it, we would have told him the facts: we do not select ads based on sensitive information, including ethnic inferences from names.”
If Google was right when dismissing Newman’s claims, then why am I able to reproduce similarly disturbing results today? Google doesn’t deny that gmail reads your emails, and serves you ads based on the content of them. You can click a link entitled “why these ads?” on the right hand side of your gmail profile, and Google will admit that at least one of the ads “was based on an email from your mailbox.”
Clearly, Google draws inferences from certain words and phrases in your mailbox. What does it see in certain names? The machine intelligence that serves the ads is often worryingly intuitive – for example, one person I spoke to while researching this article told me not only did it accurately begin offering her wedding dresses and diet plans days after her fiancé proposed to her, but also, when she started turning down ads for weight loss, the wedding dresses became adverts for maternity bridalwear.
The machine intelligence was able to look at her choices and emails and determined that the most logical reason for a bride to turn down a diet before her wedding was because she was pregnant. In this case, she wasn’t, but it just goes to show the sophistication of the algorithm.
So, why does an algorithim so sophisticated throw up results that look a lot like racial profiling? My own personal hunch is that Google’s sophisticated algorithm can tell a lot about you by the names in your emails – and that is especially true if names have other meanings, are in particular languages, replicate place names, or are particularly common within a community.
Thus, for example, in my own experience, people who are regularly emailing friends using Italian words will be more likely to be offered flights to say, Rome. Anecdotally, people who include terms like “Ramadan” or “Eid” are more likely to get trips to Mecca advertised to them; and those who add “Al-Haj” to their names see those ads disappear. It’s a very fine line from that sort of targeting by nationality, religion or interests to creating a service that effectively segregates the internet by ethnicity.
As a white man, this doesn’t bother me too much. I get free useful services, and I’m pretty sure I get offered all the good stuff. As ever, I have it easy. However, if I was a woman? Or of a different ethnicity? I think it would scare the hell out of me that I was getting a radically different online experience to benefit advertisers.
If you think this all sounds paranoid, see here for a video of Google’s chairman discussing how important knowing your real name is for their advertising purposes.
It’s yet another proof that Google services are not free – when you use a “free” Google service like Gmail, Google Maps, or Google Search, you are a product, a commodity being packaged and sold to advertisers. And with such a valuable product, who can blame them for wanting to label each product correctly?