CSWR Spring 2022 




COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  |   65   

Resist, Regulate, 
Reimagine, and 
Reinforce:  
How Social 
Workers can 
Advocate for 
Digital Inclusion

SARAH E. DILLARD



66  |  COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  

RESIST, REGULATE, REIMAGINE, AND REINFORCE

ABSTRACT

(SNVYP[OTZ�OH]L�ILJVTL�TVYL�JVTWSL_��JYLH[PUN�HY[PÄJPHS�PU[LSSPNLUJL�
(AI) with the hope it will match human decision making. They are now 
being used behind-the-scenes in areas such as healthcare, housing 
and employment, and criminal justice. These computer formulas were 
JYLH[LK�I`�H�WYP]PSLNLK�ZL[�VM�PUKP]PK\HSZ�^OV�VM[LU�WYPVYP[PaLK�WYVÄ[�
and growth over privacy and protection. This has led to gross injustices 
[OH[�OH]L�WYL]LU[LK�THYNPUHSPaLK�JVTT\UP[PLZ�MYVT�YLJLP]PUN�JHYL��
ÄUKPUN�QVIZ��VY�NHPUPUN�MYLLKVT��:VJPHS�^VYRLYZ�T\Z[�IL�HISL�[V�KPNP[HSS`�
advocate for their clients. Resisting these technologies, regulating them 
through legislation, reimagining the role one can play, and reinforcing 
what is already experienced in day-to-day interactions with AI are all 
ways social workers can be involved in creating a world that is digitally 
inclusive.



COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  |   67   

SARAH E. DILLARD

S
VJPHS�^VYRLYZ�HYL�HK]VJH[LZ�MVY�THYNPUHSPaLK�WLVWSL��
Clinicians, case workers, policy makers, and others all 
interact with populations that are considered “protected” 
VY�¸ZLUZP[P]L¹�PU�[OL�ÄLSK�VM�HY[PÄJPHS�PU[LSSPNLUJL��(0���;OLZL�

groups include, but are not limited to, women, immigrants, people of 
JVSVY��76*���3.);80��WLVWSL��[OVZL�^P[O�KPZHIPSP[PLZ��HUK�[OVZL�MYVT�
low socioeconomic status (SES).

As AI becomes more integrated into our everyday lives, algorithms are 
PUÅ\LUJPUN�H�]HYPL[`�VM�KLJPZPVUZ�WYL]PV\ZS`�OHUKSLK�I`�SP]L�O\THUZ!�
criminal risk assessment, health insurance costs, home loan eligibility, 
and employee resume review are just a few examples. 

What can a social worker do when their client has been labeled high-
YPZR�MVY�YL�VɈLUZL�KLZWP[L�UV[�OH]PUN�H�OPZ[VY`�VM�]PVSLUJL&�>OH[�JHU�H�
ZVJPHS�^VYRLY�KV�^OLU�[OLPY�JSPLU[»Z�OVTL�JHYL�ILULÄ[Z�OH]L�ILLU�J\[�
PU�OHSM&�>OH[�JHU�H�ZVJPHS�^VYRLY�KV�PM�[OLPY�JSPLU[�OHZ�OPNO�\[PSP[`�YH[LZ�
ILJH\ZL�[OLPY�JYLKP[�ZJVYL�MHJ[VYLK�PU�[OLPY�ZVJPHS�TLKPH�HJ[P]P[`&

;OPZ�WHWLY�^PSS�JV]LY�[OL�OPZ[VY`�VM�HY[PÄJPHS�PU[LSSPNLUJL��H�WHY[PHS�
overview of its current implementation and sources of bias, and 
four ways social workers can advocate for digital inclusion: through 
resistance, regulation, reimagination, and reinforcement. As 
technology is changing rapidly, the examples that follow may already 
have changed in the time that has passed between the writing of this 
HY[PJSL�HUK�P[Z�W\ISPJH[PVU��;OL�JVTT\UP[`�VYNHUPaH[PVUZ�HUK�TV]LTLU[Z�
highlighted can therefore be explored for the most recent updates.

HISTORY OF ARTIFICIAL INTELLIGENCE

 An algorithm is a set of rules or calculations—like a recipe—
[OH[�T\Z[�IL�MVSSV^LK�PU�VYKLY�[V�ZVS]L�H�WYVISLT��6_MVYK�(K]HUJLK�
Learner’s Dictionary, n.d.). In use for centuries, they were originally 
developed to aid in the construction of buildings, agriculture, and 
commerce in order to to streamline processes and create a uniform 



68  |  COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  

method for getting results (Ausiello, 2013). It was not until the mid-1950s 
[OH[�HY[PÄJPHS�PU[LSSPNLUJL�ILJHTL�WVZZPISL��H�[LJOUVSVN`�KLZPNULK�[V�
YLÅLJ[�[OL�WYVISLT�ZVS]PUN�JHWHIPSP[PLZ�VM�O\THUZ��(U`VOH���������<ZPUN�
algorithms as the structure and data as the substance, technologists 
started to use AI as a substitute for human analysis and interpretation.

6UJL�JVTW\[LYZ�PUJYLHZLK�PU�Z[VYHNL�HUK�WYVJLZZPUN�WV^LY��(0�^HZ�
HISL�[V�ÅV\YPZO��4VYL�KH[H�JV\SK�IL�Z[VYLK�VU�JVTW\[LYZ�HUK�PU�
turn used to train machine learning (ML) algorithms (Anyoha, 2017). 
Computers became more adept at problem solving and interpreting 
language. As they became cheaper, more institutions became involved 
in research and development. 

