Combating Anti-Blackness in the AI Community

Devin Guillory
University of California - Berkeley
dguillory@cs.berkeley.edu

Abstract
In response to a national and international awakening on the issues of
anti-Blackness and systemic discrimination, we have penned this piece to
serve as a resource for allies in the AI community who are wondering how
they can more effectively engage with dismantling racist systems. This
work aims to help elucidate areas where the AI community actively and
passively contributes to anti-Blackness and offers actionable items on ways
to reduce harm.

1 Introduction
“How did you go bankrupt?!” Bill asked. “Two ways,” Mike said. “Gradually and
then suddenly” – Hemmingway [11]. This oft referenced phrase aptly describes
how substantial changes that take a long time to develop can appear to happen
all at once. The extrajudicial killings of George Floyd, Breonna Taylor, Ahmaud
Arbrey, Tony McDade, and others, in combination with a poorly-managed
pandemic that is disproportionately damaging the Black community, has poured
gas on a growing blaze, calling on us to truly address the extent of anti-Black
systemic racism in the United States and abroad. Through this lens, many
in the AI community are asking: “What can I do to combat systemic racial
injustice?". The aim of this work is to help community members better identify
and understand the scale and scope of anti-Black bias within our AI community
and illustrate some concrete steps that members can take to help mitigate these
issues and build a more just community.
To summarize our contributions we first establish the necessity of recognizing
the scale and scope of anti-Blackness and how it permeates all of our institutions.
We then identify areas within academia where anti-Blackness is magnified or
reinforced and propose actions for faculty, graduate students, and conferences to
take to minimize these deficiencies.

2 Background
Students of optimization go through a stage where they see every problem as
an optimization problem. Similarly, students of racism must go through a stage

1
where they realize that anti-Black racism impacts every aspect of our society.
This acknowledgement is not sufficient to dismantle systems of racism; however,
it is a necessary first step and allows one to better appreciate the scope of the
issues that we are trying to solve. Issues such as access to healthcare[8], clean
air [4], quality education [10], credit [3], clean water [19], housing [22], public
transit [22], voter suppression [2], fair wages [7], and criminality [5] all have roots
in anti-Blackness.
We restrict our focus to the AI community and ways in which we contribute
to anti-Black racism. While the focus of recent protests have been centered
around anti-Black bias in the criminal justice system, we will expand our scope
to include more ways in which we propagate systemic harm: in our sources of
funding, in who we allow to participate in our community, in what problems we
address or exacerbate through our research and applications, and in what set of
principles we allow to guide us.

3 Examining Systems of Racism

We propose a simplifying model for examining the sources of systemic racism that
permeate our communities. The first source of systemic racism that we explore
are discrepancies in physical resources. This includes differences in wealth,
income, access to computing resources, access to clean air or water, healthcare,
and transportation, amongst other areas. Included in this list, though perhaps
more abstract, is time; time spent dealing with all other issues of race is also
time not explicitly spent improving skills that are more greatly valued than "the
ability to navigate racism." The second source that we will explore is social
resources. Who we know and who knows us are tremendous factors in explaining
the opportunities presented in our life [6]. Large percentages of open roles in
technology companies are filled via referrals; the compensation structures in
CS/AI are hidden and structured in such a way that you have to know someone
to know what is fair. Projects are rarely, if ever, completed with one individual
and who you know informs what you work on, what opportunities you are aware
of, who can vouch for you, and what social structures accept you as a member.
Jones et al.(2013) [15, 14] observe that 75% of White people have entirely White
social networks, while the average White persons social network is only 1%
Black. This remarkable racial stratification of social networks combined with the
prominent role that social networks play in our successes, lead to further racial
inequity. Lastly, we look at racial discrepancies in “measures”, i.e., anything that
is used to evaluate, punish, or reward individuals. At this moment, the most
prominent example of measures are discrepancies in policing. While Black and
White people use drugs at similar rates, those arrested and convicted for drug
offenses are disproportionately Black. Measures of aptitude, such as the SAT or
GRE, have also shown racial bias. The same can be said for teacher evaluations,
coding interviews, open-source project contributions, company promotion cycles,
and in-school suspension rates. An overwhelming number of measures of quality
have been shown to exhibit or allow for racial biases. These three categories

