Google Has a Striking History of Bias Against Black Girls

March 26, 2018 7:44 PM

Time Magazine has an excerpt from Professor Safiya Umoja Noble, author of Algorithms of Oppression: How Search Engines Reinforce Racism. In looking at Google, Noble finds that, "What we know about Google’s responses to racial stereotyping in its products is that it typically denies responsibility or intent to harm, but then it is able to “tweak” or “fix” these aberrations or “glitches” in its systems."

"What we need to ask is why and how we get these stereotypes in the first place and what the attendant consequences of racial and gender stereotyping do in terms of public harm for people who are the targets of such misrepresentation. Images of white Americans are persistently held up in Google’s images and in its results to reinforce the superiority and mainstream acceptability of whiteness as the default “good” to which all others are made invisible."

Noble was in the news last month because the Institute of Electrical and Electronics Engineers History Center tweeted out an apology to author Safiya Umoja Noble after one of its historians shared a glaringly insulting criticism of her work from the organization’s Twitter account (and then, ironically, tried to delete the entire history of the exchange and apology).

posted by TwoStride (33 comments total)

I am so glad to see this on the Blue! Noble also has a few great videos related to the topic of her book (although I encourage everyone to go buy the book, it's a great read - accessible, fascinating, and will really change the way you think about search):

2015 talk at UBC: "Just Google It": Algorithms of Oppression
2016 Personal Democracy Forum talk: Challenging the Algorithms of Oppression
2017 Open Access Symposium: Toward an Ethic of Social Justice in the World of Online Information
posted by sockermom at 8:18 PM on March 26, 2018 [5 favorites]

Interesting read-- thanks for posting. I'll have to check out the book as well.

Anecdotally I recently had two search experiences that I think are inline with this.

One was yesterday when I searched for "san da" which is a type of martial arts. When I looked for just "san da" I got martial arts results, all featuring men. When I looked for "women san da" because I wanted to see videos of women fighting, I got results for "women sandal" and a bunch of shoe shopping stuff.

The other is even more interesting, now that I try to recreate it. I had been at a protest and there was a group of Latina women doing cool work, I was trying to find them after the fact. I can't now remember the search terms I used, but I know I used "Latina" and I got a bunch of personal ads (seeking Latinas, not self-identified) and some porn.

Just now I tried it again, seeing if I could remember the terms I used. I looked for "Latina organizations" which autocorrected to Latino organizations and gave me several pages of relevant results. When I tried again, insisting on "Latina" I got a single page of random stuff, very little of it relevant.
posted by mrmurbles at 8:56 PM on March 26, 2018 [20 favorites]

I remember this! My kid wanted to see dreadlocks for little girls, and I got pages with 40-50% white girls dreadlocks from google. Like, uh, the one hairstyle that should be at least 90% POC children? I got so pissed, and we ended up looking through Tumblr instead.

And searching for Asian anything defaults to Chinese. You have to drill down to an ethnic background you want to get anyone who isn't ethnic-Han. I take it for granted that anything related to women has to be specified, because otherwise the default is always (white|male|American).
posted by dorothyisunderwood at 10:57 PM on March 26, 2018 [16 favorites]

I get 24% white girls on the same search, which of course Google's algorithms have tailored to fit my profile so it's not the same as your (or anyone else's) search results, making comparison difficult.

There's another search experiment from some years past which I'm not going to perform here at work to check if it still works, but turn of Safe Search, and then search for "European", then "African", then "American" then "Asian". Spoiler: The first few are flags, landmarks and other typical stuff, the search results for "Asian" are all porn.
posted by Harald74 at 11:33 PM on March 26, 2018 [4 favorites]

I've noticed the same sort of thing - not with regards to racism, but the general dumbing-down of search results in favour of links that can be monetised. If something can be interpreted as a search for a commercial product that result always seems to be prioritised. Also, quotation marks no longer seem to restrict searches to the literal contents as they once did, nor do minus signs always succeed in excluding terms. Consequently, it can be very hard to research things that share a name with a commercial product.

I just tested Google's current level of commercialisation by typing four random-ish words in a clean private browser window: "cypress", "confront", "pedestal", and "riot". The first result for "cypress" was Wikipedia - good! - but the rest were advertisements for a band called "Cypress Hill" and for assorted local timber suppliers. "Confront" and "pedestal" mostly gave me dictionary definitions - an OK but not inspired result IMO - but "riot" gave me:
• a very large listing for each of the branches of an art store called "Riot" (also, a map showing their locations);
• a very large link to Riot Games' support account in Twitter (including a screenshot);
• a link to a local variety show of the same name; and
• a link to the listing for a TV show, also with the same name.

Wikipedia was in the fifth place. It's hard to avoid thinking that I may have been served paid links, or at least ones that might inspire competitors to buy advertising on that page.