Today, we are seeing AI and big data—a term used to describe the vast 
amount of online information that companies are able to garner on an 
individual—come together (Bean, 2017). Digital footprints consisting of 
all the data a person has following them online are thus being used for 
(0�THJOPUL�SLHYUPUN�PU�[OL�ÄLSKZ�VM�IHURPUN��OLHS[OJHYL��JYPTPUHS�Q\Z[PJL��
employment, and national security (Anyoha, 2017; Benjamin, 2019). 

AREAS OF IMPACT

Technologists wrongfully assumed that a computer would eliminate 
bias by being based in formulas and mathematical calculations. 
;OLZL�JHSJ\SH[PVUZ�^LYL�PTWSLTLU[LK�[V�PTWYV]L�HUK�VW[PTPaL�O\THU�
decision-making, especially in areas such as criminal justice, in which 
judges were making subjective decisions (Eckhouse et al., 2019). 
/V^L]LY��[OL�OPZ[VYPJ�KPZJYPTPUH[PVU�VM�THYNPUHSPaLK�NYV\WZ�PZ�VUS`�
reproduced and reinforced by these systems in what experts are calling 
“The New Jim Code” or “Coded Bias” (Benjamin, 2019; Buolamwini 
L[�HS����������6UL�RL`�L_HTWSL�VM�[OPZ�PZ�HU�HSNVYP[OT�ILPUN�\ZLK�[V�
KL[LYTPUL�H�WLYZVU»Z�YPZR�VM�YLVɈLUKPUN�HM[LY�HYYLZ[��+L]LSVWLYZ�ILSPL]LK�
that by not including race as a data point, the machine would not 
WYVK\JL�YHJPHSS`�IPHZLK�YLZ\S[Z��;OL`�KPK��OV^L]LY��PUJS\KL�WLVWSL»Z�aPW�
codes. Geographic data—such as the area someone lives in—is often a 
proxy for race due to residential segregation and redlining (Eckhouse et 
al., 2019). By not being familiar with this history, technologists created a 

RESIST, REGULATE, REIMAGINE, AND REINFORCE



COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  |   69   

system that reinforced existing biases. Accountability must be taken by 
companies instead of operating under the guise of expertise. 

The following sections outline three areas where algorithms have 
NYLH[S`�OHYTLK�THYNPUHSPaLK�JVTT\UP[PLZ�HUK�WLYWL[\H[LK�Z`Z[LTPJ�
oppression.

HOUSING AND EMPLOYMENT 
CREDIT SCORE AND HOMEOWNERSHIP

Algorithms have been used to determine credit score since the 1980s 
(Trainor, 2015). Before that, lenders would keep their own records of who 
they believed was “trustworthy” enough to receive a loan, often barring 
THYNPUHSPaLK�JVTT\UP[PLZ�Z\JO�HZ�)SHJR��3H[PU_��HUK�1L^PZO�PKLU[PM`PUN�
WLVWSL�MYVT�WHY[PJPWH[PUN��;VKH �̀�JYLKP[�ZJVYLZ�HɈLJ[�THU`�MHJL[Z�VM�
everyday life, including loan eligibility, home ownership, utility rates, and 
social standing. 

;OL�SHYNLZ[�JYLKP[�ZJVYL�JVTWHUPLZ·,_WLYPHU��,X\PMH_��HUK�
Transunion—use data such as bill payment history, employment 
information, and current debt to determine one’s score. They also factor 
in child support payment history, arrest and incarceration history record, 
and app usage. The companies have not released information on what 
metrics are used to determine the weight of each category (Hao, 2020). 

With the rise of big data, smaller credit score companies are beginning 
to use data outside the typical sources used by larger companies. This 
includes social media information (likes, friends, locations, and posts), 
the amount of time you spend on their website, and what percent 
of income is spent on rent given geographic location (Hurley et al., 
�������(U`�JYLKP[�YLWVY[PUN�HNLUJ`��*9(��JHU�YLX\LZ[�KH[H�MYVT�ZVJPHS�
media or data scraping companies (entities one can hire/pay to collect 
vast amounts of information from people online) in order to build their 
reports. Despite the Fair Credit Reporting Act of 1970 outlining what 
KH[H�JHU�IL�\ZLK��MLKLYHS�SLNPZSH[PVU�OHZ�UV[�JH\NO[�\W�^P[O�OV^�X\PJRS`�
technology is changing and becoming more integrated in our everyday 
lives. Social media likes, for example, are not mentioned as an accepted 
or prohibited datum anywhere in the bill. 