2
of systemic bias–physical resources, social resources, and measures–are not all
encompassing and many overlap (e.g., credit may be viewed as a measure of
quality or a physical resource), but they serve as a good starting point to dissect
systems of systemic bias.

4 Academia
When examining systems, whether they be systems of injustice and otherwise,
the importance of feedback mechanisms cannot be understated. Since Academia
serves as both the touchstone of the AI community as well as the developmental
environment for new AI talent, any biases or discrepancies within Academia
will be propagated to other places that AI touches and feed back into the next
generations of Academia where it may further enhance systemic biases. As such,
Academia is not only obligated to halt practices that disproportionately impact
the Black community, but also to repair the damage that has been done through
generations of neglect.

4.1 Faculty
Our faculty set the tone for our community. If our community is to ever achieve
equity, it will inevitably require buy-in and focused effort from our faculty. Our
professors decide which topics and prior research are highlighted in courses, who
is admitted into our programs, and what criteria is required to attain a degree.
They are tasked with simultaneously advancing the state of our field and training
the next generation of leaders. Consequently, it is pertinent for AI faculty to
fully understand all the ways that systemic racism pervades our field and help
identify ways to address it.

• Who is admitted?
We assert that there is no objective way to determine research potential,
and we believe that this is the primary goal of the admissions process.
Grade point averages, letters of recommendation, research publications,
and GRE scores are some of the major factors used in determining who
will be admitted to a graduate program. Studies have shown that GRE
scores are racially biased [17]. Publications and letters of recommendation
are largely influenced by the physical resources that are available to you
and the researchers that exist in your social network. Research is seldom a
task done in isolation and modern AI research frequently requires access
to substantial computational resources. For these reasons, measures of
research output and letters of recommendations systemically disadvantage
Black applicants and also feed back into the applicant pool in a way that
amplifies existing bias. These components help to produce a system where
it is hard to make meaningful changes along the lines of racial diversity
and inclusion.

3
 What can we do? The shortest answer is “take risks.” Right now, we
screen for students that are similar to previous successful students in
some manner. Admitting students whose experiences and achievements
look differently than previous cohorts may seem risky; however, years of
studying explore-exploit trade-offs should help us to more confidently take
risks. If we continue to wait for a flood of Black candidates whose paper
applications look identical to students that have been admitted for the last
K years, we are misunderstanding the fundamentals of systemic racism
and how it contributes to what opportunities and achievements candidates
with similar potentials and different backgrounds have access to.
• Who is mentored?
If the goal of admissions is to select based upon research potential, the
goal of mentorship is to develop research potential. We use mentorship as
a catch-all term for early-career researchers (graduate and undergraduate)
whom we help to train. A useful exercise may be to examine who you have
written letters of recommendation for within the last 5 years. Does the
diversity of this list reflect the diversity you would hope to see in the field?
Unless we are very intentional, it is likely that our mentorship list is even
less diverse than the department in which we work. The reasons for this
are varied and may range from potential mentees’ awareness of research
opportunities, comfort reaching out to faculty, availability of research
funding/ability to self-fund, and social connections with students within
our labs. Ways to change this number may include actively reaching out
to promising underrepresented students in our classrooms, participating in
targeted summer REU programs, and publicizing transparent protocols on
how interested students may get involved with our labs. If our campuses
are not diverse enough to generate a robust pool of diverse students, we
may establish closer relationships with other campuses. There are several
reasons why the number of students we mentor may not reflect the diversity
we seek to achieve, many of which have nothing to do with the qualification
or talents of diverse students. Sometimes the work of allies is to put in
extra effort to make themselves as accessible as possible.
• Who are your collaborators?
Current students should be your primary source of collaboration, yet who
you work with at other institutions matters greatly. While collaborating
with familiar faces and institutions has its benefits, it also means that
you are less likely to encounter diverse people through the course of your
research. This affects not only whose work you are familiar with, but also
who is familiar with your work.

