Google Racial Bias Created by Search Users Shows up on AdWords

The British police have an interesting interrogation method that’s a ramped-up version of the 20-question game. Namely they take turns to ask you the same thing in as many different ways as you can imagine. The principle behind it is that if you ask enough questions, enough times of the same subject, eventually the real truth begins to color their replies.

Search is no different but it offers a twist: the interrogator and the subject here are heuristically linked through intent. In other words what the search engine reflects back at us is the direct result of the search query we have input there. At the risk of mixing metaphors here, Google search is not unlike the “Mirror, mirror on the wall…” where the question asked, reveals (in the fairy tale) the Evil Queen’s narcissistic obsession.

A new study carried out by Harvard professor, Latanya Sweeney, indicates that Google Ads that appear when racially biased names are used in Google search, predominantly display adverts that have to do with criminal records, identity verification services and criminal background information services.

The key to understanding the importance of the findings lies in understanding the mechanics of search and the Google Adwords system. Search results in Google are based upon the search engine’s understanding of the intention of the search query. To achieve this the Google search algorithm cross-checks an incredible amount of information that relates to the statistics of similar searches and the search profiles of those who carried them out, the search profile of the person asking the search query and the behavioral pattern of those who have clicked on results in the past. In addition it adds localization and, when logged into your personal profile, social connections and it carries out an analysis of the social signal of specific websites, their content, links profile and authority. It does all this in thousandths of a second in order to deliver results fast.

The Google AdWords system allows the end user to specify keywords that will trigger the advert and make it appear. This is also subject to a degree of optimization within Google’s data world so that adverts that get a response (i.e. a click for the advertiser and a profit for Google) to a specific query are weighed to appear more for those queries than adverts that appear and do not get a response.

The two systems are set up to be independent but contextual so that relevant ads appear in response to specific search queries. The idea behind this is that relevancy leads to better click through rates from ads and overall end-user satisfaction.

Look at the example of Google Adwords ads that show up for the US-based query for: "Latanya Sweeney"

Google Adwords Show Racial Bias created by Search Users

So, What’s The Problem?

The problem lies in the set up, or rather what the set up is designed to deliver: higher end-user satisfaction. This means that it learns from successful search queries and fine-tunes the ads displayed with specific search terms to deliver what is most likely to appeal to the end user.

In our particular case, highlighted by the study this is that when names that are ethnically black are input in search the adverts associated with them suggest that they have a criminal record, have been arrested or perhaps should not be trusted.

Like the mirror on the wall and the Evil Queen’s obsession with her looks, this is a bias that is revealed by our own interaction with search. What Google delivers is what is expected to work according to the data it has from us. So, what it reflects back at us are our own biases.

Now, I understand enough about search to know that it is not universally that clear cut. Semantic search has a nuanced and ever-developing serendipity discovery algorithm designed, specifically to help us break free of our own narrow search bubbles. That however kicks in under specific conditions. When there is an overwhelmingly clear-cut answer to a search query and where AdWords are concerned, serendipity does not play a massive role.

Knowledge is Power

The difference between us and the Evil Queen is that in her case the mirror she used was not subject to the same degree of analysis as Google search results are.

As a species we are unique in seeking to better understand ourselves and then acting upon that knowledge.

We have come a long way from the times when race and skin color produced an instant knee-jerk response in us. Collectively, we may not have come quite long enough but, at least, now, knowing this, we can begin to take the steps necessary to address it.

With resumes and online profiles playing an ever increasing importance in job-seeking, online trust and our sense of identity, to fail to do anything is to fail to play a role in the digital world we want to inhabit.

UPDATE February 2014: For a fresh example of how search can reflect back our own biases check out the case of and Autocomplete.

What You Missed

If You Want to Get a Job Learn About SEO Bots
How to Use Social Media to Get a Job
Social Media and the Death of the Resume
How Content Became the Currency of the Connected Economy
How to Build Trust in Your Business (Podcast)
Our village communities revolved around ‘influencers’ and ‘authority figures’
How to Build a Culture of Trust in Your Business (Podcast)