SARAH E. DILLARD



70  |  COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  

There is a large racial discrepancy between those with good vs. bad 
credit (Singletary, 2020). This directly correlates with the biased history 
of credit scoring and systemic oppression that is inherent in the rating. If 
UV�VUL�PZ�^PSSPUN�[V�WYV]PKL�H�SVHU��I\PSKPUN�JYLKP[�ILJVTLZ�ZPNUPÄJHU[S`�
TVYL�KPɉJ\S[��9LKSPUPUN�VM�)SHJR�HUK�3H[PU_�ULPNOIVYOVVKZ�THKL�P[�
PTWVZZPISL�MVY�MHTPSPLZ�[V�X\HSPM`�MVY�TVY[NHNLZ�I`�ZHUJ[PVUPUN�[OLZL�
areas as “risky” for lending (Lerner, 2020). At the same time, these 
MHTPSPLZ�WH`�HU�H]LYHNL�VM�����TVYL�PU�[H_LZ�JVTWHYLK�[V�^OP[L�
MHTPSPLZ�SP]PUN�PU�OVTLZ�VM�LX\P]HSLU[�]HS\L��

0U�[OL�<:��VUS`�������VM�)SHJR�OV\ZLOVSKZ�V^U�[OLPY�OVTL�JVTWHYLK�
[V�������VM�^OP[L�OV\ZLOVSKZ��PU�WHY[�K\L�[V�YHJPZ[�YLKSPUPUN��*HTWPZP��
2021). Since credit score focuses on ownership through mortgages, the 
majority of Black Americans do not have this assurance to add into the 
algorithm. If rental payments, however, were taken into consideration, a 
NYLH[LY�Z^H[O�VM�THYNPUHSPaLK�PUKP]PK\HSZ�^V\SK�IL�HISL�[V�I\PSK�IL[[LY�
credit. 

;OL�SHJR�VM�OVTLV^ULYZOPW�PU�THYNPUHSPaLK�JVTT\UP[PLZ�PZ�WLYWL[\H[LK�
by AI in other ways. Without any privacy regulations or civil rights laws 
PU�WSHJL�[V�YLN\SH[L�[OL�\ZL�VM�LSLJ[YVUPJ�KH[H��SLUKLYZ�HYL�HISL�[V�ÄS[LY�
candidates using racial proxy data, resulting in digital discrimination 
and continued historic exclusion. For example, Black and Latinx 
individuals are charged more for home loans, amounting to an 11 to 
����HKKP[PVUHS�WYVÄ[�MVY�SLUKLYZ��*V\U[Z���������;OPZ�HKKZ�\W�[V������
$500 million annually from Black and Latinx individuals. Despite Foggo 
et al. reporting that lending discrimination being on a “steady decline,” 
the authors did not indicate how that was measured (2020). Most 
importantly, any lending discrimination directly impacts the potential to 
buy a home, one of the main ways a family can build generational wealth 
�+L4H[[LV���������6^UPUN�H�OVTL�TLHUZ�VUL�JHU�YLÄUHUJL�VY�ZLSS�H[�H�
OPNOLY�WYPJL�[OHU�[OL�OVTL�^HZ�W\YJOHZLK�MVY��YLZ\S[PUN�PU�H�WYVÄ[�[OH[�
can be passed down to children or other dependents. 

Achievements such as the Fair Housing Act of 1968—which disallowed 
KPZJYPTPUH[PVU�PU�[OL�I\`PUN��YLU[PUN��VY�ÄUHUJPUN�VM�H�OVTL·HYL�\UHISL�
to protect those they were meant to. Algorithms are often protected 
HZ�[YHKL�ZLJYL[Z�I`�[LJOUVSVN`�JVTWHUPLZ��>P[O�UV�^H`�[V�HUHS`aL�

RESIST, REGULATE, REIMAGINE, AND REINFORCE



COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  |   71   

VY�YLZLHYJO�[OL�MVYT\SHZ�ILPUN�\ZLK��P[�PZ�KPɉJ\S[�[V�I\PSK�H�JHZL�[OH[�
proves someone’s civil rights were violated. 

HIRING

AI is also being used by companies to accelerate the hiring process. 
(SNVYP[OTZ�JHU�NV�[OYV\NO�[OV\ZHUKZ�VM�HWWSPJH[PVUZ�X\PJRS`·JOVVZPUN�
[OL�JHUKPKH[LZ�^OV�ILZ[�Ä[�^OH[�PZ�JVKLK�PU[V�[OL�Z`Z[LT�HZ�PKLHS�
�/LPS^LPS����� ���6UL�VM�[OL�THPU�ILULÄ[Z�PZ�[OH[�TVYL�WLVWSL�JHU�IL�
considered for a position than, for example, when an individual in the 
HR department had to manually review resumes. However, AI is also 
being used for facial recognition to deduce applicants’ personalities 
based on their expressions and appearance (Castelvecchi, 2020). 
6M[LU[PTLZ��[OLZL�WOV[VZ�HYL�VI[HPULK�[OYV\NO�X\PJR�VUSPUL�ZLHYJOLZ�
of a candidate’s social media platforms, such as LinkedIn or Facebook. 
The practice of discerning personality traits from face recognition 
algorithms has been proven generally inaccurate but some companies 
are still deploying this technology (Wells, 2020). 

In addition, facial recognition technology is shown to be less accurate 
on dark skinned faces and women/femmes’ faces (Buolamwini et al., 
�������;OPZ�YLZ\S[Z�PU�PUKP]PK\HSZ�MYVT�[OLZL�OPZ[VYPJHSS`�THYNPUHSPaLK��
intersecting communities often registering as “non-human” to these 
computer systems. This will be discussed more in a later section.  