• What topics are emphasized and de-emphasized?

What topics are valued in our community? Our community has done a
tremendous job of simultaneously being driven by empirical performance

4
 of applications and avoiding consequential discussions on the implications
of the applications that we focus on.
We write this piece in the midst of an intense international spotlight on anti-
Black violence enacted and enabled through our policing systems. Artificial
Intelligence systems have enabled more efficient and pervasive tracking,
monitoring, criminalization, and repression of Black people. Historically,
law enforcement organizations have deployed intense and technologically
sophisticated surveillance campaigns with the goals of dismantling move-
ments for the civil rights of Black people. We mention all of this to illustrate
that while the AI community has been reticent to proffer regulations or
even guidance on how AI technology should be ethically used, it has al-
ready been adapted and applied in various settings to exacerbate existing
inequalities within our society. Diversifying our field will not remedy the
harms that our systems have been agents in causing, but not addressing
these harms may further drive marginalized groups from our community.
This can reasonably be seen as being complicit in the harms our systems
produce.
While work on fairness and ethical Artificial Intelligence exists, it has not
received the same attention as more popular and controversial works such as
facial recognition. The most-cited example of ethical AI that I have found
has ∼ 1300 citations [1], while several facial recognition works, e.g., FaceNet,
from the same time period have ≥ 3000 [18, 21, 20]. Emphasising and
centering research which deals with the societal implications of Artificial
Intelligence is necessary to ensure that Artificial Intelligence has a positive
impact on our societies.
• What schools/labs are feeders?
Analyses of professors of Computer Science programs[12] indicate that
10 Computer Science programs are responsible for producing more than
50 percent of all CS faculty across the United States. Disproportionate
representation in these programs produces cascading effects that influence
all other departments in the U.S. Sourcing graduate students or lab visitors
primarily from institutions that have issues with retaining or attracting
Black students to their programs further compounds the problem of under-
representation because it allows these issues to spread to other schools. It
is tempting to assume our problem with representation stems from a lack
of suitable candidates. However, Black students making up ≥ 4% of CS
undergraduate degrees yet ≤ 1% of PhDs, indicates that PhD programs
suffer from a distinct drop-off that undergraduate diversity issues fail
to explain. The reasons for this drop off are complex, but we believe
one of them might be based on which schools serve as feeder schools
into these PhD programs. Despite having 1/10th the students, the 101
HBCUs produce more Black Computer Science Bachelor’s degrees than
the 115 R1 institutions in the U.S. PhD applications are designed to favor
research experiences thus, the fact that such a large portion of Black CS

5
 undergraduates stem from institutions where research is not funded and/or
prioritized as much as it is at R1 institutions puts Black applicants at a
fundamental disadvantage. To conduct a quick self audit, you may look at
the 11 HBCUs listed below, which graduate the largest numbers of Black
CS students, and see how connected they are to your department [9, 16].

– North Carolina A & T

– Southern University
– Norfolk State University
– Johnson C. Smith University
– Florida A & M
– Alabama A & M
– South Carolina State University
– Lane College
– Rust College
– University of Arkansas Pine Bluff
– Virginia State University
– Morehouse College

• What companies are students recruited from/funneled into?

