What is Discrimination?

 

On Thursday December 3rd, Google fired their leading ethical AI researcher Timnit Gebru. It shook the technology and AI research community. At the time of writing this, over 1200 Google employees have signed a solidarity letter condemning the firing and asking for more  transparency on the situation. Google firing Timnit, who is one of the leaders of her field, might be deemed justifiable for a plethora of reasons. The ones given by Jeff Dean was that one of her emails to a “Women and Allies” group inside Google wasn’t becoming of a google manager and that her paper, which prompted the email, didn’t properly give Google credit for it’s efforts to mitigate the criticisms about their ethical AI practices. Feel free to read Dr. Gebru and Jeff’s emails here.

The question I want to answer on this post is: is this discrimination? Dr Gebru is a black woman and Jeff Dean is a white man. According to Jeff’s email, there were “rules” that weren’t followed that might have justified her dismissal. In this exercise I want to breakdown what is discrimination, is this an instance of discrimination and, if so, how can we fight discrimination like this.

During the fallout from this happening was a tweet from a McGill prof Nicolas Le Roux, that I felt was aptly put:

Now might be a good time to remind everyone that the easiest way to discriminate is to make stringent rules, then to decide when and for whom to enforce them.
My submissions were always checked for disclosure of sensitive material, never for the quality of the literature review.— Nicolas Le Roux (@le_roux_nicolas) December 3, 2020

The simplest method of discrimination is to make stringent rules that are almost impossible to follow and then selectively decide when to enforce these rules.

This incident presents us with a case study on discrimination and, if we’re being honest, gaslighting. The benefit of this implementation of discrimination is that when you apply your discrimination you can point to the victim and say “Well you didn’t follow the rules”. This shifts the burden from the discriminator to the victim and, if anything, uplifts the discriminator as a scrupulous rule follower and “law and order candidate”.

Let’s break this down into the a couple sections: what is the ‘stringent’ rule and what is the actual expectation when that rule is broken.

What is the stringent rule? There are a few different takes here. There are three major things that Jeff’s email referenced: Issues surrounding her paper, issuing an ultimatum for “working on an end date” and her email to the women’s group criticizing Google’s practices. He said her paper “ignored too much relevant research” while not following the “internal review process” criteria. Dean’s email also stated that she requested the name of her reviewers and gave an ultimatum “that if [Google] didn’t meet these demands, she would leave Google and work on an end date”. Dean also goes on to mention that her email to the women and Allies group was “inconsistent with the expectations of a Google manager.”

So that looks like the rules were broken by these actions, right? That is, unless the actual practice and expectation is for people to do these things and continue working or even get rewarded for that behaviour.

On the merits of her research paper, the 6 other contributors mentioned that they have referenced 128 relevant sources and when pressed about the submission time and procedure of the paper:

I used to work as part of this process Jeff is talking about when I did google AI comms. The white guy AI researchers rarely followed the process & for some unknowable reason they didn’t get fired. https://t.co/qDpJUL04zD

— William Fitzgerald (@william_fitz) December 3, 2020

On making ultimatums for resignation:

I can confirm that — as a white guy at Google Research that submitted a condition letter of resignation — I received dramatically different results.

I was set up with a sequence of meetings with management — culminating in the (very disappointing) final boss of Dean.

— Jack Poulson (@_jack_poulson) December 3, 2020

On being critical of Google’s hiring practices and Google in general:

Lastly, let’s use this definition to assess a number of other claims of discrimination and systematic racism. We’ll list out a number of discriminatory instances and try to answer those two questions, what’s the stringent rule and what is the usual expectation when this is broken.

Police brutality

Stringent Rule: Comply with the police officer when they ask you to do something. Just obeying and not pushing back against police instruction means that you wont be antagonized and especially shot by a police officer. This can be stringent because sometimes, they’re yelling conflicting instructions. You can’t ask any questions. Just comply.