Gender bias in hiring algorithms was most notably reported in 2018 
^OLU�(THaVU�OHK�[V�NL[�YPK�VM�[OLPY�Z`Z[LT�^OPJO�WLUHSPaLK�JHUKPKH[LZ�
who had “women’s” in their application—that is, attended a women’s 
college or were in a women’s group (Vincent, 2018). According to 
sources at the company, this was because the algorithm was trained on 
L_PZ[PUN�LTWSV`TLU[�PUMVYTH[PVU��<ZPUN�[OL�KH[H�[OH[�TVZ[�VM�(THaVU»Z�
employees are men, the algorithm decided that applications with the 
word “woman” or “women” should be rejected, reinforcing the pre-
existing gender bias at the company. By learning from data based on 
L_PZ[PUN�PULX\P[PLZ��[OL�THJOPUL�PUOLYP[LK�H�IPHZ�HUK�[O\Z�WLYWL[\H[LK�
this unconscious preference in Silicon Valley. 

SARAH E. DILLARD



72  |  COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  

HEALTHCARE
INSURANCE COSTS

Lifestyle data—the food you eat or how much you watch TV—is now 
readily available as industries collect information they hope to use 
to keep you as a customer. In addition, many insurance companies 
are also using this data to determine a patient’s risk of incurring high 
medical costs (Allen, 2019). Concerns are mounting over whether or 
UV[�[OPZ�KH[H�PZ�PTWHJ[PUN�[OL�JVZ[�H�WLYZVU�PZ�X\V[LK�MVY�[OLPY�TVU[OS`�
health insurance rate. In addition, the accuracy of the predictions is 
PU�X\LZ[PVU��HZ�[OL`�YLÅLJ[�KPZJYPTPUH[VY`�HZZ\TW[PVUZ�HIV\[�JLY[HPU�
groups of people.

The Health Insurance Portability and Accountability Act (HIPAA) only 
covers medical information that was collected through a “covered 
entity,” which limits the bill's protective capabilities for health and mental 
health facilities. In recent years, health insurance companies such as 
Aetna and UnitedHealth have been collecting (either independently 
or through contracts) personal or lifestyle data such as social media 
activity, hours spent watching TV, education status, place of residence, 
and net worth (Allen, 2019).

By raising health insurance costs based on certain social demographics, 
LZWLJPHSS`�Z[H[PJ�MHJ[VYZ�Z\JO�HZ�WHYLU[Z»�LK\JH[PVU�SL]LS��THYNPUHSPaLK�
communities become stuck in a cycle of poor health and poverty as the 
assessment is based on metrics they cannot change. In addition, by 
\ZPUN�KH[H�WVPU[Z�[OH[�KPZWYVWVY[PVUH[LS`�PTWHJ[�76*��Z\JO�HZ�HYYLZ[�
records, health insurance companies perpetuate racist oppression. 
Thus, the algorithmic results are inherently biased. 

AT-HOME CARE HOURS

The use of algorithms to make healthcare decisions is becoming more 
widespread as industries try to streamline processes in order to cut time 
and cost while also eliminating human bias. In Arkansas, a software was 
implemented to determine how many hours of at-home-care Medicaid 
WH[PLU[Z�ULLKLK��3LJOLY���������6ɉJPHSZ�ZH`�[OH[�ILMVYL�[OPZ�Z`Z[LT��
their assessments were done by individuals who would make decisions 
that favored some and were arbitrary with others. 

RESIST, REGULATE, REIMAGINE, AND REINFORCE



COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  |   73   

After the algorithm, which was developed by a group of health 
researchers at InterAI,  was implemented, many people had their hours 
J\[·IV[O�WH[PLU[Z�YLJLP]PUN�ZLY]PJLZ�HUK�Z[HɈ�WYV]PKPUN�H[�OVTL�
assistance (Lecher, 2018). Legal Aid of Arkansas started receiving calls 
MYVT�PUKP]PK\HSZ�^P[O�JVTWSHPU[Z��ZVTL�VM�^OVT�^LYL�OVZWP[HSPaLK�K\L�
to lack of care. 

When the president of InterAI was interviewed about transparency in 
the algorithm’s metrics, he argued that one should trust that “a bunch of 
smart people determined this is the smart way to do it” (Lecher, 2018). 
However, during court proceedings it was revealed that the wrong 
calculation was being used for at least one case. This kind of error 
could have been caught if someone had overseen the deployment and 
checked all results. 

POTENTIAL ILLNESSES

A risk-assessment tool used by large health systems in the United 
States was shown to give sick Black patients the same score it was 
NP]PUN�[V�OLHS[OPLY�^OP[L�WLVWSL��6ILYTL`LY�L[�HS������ ���9LZLHYJO�
ZOV^LK�[OH[�Ä_PUN�[OPZ�KPZWHYP[`�^V\SK�OH]L�JH\ZLK�HU�PUJYLHZL�PU�
)SHJR�WH[PLU[Z�^OV�YLX\PYLK�L_[YH�JHYL�MYVT�������[V��������;OPZ�
algorithm did not use race as one of its data points; it did, however, use 
insurance claims data over a certain year (information such as age and 
sex, insurance type, diagnosis, medications, and detailed costs). In the 
end, it predicted accurately what people would spend on healthcare the 
following year; it did not predict who was more in need of improved care 
due to adverse health conditions. 

