Algorithms of Oppression – Book Review

Artificial Intelligence is a system of replicating the past. Algorithms of Oppression explains the surprising origin of bias in search results. When predicting the future of Ai, you’d be surprised at how well the past predicts the future.

Ai Parenting is a judgement free community moving from screen time to quality time and our motto is don’t sedate, relate to create. The purpose of this blogpost is uncover the sedation around bias in Google Search. You’ll see how bias in search is similar to bias in library sciences and how bias in search is different since Google is a private company rather than a public good like your library.

This is a book review of Algorithms of Oppression by Professor Safiya Noble. Her books is available on Amazon. When it comes to content moderation, I’m also going to draw from Sarah T Roberts book called Behind the Screen

Today’s session is the third piece of my Foundations of Artificial Intelligence series, a look into the history behind of Artificial Intelligence and how this gets repeated today. Knowing the origins of the practical Ai systems that we use every day will help you understand their strengths and weaknesses and it will also help you predict where things will move in the future.

So let’s jump into it. The three topics that we will cover are

  1. How bias moved from libraries to search
  2. What happens when search moves from a Public to a Private good
  3. How you‘re creating a database of intentions

How bias moved from Libraries to Search

In 2012, Professor Safiya Noble was dared by an academic colleague to search black girls on Google. She was shocked by the result. Which in 2012 was mostly pornographic images of black women [1].

In 2016 MBA Student Rosalia did a Google Image searches for “unprofessional hair styles for work” that returned only images of black women, while an image search for “professional hair styles for work” returned only images of white women [2].

An image search for “three white teenagers” showed smiling teens playing sports, while a search for “three black teenagers” or “three brown teenagers” showed three police mugshots. A black student tweeted about how Google mislabeled an image of him and his friend as two Gorillas[3].

Does this bias only exist inside Google Search? Well if we look at the history of Libraries we see similarly alarming results. 

When I went to school visiting a library meant asking a librarian for a book related to a topic that I wanted to search. The Librarian (usually a white woman) understood the catalogue better than I did so she would usually write down a catalogue and index that I would then need to find on the library shelves. When it came to these catalogues there were three main catalogues that are used: the Dewey Decimal System later replaced by the Universal Decimal System, and the Library of Congress System still in use today. 

Library science was generally reserved for the most privileged in society so it might not be surprising that the Library of congress had books under the category “Illegal Aliens” and only changed this to “non-citizens” and “unauthorized immigration” in 2016. What is surprising that the House of Representatives overturned this ruling a few months later with the goal of duplicating the language of federal laws written in Congress, it’s very rare to have politicians rule on library categorization but renaming illegal aliens was crossed a political line. It shows how this bias exists even today. 

In the 1970s “African-Americans” replaced the category “Negroes”, this was replaced with blacks in 2001.  “Mentally handicapped” or “Retarded persons” was replaced with “People with Mental Disabilities” in 2001. LGBTQ2+ content was categorized as “Sexual Deviations” until it was changed to “sexual minorities” in 1972, to “sexual persons” 2016 [4].

This means that it’s not only Google that has issues in searches but libraries as well. For example, the library of Congress had racist category names for Jewish, Asian, and blacks. Image searches on library databases yields similarly troubling results. 

Generally these categories and search results reflect a larger bias within society but we also need to ask the question of their impact on people of colour. For example, how will a black child see themselves when image results are mainly sexualized or mugshots? 

This leads into the question of who is responsible for correcting these results.

What happens when search moves from a Public to a Private good?

We can see that over time due to political pressure Libraries made corrections to their categorization to use more politically correct terminology such as people with mental disabilities, blacks, and sexual persons. 

When Google was presented with the pornographic search results for black girls, they said that Google is not responsible for search results. The results come from user choices that they do not control. So Prof Noble asked if Google is not responsible for the search results of black girls then who is? 

Safiya argues that most people see Google as a public utility like a service that we pay for with our taxes. We often trust the results that we get from Google as the objective truth. However, Google is not a public utility. Since it sells advertisements companies, industries, and governments can pay for better rankings through ads and search engine optimization. So we can think of the results of a Google search not as the truth but as what some interested party wants you to know. 

These results are nearly impossible for regular citizens to change even when they cause harm. For example, Hristo Georgiev was labeled a serial killer by a Google Search result since his name corresponded with a deceased convict. In response, a Google spokesperson said “Organizing billions of entities in our Knowledge Graph requires that we use automated systems, which often work well but are not perfect.” Who bears the responsibility people like Hristo who can’t get a job because of these search results? This falls on the public through unemployment or worse the prison system. 