From the viewpoint of an advertising company, this makes sense: they don't make money out of non-commercial searches, so why pander to the abstract curiosity of intellectuals? But in my opinion the loss of the accumulated knowledge represented by Google's mind-map (for they surely won't keep it indefinitely, and in any case it can no longer be accessed) is a cultural crime comparable to the destruction of the great library of Alexandria.
posted by Joe in Australia at 11:43 PM on March 26, 2018 [14 favorites]

Google was once a fine and useful thing that produced a ton of relevant and often unexpected information. It is now a fairly useless and frustrating thing, routinely changing search terms and returning tons of irrelevant ways for somebody to solicit money.

But this:
I looked for "Latina organizations" which autocorrected to Latino organizations ...
Well, that's just fucked up.
posted by Kirth Gerson at 3:27 AM on March 27, 2018 [10 favorites]

but the general dumbing-down of search results in favour of links that can be monetised.

I'm seeing this in news. I curate African economic and business news frequently on a weekly basis and the degradation of content is obvious.
posted by infini at 3:30 AM on March 27, 2018 [20 favorites]

I use a couple of other search engines before I turn to Google- and then only if the others fail to find a useful link.
posted by Burn_IT at 4:16 AM on March 27, 2018 [2 favorites]

Burn_IT: I use a couple of other search engines before I turn to Google- and then only if the others fail to find a useful link.

What are the better alternatives? I've been growing increasingly dissatisfied with Google myself.

And how do the alternatives perform on the searches mentioned in the article? A quick check of Bing image search isn't faring much better on "unprofessional hairstyles for work", at least until you get three pages down to the guy with "SAUSAGE" reverse-shaved onto his head.
posted by clawsoon at 4:35 AM on March 27, 2018 [2 favorites]

What are the better alternatives?

I use DuckDuckGo.
posted by Pendragon at 5:06 AM on March 27, 2018 [10 favorites]

before I turn to Google- and then only

Yes, similar here, Google only to verify that maybe I made a bad search query. But now, maybe not a good test after all.

Having a hard time seeing how huge growth in companies benefit anyone and an easy time seeing how it amplifies these kinds of biases.
posted by filtergik at 5:38 AM on March 27, 2018

I was appalled when I pulled out a worksheet I'd made a couple years ago that had pictures of athletes doing different sports on it, and realized they were all white men. I hadn't set out to get pictures of white men, of course. But when I googled each sport and grabbed a pic off the first page of results...
posted by BrashTech at 5:42 AM on March 27, 2018 [1 favorite]

whenever I search for a public person x, "x net worth" and "x wife" are in the suggested results, which I find creepy af
posted by thelonius at 5:48 AM on March 27, 2018 [5 favorites]

they don't make money out of non-commercial searches, so why pander to the abstract curiosity of intellectuals?

it's also sad how flat and predictable are the results for plain searches on history or literature or philosophy: wikipedia, a few history sites aimed at high school students, various online Cliff's Notes type things, a few other reference sites.
posted by thelonius at 5:57 AM on March 27, 2018 [4 favorites]

it's also sad how flat and predictable are the results for plain searches on history or literature or philosophy: wikipedia, a few history sites aimed at high school students, various online Cliff's Notes type things, a few other reference sites.

I wonder if this is tied to the release of Google Home, or if it will be accelerated by the increased advertising push to get people doing verbal Google searches rather than pulling out a device. Flat, predictable, and simple results are much easier for the assistant to read out to the asker.
posted by halation at 6:09 AM on March 27, 2018 [1 favorite]

There's now a "verbatim" search option under the "tools" button which you must use for verbatim searches. Because, of course, quotes no longer mean "verbatim". Sheer patronizing ridiculousness in service of Google Knows Best.
posted by inconstant at 6:21 AM on March 27, 2018 [39 favorites]

Google also broke the + switch in their search a decade ago in their hamfisted cargo cult attempt at a social network service. They still haven't fixed it, and I still don't know how to say require word X near word Y in the results.

So I mainly use DuckDuckGo. Which doesn't work as well as Google but at least they don't break their own syntax for stupid reasons.
posted by tclark at 7:20 AM on March 27, 2018 [7 favorites]

I love how Google now returns results with only some of your search terms as if you typed half of them in by mistake or something.
posted by octothorpe at 7:22 AM on March 27, 2018 [25 favorites]

An evergreen talk on this is "Programming is Forgetting" by Allison Parrish, on digital imperialism, on organized forgetting as discovery (♥️ u uklg, rip), on the ethical code we bake into everything we create.
posted by tapesonthefloor at 8:07 AM on March 27, 2018 [5 favorites]

I use DuckDuckGo.