Proxies for race are often unknowingly used in developing algorithms, 
which then produce biased results. Ruha Benjamin refers to this as 
¸JVKLK�PULX\HSP[`¹�HUK�[OL�LU[PYL�Z`Z[LT�HZ�¸;OL�5L^�1PT�*VKL¹�
(Benjamin, 2019). Without proper knowledge of systemic racism, 
the individuals working for companies such as InterAI continue to 
WLYWL[\H[L�[OL�VWWYLZZPVU�VM�THYNPUHSPaLK�NYV\WZ�^OPSL�NP]PUN�TVYL�
power to the privileged. The notion that healthcare should be provided 
to an individual based on the amount of money they are able to spend 
M\Y[OLYZ�J\YYLU[�YHJPHS�KPZWHYP[PLZ�PU�SPML�L_WLJ[HUJ`�HUK�ILULÄ[Z�[OVZL�
with greater capital. 

SARAH E. DILLARD



74  |  COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  

CRIMINAL JUSTICE
RISK ASSESSMENT

In the 1980s, lawmakers across the United States passed legislation 
for harsh, mandatory minimum sentencing in order to eliminate human 
bias in decision making (Forman, 2017). This meant an individual 
had to spend a certain amount of time in prison based on the crime 
they committed. With the crack-cocaine epidemic ravaging Black 
JVTT\UP[PLZ��Z\IZ[HUJL�\ZL�^HZ�M\Y[OLY�JYPTPUHSPaLK��;VKH �̀�[OL�WYPZVU�
industrial complex (PIC) in the U.S. has in part expanded because of this 
legislation as the number of people incarcerated rose from hundreds 
of thousands to millions over the following decades (The Sentencing 
Project, 2021). The need for improved criminal risk assessment therefore 
became present and private companies started creating algorithms 
in order to more accurately predict the probability of a defendant 
YLVɈLUKPUN��

6UL�VM�[OLZL�[VVSZ��[OL�*VYYLJ[PVUHS�6ɈLUKLY�4HUHNLTLU[�7YVÄSPUN�MVY�
(S[LYUH[P]L�:VS\[PVUZ��*647(:���WYVK\JLZ�[OYLL�JH[LNVYPLZ�VM�YPZR·SV �̂�
medium, or high—and has been shown to reproduce racial disparities 
in its results (Angwin, 2016). Black people are twice as likely as white 
WLVWSL�[V�IL�SHILSLK�H�OPNOLY�YPZR�I\[�UV[�YLVɈLUK��6]LYHSS��[OL�ZVM[^HYL�
^HZ�ZOV^U�[V�IL�HJJ\YH[L�����VM�[OL�[PTL�

The biased results are not the only problem. The labels produced by 
[OL�*647(:�HSNVYP[OT�KV�UV[�JVYYLSH[L�^P[O�H�Z[H[PZ[PJHS�JOHUJL�VM�
YLVɈLUKPUN"�[OL`�HYL�NLULYHSPaH[PVUZ�VY�LZZLU[PHSS`�YHUKVTS`�HZZPNULK�
U\TILYZ��(U`VUL�JHU�PU[LYWYL[�[OL�YH[PUN�KPɈLYLU[S`"�H�OPNO�ZJVYL�VY�
OPNO�JOHUJL�VM�YLVɈLUKPUN�KVLZ�UV[�JVYYLSH[L�[V�H�U\TILY�VM�KH`Z��
months, etc. In addition, these results are shown to judges without any 
explanation of the data that went into them or the formula used. 

In 2016, one defendant challenged a Wisconsin court’s ruling and 
the label produced by the risk-assessment. The judge decided that 
because the algorithm was not deterministic in the ruling, there was no 
^H`�[V�WYV]L�P[�OHK�Z\JO�H�NYH]L�PUÅ\LUJL�VU�[OL�KLJPZPVU��,JROV\ZL�
L[�HS������ ���/V^L]LY��[OL�JV\Y[�MHPSLK�[V�YLJVNUPaL�[OH[�[OPZ�KLJPZPVU�
goes against the purpose of using an algorithm—eliminating human 

RESIST, REGULATE, REIMAGINE, AND REINFORCE



COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  |   75   

bias—by adding the judge’s input on top of the low, medium, or 
high result, and by not using the algorithm in a deterministic way, its 
objectivity (assuming they were objective, which they are not) is not 
being employed. At the end of the day, a judge—a human with bias—
PZ�THRPUN�[OL�KLJPZPVU�HUK�[OH[�KLJPZPVU�PZ�UV^�ILPUN�PUÅ\LUJLK�I`�
inaccurate algorithms. 

In the Wisconsin case, the judge declared that since the defendant 
was able to see the results of the algorithm, there was nothing else 
that needed to be revealed (Eckhouse et al., 2019). However, the data, 
metrics, and formulation all impact the algorithm’s output and can all 
be sources of bias (Miron, 2020). As stated previously, using static 
PUMVYTH[PVU��aPW�JVKL�H[�IPY[O��SHZ[�UHTL��WHZ[�JYPTPUHS�OPZ[VY`��OHZ�ILLU�
shown to correlate with the social factors of sensitive groups more so 
than dynamic information (current substance use, peer rejection, hostile 
behavior). 