Despite the aspirational nature of Artificial Intelligence’s ideals, applica-
tions of AI have inflicted real harm on Black people. How you engage with
companies that profit from anti-Black policies matters. Who do you take
funding from? Who do you allow to recruit on campus? Who do you have
informal relationships with? Where do your students end up working? I
offer these as provocative questions and do not purport to offer prescriptive
answers. However, I think it is important for our organizations to have
these conversations and work on deciding our own ethical lines, if for no
other reason than to mitigate the risk of scandal for money that we have
accepted from nefarious sources.
• Who is hired?
As we climb the ladder of academia, institutions become more and more
risk-averse. Whereas there might be a willingness to take a chance on an
undergraduate or summer visitor, this willingness drops when it comes
to graduate students, post-docs, and especially faculty. If we want to
truly move the needle on anti-Blackness in our community, the hiring
and retention of Black professors must be addressed. The first stage of
a research project is establishing relationships and, historically speaking,
Black faculty have done a much better job of establishing relationships
with Black students than other faculty. Fixing this discrepancy is not just
about who is visible at the top, but about who is better positioned to
succeed at all levels of your pipeline.

6
 Producing an equitable hiring process will require evaluating all aspects of
the procedure with an eye for where biases may manifest. Are diversity
efforts appropriately valued with sufficient rigor and weighting? Are we
drawing from representative pipelines and accounting for discrepancies in
physical and social resources?
• How much do you pay?
Systems that require financial “sacrifice" in order to attain professional
advancement are extremely effective at removing high achieving Black
people from the leadership pipeline. The length of sacrifice and the drastic
difference in earning potential and realized compensation of Computer
Science PhD students makes this community one of the worst known
examples of this discriminatory mechanism.
While discrepancies in graduate student pay affect all students, the 10x
difference in median wealth between Black and White families means that
Black students or potential students will be disproportionately impacted
by a drop in income. This design choice, which has gone unaddressed
for generations, creates a scenario in which a large number of qualified
and interested Black students simply cannot afford to attend graduate
school. Some schools and programs are aware that this opportunity cost
is driving talented researchers away from academia and have developed
quasi-official corporate partnerships that provide AI graduate students
with supplemental incomes. However, these relationships are often opaque
and invisible to students and communities without close insider connections.
The unfortunate impact of this is that even if your program happens to
be one that has made significant improvements to the compensation and
quality of life of expected graduate students, potential Black students
who would benefit the most from these changes are likely unaware of any
improvements that you have made in this arena.
• What is the campus environment?
Are the Black students on your campus adequately supported? This is a
complicated question to answer and perhaps the best method of answering
this question is by talking to your current and former Black students. A lot
of students make it through graduate programs with traumatic experiences
and unless there are Black professors, may not trust any faculty members
enough to be candid about the problems in their environment. If and when
you hear complaints from Black students, recognize that sharing these
issues often constitutes a risk for these students and it is imperative to
listen without becoming defensive.
• What programs and resources are in-place?
Systems that perpetuate anti-Black bias exist within the AI community
just as strongly as they do in most other spaces. What resources have you
committed to combating them? How has your lab, your department, your
university, and larger community organizations in which you are a member

7
 addressed issues related to systemic anti-Blackness? Instead of proclaiming
your beliefs and convictions, demonstrate them through investments and
actions. Making significant changes in racial equity in our institutions is
achievable but will require tangible effort from people other than those
who are discriminated against.