Currently Google has improved its results for the "black girls" search (as the article points out) while DuckDuckGo still returns mostly porn. Unfortunately this is consistent with my experience with DuckDuckGo on the several occasions I've tried to switch—the result quality always lags behind Google.
posted by enn at 8:18 AM on March 27, 2018 [4 favorites]

If you don't want porn, use the safe search: strict option.
posted by Pendragon at 8:27 AM on March 27, 2018

Another book that touches on tech's bias is Technically Wrong: Sexist Apps, Biased Algorithms, and Other Threats of Toxic Tech. It's not strictly focused on race, but it was a really good read. I have Algorithms of Oppression ready to read after I finish my current book and I am really excited about a deeper dive.

Related to this, though it's been great to see more diverse stock photo sites come up recently—TONL, Colorstock—I want the balance to shift more quickly. I'm a graphic designer who does work with a behemoth infant nutrition company and finding current, warm and fuzzy images of a non-white mom breastfeeding a baby is impossible. We often end up using the same few shots even though we are always looking for more. It's hard enough to talk your client into the need to show diversity (not just ethnicity but also diversity of body size and shapes) but if you can't find quality images then it's even harder.

If you search breastfeeding and select Black on sites like Getty or Masterfile that let you specific model ethnicity, you will likely get a few images that were shot in the 90's and then you'll get Nat Geo type images of women in Africa breastfeeding. Or the non-white mom's in stock photos won't look happy, they'll look angry or sad or be on their phones. Or they'll be feeding formula instead of breastfeeding. (No judgement on the latter, it's just that I'm searching for "breastfeeding" and the substitution doesn't occur when the model is white.) It's worse for Latinx search queries, slightly better for Asian, but as dorothyisunderwood mentions, there a lack of diversity there too.
posted by the thorn bushes have roses at 8:47 AM on March 27, 2018 [9 favorites]

If you don't want porn, use the safe search: strict option.

The problem isn't How To Avoid Porn, the issue is search results skewing towards porn because of race/ethnicity. Safe Search also has a history of filtering out non-porn pages which focus on things like sexual health, art, etc. Imagine for instance trying to search for sexual health resources/information that focuses on Black or Asian women...
posted by Secret Sparrow at 8:55 AM on March 27, 2018 [24 favorites]

The problem isn't How To Avoid Porn, the issue is search results skewing towards porn because of race/ethnicity.

I was talking about DuckDuckGo.

If you use the safe search: strict option on DuckDuckGo and search for "black women health", you get some good ( I think ) results.
posted by Pendragon at 10:50 AM on March 27, 2018

The point is that you shouldn't need safe search: strict when searching for "black women health" to avoid porn.
posted by TwoStride at 11:07 AM on March 27, 2018 [13 favorites]

Why not ? This is the internet, 50% of it is NSFW. Any search term you use leads to porn. You can't expect a search engine to know what you want to search.
posted by Pendragon at 12:12 PM on March 27, 2018 [2 favorites]

And the default setting of DuckDuckGo is strict. So, you don't have to set anything. If you want adult content, you set it to off.
posted by Pendragon at 12:16 PM on March 27, 2018

Pendragon, are your first thoughts pornographic when you think about black women's health?
posted by TwoStride at 12:36 PM on March 27, 2018 [9 favorites]

Any search term you use leads to porn.
I'd love to see the actual evidence for this, particularly since it directly contradicts the arguments Dr. Noble makes in the book and videos that we are discussing in this thread.
posted by sockermom at 1:15 PM on March 27, 2018 [5 favorites]

To add: Searching for information is a core part of my livelihood and I am not met with porn on the dozens of searches I do every day. There is something profoundly broken when we search for "black girls" and are met with pornography, and to pretend that the problem is that users aren't enabling safe search: strict is entirely and willfully missing the point.
posted by sockermom at 1:18 PM on March 27, 2018 [20 favorites]

The issue is not porn, per se, it's that it's getting ranked differently depending on skin color (or skin color-oriented search terms), and apparently often appearing on the first page of results.
posted by rhizome at 3:15 PM on March 27, 2018

the degradation of content is obvious

So obvious we stopped calling it what it was and started calling it, generically, "content," and the people who create it , "content creators."
posted by klanawa at 4:56 PM on March 27, 2018 [1 favorite]

You can't expect a search engine to know what you want to search.
I just went and taught a class about search and thought about this the entire time I was in there. Countless research studies demonstrate that people rarely look beyond the first page of results, despite the fact that when faced with results from pages two, three, and four, those people find as much relevant content to their topic as they do on the first page.

If Google didn't sort of know what I was looking for when I entered a search term, they would not have become the giant that they are. They literally are synonymous with search. To assert "how could Google possibly know if I'm looking for pornography?!" boggles my mind. But the problem isn't pornography. Do a search for professor, as Noble does in her book, and take a look at what you see. Anything missing? The problem is widespread structural oppression. Everything that we create has our biases embedded in it. Search is no exception. Artifacts have politics!

I mean, Microsoft is able to identify people who have pancreatic cancer just based on what they are searching for on Bing before they have been diagnosed with cancer. These companies can do better. They know what you're searching for even when you don't.
posted by sockermom at 6:06 PM on March 27, 2018 [8 favorites]