FACIAL RECOGNITION

)PVTL[YPJ�PKLU[PÄJH[PVUZ��ÄUNLYWYPU[Z��]VPJL��HUK�PYPZ�ZJHUZ��OH]L�
been used by the criminal justice system for decades (Najibi, 2020). 
In addition, TSA’s advanced imaging technology present at airport 
JOLJRWVPU[Z�YLX\PYLZ�HNLU[Z�[V�ZLSLJ[�VUL�VM�[^V�I\[[VUZ�^OLU�WLVWSL�
enter the machine: man or woman. This means anyone who does not 
Ä[�^P[OPU�[OPZ�VWWYLZZP]L�NLUKLY�IPUHY`�NL[Z�W\SSLK�HZPKL�HUK�ZLHYJOLK�
�*VZ[HUaH�*OVJR����������

6\[�VM�HSS�[OL�HIV]L�IPVTL[YPJ�L_HTWSLZ��MHJPHS�YLJVNUP[PVU�[LJOUVSVN`�
is being deployed across the widest variety of industries, including 
law enforcement, employers, manufacturers, and government housing 
authorities (Klosowski, 2020). In 2018, the Gender Shades study found 
[OH[�[OYLL�KPɈLYLU[�JVTTLYJPHS�HSNVYP[OTZ�^LYL�NYH]LS`�PUHJJ\YH[L�H[�
PKLU[PM`PUN�KHYRLY�ZRPUULK�^VTLU��^P[O�LYYVY�YH[LZ�HZ�OPNO�HZ�������
�)\VSHT^PUP�L[�HS����������*VTWHYLK�[V�H������LYYVY�YH[L�MVY�SPNO[LY�
skinned males, the disparity is astonishing. 

However, the impact of this bias is more frightening. In a test conducted 
I`�[OL�(*3<�VM�(THaVU»Z�MHJPHS�YLJVNUP[PVU�[VVS��^OPJO�^HZ�H]HPSHISL�MVY�
HU`VUL�[V�\ZL��[OL�[VVS�PUJVYYLJ[S`�PKLU[PÄLK����TLTILYZ�VM�*VUNYLZZ�

SARAH E. DILLARD



76  |  COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  

HZ�JYPTPUHSZ��:UV �̂��������)SHJR�*VUNYLZZ�TLTILYZ�THKL�\W�����
VM�[OVZL�TH[JOLZ�KLZWP[L�VUS`�THRPUN�\W�����VM�[OL�/V\ZL��;OLZL�
results reinforce the historic over-policing of the Black community and 
JYPTPUHSPaH[PVU�VM�PUKP]PK\HSZ�IHZLK�VU�[OLPY�ZRPU�[VUL�

/H]PUN�TVYL�HJJ\YH[L�MHJPHS�YLJVNUP[PVU�[LJOUVSVN`�^V\SK�UV[�Ä_�
the problem of over-policing; in fact, it might exacerbate it. During 
ZSH]LY`�PU�[OL�<:��¸SHU[LYU�SH^Z¹�^LYL�LUHJ[LK�PU�5L^�@VYR�YLX\PYPUN�
enslaved people to carry a light by their faces in order to remain visible 
(Najibi, 2020). This same tracking of Black individuals could thus be 
done by high resolution cameras disproportionately located in certain 
neighborhoods which capture images and use them for databases.  

DIGITAL INCLUSION: WHAT CAN SOCIAL WORKERS DO?
Even as technology expands and overtakes many human jobs, social 
workers are here to stay. According to a 2015 study done by NPR, 
mental health workers are the least likely profession to be automated by 
H�THJOPUL��)\P���������;OPZ�TLHUZ�[OH[�MVY�HZ�SVUN�HZ�(0�HɈLJ[Z�V\Y�SP]LZ��
there will be social workers ready and able to advocate. 

According to the NASW Code of Ethics, social workers must challenge 
social injustice and address social problems (NASW, 2021). With 
technology companies often unknowingly perpetuating systemic 
VWWYLZZPVU�VM�THYNPUHSPaLK�NYV\WZ�[OYV\NO�V]LY�WVSPJPUN��PUHKLX\H[L�
healthcare, or discrimination, social workers have the responsibility to 
advocate for those targeted by these practices. The following outlines 
current models addressing algorithmic harm and ways social workers 
can be involved in mitigating the gap of algorithmic knowledge, digital 
PULX\HSP[ �̀�HUK�JVKLK�IPHZ�

RESIST

;OLYL�HYL�THU`�VYNHUPaH[PVUZ�^VYRPUN�[V�IHU�[OL�\ZL�VM�MHJPHS�
YLJVNUP[PVU�ZVM[^HYL�I`�WVSPJL��6aLY�L[�HS����������;OL�^LIZP[L�
IHUMHJPHSYLJVNUP[PVU�JVT�PZ�Z\WWVY[LK�I`�KVaLUZ�VM�NYV\WZ�HUK�[OL`�
provide an interactive map marking places where facial recognition 
is used (Ban Facial Recognition, n.d.). This not only includes law 

RESIST, REGULATE, REIMAGINE, AND REINFORCE



COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  |   77   

LUMVYJLTLU[�HNLUJPLZ��I\[�(THaVU�9PUN�KL]PJLZ�HZ�^LSS��

It is nearly impossible today to avoid an online footprint. However, 
resisting the use of AI in one’s everyday life is one of the main forms of 
not only advocacy but protection. Social workers can both inform their 
clients and resist these technologies in their own lives. Guidelines to 
follow include limiting the amount of information shared online, refusing 
[V�VW[�PU�[V�TVUP[VYPUN�ZLY]PJLZ��HUK�[\YUPUN�VɈ�ZTHY[WOVUL�MLH[\YLZ�
[OH[�NYV\W�WOV[VZ�IHZLK�VU�PKLU[PÄLK�MHJLZ��2SVZV^ZRP���������

6M[LU[PTLZ�^OLU�ZPNUPUN�\W�MVY�HU�VUSPUL�HJJV\U[��^LIZP[LZ�^PSS�HZR�
for personal identifying information (PII) such as full name, birthdate, 
and address. Unless absolutely needed, providing these sensitive 
facts about oneself can result in unwanted tracking and associations. 
Analytics such as cookies are another way websites use online history 
[V�ÄS[LY�HKZ�HUK�ZLHYJO�YLZ\S[Z��;OL`�ZH]L�JLY[HPU�[`WLZ�VM�KH[H�PU�VYKLY�
to track what individuals are clicking on, looking at, and engaging 
with. The social isolation this causes limits online content and can be 
KHUNLYV\Z�MVY�JSPLU[Z�^OV�ÄUK�[OLTZLS]LZ�SVJRLK�PU[V�TPZPUMVYTH[PVU��
Meanwhile, under the guise of social connection, facial recognition 
ZVM[^HYL·ZWLJPÄJHSS`�PU�P7OVULZ·HSSV^Z�\ZLYZ�[V�[HN�[OLPY�MYPLUKZ��
However, this data is being shared beyond one’s personal device. The 
ZL[[PUN�T\Z[�IL�[\YULK�VɈ�THU\HSS �̀�

Resistance can come in many forms. Creative ways of avoiding the 
[LJOUVSVN`�HYL�WYL]HSLU[�LZWLJPHSS`�PU�[OL�WHZ[�Ä]L�`LHYZ��TVZ[�UV[HIS`�
the Umbrella Movement in Hong Kong, in which protestors used open 
umbrellas to shield their faces from government surveillance cameras 
(BBC, 2019). 

REGULATE

Currently, there are no federal laws in the US regulating AI. Governing 
bodies lack the expertise and knowledge to properly create legislation 
that protects privacy, limits surveillance, and bans discrimination 
�7HaaHULZL���������;OLZL�[LJOUVSVNPLZ��HZ�V\[SPULK�HIV]L��YLÅLJ[�[OL�
structural biases that have been present in society for centuries, and 
[O\Z�JVU[PU\L�[V�OHYT�THYNPUHSPaLK�JVTT\UP[PLZ��7YP]HJ`�SLNPZSH[PVU�

SARAH E. DILLARD



78  |  COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  

from the 1960s-80s are now out of date. Data protection only covers 
government and medical databases while anti-discrimination in housing 
and employment does not extend to a computer formula (Bock, 
�������;OLZL�WVSPJPLZ�ULLK�[V�IL�YLMYLZOLK�[V�YLÅLJ[�[OL�]HZ[�NYV^PUN�
implementation of AI. 

Technology companies monitor their systems in-house and rarely 
WYV]PKL�[OL�L_HJ[�KL[HPSZ�VM�[OLPY�HSNVYP[OTZ�MVY�X\HSP[`�JOLJRZ�I`�V\[ZPKL�
researchers. They claim their system is protected by being a trade 
secret: intellectual property that cannot be released because it is integral 
[V�[OL�ÄUHUJPHS�^LSS�ILPUN�VM�[OL�JVTWHU`�HUK�JV\SK�W\[�[OLT�V\[�VM�
I\ZPULZZ�PM�JVWPLK��<UP[LK�:[H[LZ�7H[LU[�HUK�;YHKLTHYR�6ɉJL��U�K����
However, this claim prevents diverse and informed research entities from 
TP[PNH[PUN�IPHZLK�V\[W\[Z�VY�YLZ\S[Z�^OPJO�YLÅLJ[�OPZ[VYPJ�KPZJYPTPUH[PVU��
+\L�[V�H�MLHY�VM�SVZPUN�WYVÄ[Z�PM�[OL�JVTWHU`»Z�YLW\[H[PVU�PZ�OHYTLK��
many data-driven industries hide behind this trade secret policy, which 
intentionally obscures them from public review. 

Social workers in policy can educate themselves on the uses of AI in a 
ÄLSK�[OL`�HYL�L_WLY[Z�PU��OLHS[OJHYL��JYPTPUHS�Q\Z[PJL��VY�HUV[OLY��;OL`�
can write briefs on biased algorithms and the need for federal regulation 
as members of SAFElab at Columbia University did (Anguiano et al., 
2021). Cities such as San Francisco and Boston have passed their own 
legislation disallowing facial recognition technology, ahead of federal 
changes (Associated Press, 2021). 

Petitioning lawmakers to focus on AI and its potential for harm is 
another way social workers can get involved in advocating for digital 
inclusion. As stated before, with biometric systems such as facial 
recognition spreading surveillance, it is likely that a more accurate 
algorithm will be used to continue the over-policing of Black individuals. 
Social workers, who are educated in the historic and systemic harms 
KVUL�[V�THYNPUHSPaLK�JVTT\UP[PLZ��JHU�PUMVYT�[OVZL�^P[O�WVSP[PJHS�WV^LY�
the ways in which AI perpetuates this oppression.