Another example of how search results are not objective reality is by searching for competitors of Google when they are not the dominant platform. In June 2021 Google was fined $270M in France for anti-competitive behaviour for favoring their advertising network over others in search results [5].

When we look at these and countless other examples we see that search result manipulation is merely another form of oppression used by those with the money to manipulate it. This helps the rich and hurts our poorest and most vulnerable in society. 

What’s alarming was not the image search results but but rather it was Google’s Corporate response: “Sometimes unpleasant portrayals of sensitive subject matter online can affect what image search results appear for a given query. These results don’t reflect Google’s own opinions or beliefs – as a company, we strongly value a diversity of perspectives, ideas and cultures.” [6]

So racism doesn’t represent your company values but you won’t be responsible for racist images that appear in your search results? By the way the results for three black teenagers and three brown teenagers is relatively unchanged today over 5 years later. 

When faced with an ethical argument like gender discrimination in results Google will often point to free speech so they don’t need to follow the expectations associated with a public good or a public utility. Yet the public treats the service like a public good and a source of truth. 

How you’re creating a database of intentions

John Batelle who helped launch Wired and the Industry Standard described search as one of the most important cultural artifacts of our time: the database of intentions. The aggregate result list of every search result list, every click, and every path taken as a result. A massive click-stream database of needs, wants, and desires that can be exploited for all sorts of ends [7].

This fits very closely with what I’ve been referring to in terms of our Artificial Unconsciousness, the connection between what results you are shown and what you end up clicking on provides a detailed understanding of our unconscious desires and is why we often end up with search results, feeds, and videos that feed these desires.  

For example, when Safiya looked into why porn ranked among top result for black girls she found that online porn companies are ranking higher by building SEO optimized porn pages that match specific fetishes. So it’s giving some people exactly what they are looking for. The term is vague so other results do not get as many clicks. The only way to fix this is to generate enough media attention in order to force a change by Google (like no porn on the first page of search results please). Or to create SEO optimized pages that look similar to the porn result but direct them to something else. If we don’t learn Search Engine Optimization or at least how to draw the attention of companies such as Google these results will not change. Companies make money by being on the top of the search results, so unless there is a financial incentive we will likely always be behind. 

Safiya argues that Google should be broken up into multiple independent companies since it is exactly this merging of data that creates a very detailed picture of who we are. So it’s ironic that in 2015, Google combined the data from all it’s properties into a single merge database of desires. You probably saw this as the “One account All of Google” campaign. This included data from the online advertising platform DoubleClick whose clients include Microsoft, GM, Coca-Cola, Motorola, L’Oreal, Apple, Visa, and Nike meaning that most of your actions on webpages since 1995 have been added to your Google profile [8]. For many this was just another terms of service that we had to agree to, for Google this provided a history and evolution of desires over time.  

Prof Noble argues that governments that choose to make the Internet a public utility like water need to also look at making certain Internet services such as search public utilities as well. Otherwise they will introduce a public utility rife with racial and gender bias.

Speaking of politics, we’ve established that Ai has a very hard time critical thinking so content moderation often falls onto outsourced workers who use rubrics. UCLA professor Sarah Roberts, whom I had the pleasure of interviewing in the past showed how these rules tend to bias towards allowing Liberal perspectives while rejecting many others [9]. A Judiciary Committee Hearing with a Google Executive revealed that 99% of US political donations from Google executives and employees were to Democrats [10]. I’m sure many of you had read about the class action lawsuit by Donald Trump this week against big tech for the same types of bias, this should lead to so interesting discussions about political bias in Big Tech. 

Once we have a database of desires it’s easy for it to be exploited for all sorts of ends. Be they financial, competitive, or political. This is another reminder why your data and your privacy is so important. You are volunteering your vote and your voice and you won’t have rights to have these rankings changed. Ask what’s the financial incentive for this to be the top search result ranking?

From the bottom of our hearts, thank you for joining us on Ai Parenting Live Today. Next week we’ll do a deep dive into the history of media. 


  1. Algorithms of Oppression available on Amazon
  2. Search results for hair styles for work
  3. Three black teenagers controversy
  4. Bias in Libraries
  5. Google $270M fine in France
  6. Google’s Apology for bias in image searches
  7. John Battelle on the Database of Intentions
  8. Senator Demings asks Google about DoubleClick data merger
  9. Sarah T Roberts explores the practices of moderation in her book Behind the Screen
  10. Senator Hawley asks about Political donations during Judiciary Committee Hearing

Leave a Comment

Your email address will not be published. Required fields are marked *

Shopping Cart