4.2 Graduate Students

• Who benefits from diversity?
While substantial research has shown that diverse teams achieve better
performance[13], we reject this predatory view of diversity in which the
worth of underrepresented people is tied to their value add to in-group
members. We argue for combating anti-Blackness through the lens of
justice. As such, all members of the community should be invested in
developing a more inclusive and less discriminatory environment.
• Who works towards diversity?
Black graduate students (and faculty) shoulder an extraordinary amount of
the burden of "diversifying" university campuses. One of the most impactful
actions an ally can take is help to do this work. This participation will
not only help to create a more diverse campus, but will also allow your
Black peers to spend more time on their research and help them to attain
successful graduate school careers. Understanding how to best assist in
anti-Racism efforts may seem challenging or intimidating at first, but
many of the necessary tasks are relatively simple and do not require an
advanced degree in Black studies to result in meaningful contributions.
Show up early, ask what you can do to support, volunteer to take over
a non-leadership role. For example: "Hey, I would like to help secure
food for all the NSBE meetings this semester." While the current levels of
attention on anti-racism might be fleeting, the magnitude of effort required
for change is not; show up consistently and be willing to be lead.
• Who collaborates or studies with whom?
Lack of social integration is a significant factor in disparate outcomes and
experiences Black graduate students face in Academia. Lunches, dinners,
happy hours, and game nights are activities that bear no direct relationship
with technical merits but significantly contribute to the sense of safety
and belonging that Black students in your program seek. As with most
professions, social interactions often lead to professional opportunities.
Within graduate school, study groups, research collaborations, and work-
shop organizers are often selected through a mix of social and academic ties.
Reforming our social networks, as well as those of our entire community, is
a formidable task, but while we are in the process of doing that we can
actively acknowledge when biased social ties are influencing professional
advancement and do our best to mitigate this. To state it frankly, put
in the extra effort to know your Black colleagues’ skills and interests and

8
 when opportunities present themselves that might align well with these
students be sure that they are given fair consideration and not excluded
because they are not in the right social groups.
• Which undergrads do you mentor?
Who we invest our time in matters. The undergraduates that we mentor
today will likely constitute the next generation of graduate students so it is
imperative that we intentionally invest our time and mentorship capacity
in diverse student populations. Oftentimes, the mechanisms of how to
become involved in undergraduate research are hidden and subjective. One
of the biggest levers we have to increase diversity in graduate school is to
increase diversity in undergraduate research. To do this well, we should
focus on both encouraging more underrepresented minorities to participate
in undergraduate research and recruiting more graduate students to serve
as mentors.
Additionally, it is necessary that we focus on finding funding for under-
graduate researchers. Unpaid internships have a long history in the U.S.
and one of their functions is to reinforce racial inequalities. We must work
diligently at prohibiting this practice at our institutions. The ability to
perform unpaid labor is often a privilege only afforded to those of substan-
tial financial means. The following framing may help others more fully
understand the role that compensation for undergraduate researchers plays
in amplifying socio-economic inequality in our field. Let C = U + P repre-
sent the total capacity for undergraduate mentorship within a department.
While U and P represent the unpaid and paid researchers. The unpaid
research slots, U , are reserved exclusively for high socio-economic students
and the paid slots P are to be split between students of all socio-economic
backgrounds. Ignoring all other sources of systemic bias, this arrangement
alone produce a system that significantly disadvantages Black students
while serving as the primary pool for future graduate students. Calculating
the true numbers for P and U in your institution should be a relatively
straightforward process that can effectively illustrate just how skewed some
of our current systems are.
• Who does the "devil’s advocate" serve?
Racism exists. It exists in our neighborhoods, in our departments, and in
our labs. Conversations that question its existence are either intellectually
lazy or conducted with ill intent. To have productive conversations and
institute changes, we must avoid continually rehashing whether this well-
documented phenomenon is a well-documented phenomenon. Just as a
budget meeting filled with constant debate about the existence of currency
is unproductive, so too is a discussion around anti-Blackness in which
we refuse to name it or engage in meaningless debate over its existence.
To address anti-Blackness, we cannot continue to indulge the whims of
misinformed individuals or bad actors and offer them as much space as
those working through active solutions.

9
5 Discussion
Through this work, we discuss several areas within the Artificial Intelligence
community where systems perpetuate Anti-Blackness. This work falls woefully
short of being comprehensive and intentionally does not discuss systems of racism
within industry, conferences, or pipelines of capital. This work also does not
address issues of sexism within our community, whose effects are compounded
with racism to disproportionately impact Black women. The ultimate goal of this
work is to assist in creating a more equitable Artificial Intelligence community
whose broader impact includes reducing Anti-Blackness in our society. While
not simple, we wholeheartedly believe that this is a tractable task and call upon
our community to help make this happen.