Actual Experience: Although there aren’t statistics on non-compliance with police interactions by race, we can use surrounding data points to infer some details. For example, in Pew Research Center’s survey earlier this year, a majority of all adults said that, in dealing with police, blacks are generally treated less fairly than whites (84% of black and 63% of whites). This means perceptions reflect a bias in implementation. The San Francisco Attorney General filed a report that found “although Black people accounted for less than 15 percent of all stops in 2015, they accounted for over 42 percent of all non-consent searches following stops.” Of all people searched without consent, Black and Hispanic people had the lowest [almost half as many] ‘hit rates’ as their white counterparts. PBS citing a study conducted in Oakland found that when evaluating police responses to drivers “white drivers were 57 percent more likely to have an exchange filled with a highest degree of respectful language, whereas black drivers were 61 percent more likely to experience an exchange that fell into the category of least respectful.”. This means when policemen encountered black drivers, they were more likely to show disrespect to the driver – ultimately influencing the driver’s ability to comply. A DOJ investigation into Ferguson’s police department found that with black residents officers were “inclined to interpret the exercise of free-speech rights as unlawful disobedience, innocent movements as physical threats, indications of mental or physical illness as belligerence.” If we want to look at a case study of who is most likely to comply, lets asses police officers dealing with other police officers.  A government taskforce found that 9 out of the 10 off-duty officers killed by other officers in the United States since 1982 were black or Latino.

Redlining/Housing Discrimination

Stringent Rule: You need a pre-approved mortgage. You need a certain level of income before being considered for a house. Your relator is expected to place you in the best neighborhood for you and your family.

Actual Experience: From what we can gather,  lenders deny mortgages for Black applicants at a rate 80% higher than that of White applicants, according to 2020 data from the Home Mortgage Disclosure Act. This sits at a ~30% rate compared to the national average of 17%. According to a study at UC-Berkeley found that with all else being equal, traditional face-to-face services are 5% more likely to reject Black and Latinx applicants. For the borrowers that do get loans, lenders charge otherwise-equivalent Latinx/African-American borrowers 7.9 basis points higher rates to refinance their mortgages, which costs minority borrowers ~$765M yearly!

Hiring and University Acceptance practices

Stringent Rule: “We just hire the best people for the job, regardless of race.” You didn’t get the job or gain acceptance because the criteria is very high and you didn’t meet that criteria.

Actual Experience: When looking at admittance for jobs or schools, we consider “affirmative action” as the ‘unfair advantage’ for racial diversity. What we don’t factor in is how legacy, sport scholarship and “networking” favor the status quo of racial homogeneity in institutions. It’s been estimated that “legacies” receive a bonus equal to about 160 SAT points and athletes a bonus of about 200 points. Not to mention how the highest beneficiaries of affirmative action are actually white women and not BIPOC. When examining bias in workplace admissions, results from a Poverty Action Lab experiment found that résumés with white-sounding names received 50 percent more callbacks than those with black-sounding names, while all other things remain equal.

A cheeky bonus example, see if you can spot the rule and exemption:

My university calls my PhD a “terminal degree,” and everyone has to have one. The only exceptions I know of are artists with MFAs.

I knew a (white male) professor of Epstein’s generation at my university. He got his job when his advisor called a friend at USC. That’s it.

— viet CALL ME DR. nguyen (@viet_t_nguyen) December 13, 2020

Now that we see this pattern and can recognize it, the onus is on us to break it. The most direct way to combat this trend is by calling out the ‘stringent’ rule and especially stopping and aggressively combatting the selective application. One effective way to call out the rule is for non-minority allies to own up to stringent exclusionary rules. This can be seen when people in power make roles more accessible, or allies being vocal about the exceptions to rules that were made for them (ie. acknowledging privilege) instead of dismissing the unmerited favor. That’s a habit that needs to be owned daily but during acute situations of discrimination, like what occurred at Google AI, we need to call out the pattern of selective application and stand up for victims even at the detriment of our privilege. When you see BIPOC being the recipient of rule-based discipline, it’s important to step back and think about if this rule is applied consistently and fairly across your organization. If not, use your voice and power to advocate for removing the rule and absolving and supporting your colleague. This is especially important for those in leadership positions. With the knowledge to identify discrimination comes a responsibility to pursue fairness, because our best chance of winning is when we all have a seat at the table.