References
[1] Solon Barocas and Andrew D Selbst. Big data’s disparate impact. Calif. L.
Rev., 104:671, 2016.
[2] Keith G Bentele and Erin E O’brien. Jim crow 2.0? why states consider and
adopt restrictive voter access policies. Perspectives on Politics, 11(4):1088–
1116, 2013.

[3] David G Blanchflower, Phillip B Levine, and David J Zimmerman. Dis-

crimination in the small-business credit market. Review of Economics and
Statistics, 85(4):930–943, 2003.
[4] Mercedes A Bravo, Rebecca Anthopolos, Michelle L Bell, and Marie Lynn
Miranda. Racial isolation and exposure to airborne particulate matter and
ozone in understudied us populations: environmental justice applications of
downscaled numerical model output. Environment international, 92:247–255,
2016.
[5] Rose M Brewer and Nancy A Heitzeg. The racialization of crime and
punishment: Criminal justice, color-blind racism, and the political economy
of the prison industrial complex. American Behavioral Scientist, 51(5):625–
644, 2008.
[6] Antoni Calvo-Armengol and Matthew O Jackson. The effects of social net-
works on employment and inequality. American economic review, 94(3):426–
454, 2004.
[7] Major G Coleman. Job skill and black male wage discrimination. Social
Science Quarterly, 84(4):892–906, 2003.
[8] Joe Feagin and Zinobia Bennefield. Systemic racism and us health care.
Social science & medicine, 103:7–14, 2014.

10
 [9] National Center for Science and Engineering Statistics (US)(NCSES).
Women, minorities, and Persons with Disabilities in Science and Engi-
neering, 2018.
[10] Shaun R Harper, Lori D Patton, and Ontario S Wooden. Access and equity
for african american students in higher education: A critical race historical
analysis of policy efforts. The Journal of Higher Education, 80(4):389–414,
2009.
[11] Ernest Hemingway. The sun also rises. Simon and Schuster, 1926.
[12] Jeff Huang. Analysis of Over 2,000 Computer Science Professors at Top
Universities, May 2014 (accessed June, 2020).

[13] Vivian Hunt, Dennis Layton, and Sara Prince. Diversity matters. McKinsey
& Company, 1(1):15–29, 2015.
[14] Robert P Jones. Self-segregation: Why it’s so hard for whites to understand
ferguson. The Atlantic, 21, 2014.

[15] Robert P Jones, Daniel Cox, and Juhem Navarro-Rivera. The 2013 american
values survey: In search of libertarians in america. Public Religion Research
Institute, 2013.
[16] Digital Learning Lab. Directory of HBCU Computer Science Programs
Bachelors & Masters, Version 2.0, 2016.

[17] Casey Miller and Keivan Stassun. A test that fails. Nature, 510(7504):303–
304, 2014.
[18] Omkar M Parkhi, Andrea Vedaldi, and Andrew Zisserman. Deep face
recognition. 2015.

[19] Laura Pulido. Flint, environmental racism, and racial capitalism, 2016.
[20] Florian Schroff, Dmitry Kalenichenko, and James Philbin. Facenet: A
unified embedding for face recognition and clustering. In Proceedings of the
IEEE conference on computer vision and pattern recognition, pages 815–823,
2015.

[21] Yaniv Taigman, Ming Yang, Marc’Aurelio Ranzato, and Lior Wolf. Deepface:
Closing the gap to human-level performance in face verification. In Pro-
ceedings of the IEEE conference on computer vision and pattern recognition,
pages 1701–1708, 2014.
[22] Richard Williams, Reynold Nesiba, and Eileen Diaz McConnell. The chang-
ing face of inequality in home mortgage lending. Social problems, 52(2):181–
208, 2005.

11