>P[OV\[�YLN\SH[PVU��[LJOUVSVN`�JVTWHUPLZ�^PSS�IL�\USPRLS`�[V�ZJY\[PUPaL�
their systems to the same degree as outside researchers. Limiting the 
uses of a product, whether by disallowing hate groups from posting on a 

RESIST, REGULATE, REIMAGINE, AND REINFORCE



COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  |   79   

WSH[MVYT�VY�I`�LUKPUN�KH[H�WHY[ULYZOPWZ�^P[O�V[OLY�ÄYTZ��TLHUZ�SPTP[PUN�
I\ZPULZZ�HUK�[OLYLMVYL�WYVÄ[��;OLYL�ULLKZ�[V�IL�H�TVUL[HY`�PUJLU[P]L�
in the form of a tax (ideally on data storage) that encourages these 
companies to delete digital footprints. 

REIMAGINE 

There are many other roles that social workers can take in advocating 
for digital inclusion. Technology companies are now creating jobs 
PU�ÄLSKZ�Z\JO�HZ�YLZLHYJO�L[OPJZ�HUK�JVTT\UP[`�YLSH[PVUZ�HUK�HYL�
attempting to diversify their hiring practices through apprenticeships 
for people with unconventional backgrounds. With an extensive 
\UKLYZ[HUKPUN�VM�Z`Z[LTPJ�IPHZ��ZVJPHS�^VYRLYZ�HYL�^LSS�LX\PWWLK�[V�IL�
a part of these discussions.

,[OPJHS�KL]LSVWTLU[�HUK�KLWSV`TLU[�VM�(0�PZ�VUL�LTLYNPUN�ÄLSK�ZVJPHS�
workers must be a part of. Knowledge of criminal justice and healthcare 
is integral in decisions concerning what data should be used, whether 
that data is a proxy for race, and if the data results in biased outputs 
[OH[�OHYT�THYNPUHSPaLK�JVTT\UP[PLZ��(WWS`PUN�[OPZ�Q\KNTLU[�HUK�
empathy will be a growing necessity as automation continues to expand 
(Johnson, 2021). 

In research, teams improving machine learning algorithms need 
annotators from a wide range of backgrounds in order to capture 
the nuances of human expression (Johnson, 2021). By including 
stakeholders with varying sources of knowledge, discussions open 
up and opinions are provided which could not have been captured by 
people who mostly think the same. Time and diligence are also needed, 
something tech companies try to cut by paying annotators by the social 
media post. Working with a group means a consensus must be reached, 
rather than allowing one person to determine the meaning behind a post 
(Patton et al., 2020). 

(Z�[LJOUVSVN`�JVTWHUPLZ�ZLLR�[V�KP]LYZPM`�[OLPY�Z[HɈ�PU�VYKLY�[V�PTWYV]L�
the systems they create, social workers can be consultants for unbiased 
hiring practices. Firms such as Race Forward are employing people to 
SVVR�H[�Z[Y\J[\YHS�VWWYLZZPVU�HUK�ÄUK�^H`Z�[V�LSPTPUH[L�P[�PU�KPɈLYLU[�

SARAH E. DILLARD



80  |  COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  

HYLHZ��9HJL�-VY^HYK��U�K����)PN�JVUZ\S[PUN�ÄYTZ�HYL�HSZV�[HRPUN�VU�
projects to create more inclusive employment searches and outreach, 
HUK�ZVJPHSS`�YLZWVUZPISL�[LJOUVSVN`�PZ�HU�LTLYNPUN�ÄLSK�VM�YLZLHYJO���

REINFORCE

(SNVYP[OTPJ�RUV^SLKNL�NHWZ�HYL�HUV[OLY�MVYT�VM�KPNP[HS�PULX\HSP[`�
PTWHJ[PUN�THYNPUHSPaLK�JVTT\UP[PLZ��*V[[LY�L[�HS����������<UKLYZ[HUKPUN�
how personal data is used, where one may encounter bias due to AI, 
and ways to protect oneself are all crucial for agency in the digital world. 

Socioeconomic status is viewed as the main determinant for algorithmic 
knowledge (Cotter et al., 2020). In the US, class often relates to one’s 
race, as a disproportionate number of Black and Latinx individuals 
SP]L�ILSV^�[OL�WV]LY[`�SPUL��*YLHTLY���������/V^L]LY������VM�)SHJR�
people in the US use social media (Pew, 2021). This means a vast 
majority of Black users—given the disproportionate number of Black 
individuals who experience intersecting poverty—likely are not aware of 
the underlying algorithms, data scraping, or implications of their online 
presence in their physical lives. 

9LPUMVYJPUN�IHZL�RUV^SLKNL�VM�[LJOUVSVN`·ZWLJPÄJHSS`�(0�HUK�OV^�
it is used—is another way social workers can support digital inclusion 
LɈVY[Z��;OL�(SNVYP[OTPJ�1\Z[PJL�3LHN\L��MVY�L_HTWSL��[VVR�H�JYLH[P]L�
approach by creating a workshop called “Drag vs. AI” (AJL, 2020). 
Participants learn about facial recognition software and then learn 
from drag performers how to do their makeup in order to escape the 
THJOPUL»Z�¸JVKLK�NHaL�¹�0[�LUKZ�^P[O�H�ÄUHS�Y\U^H`�ZOV^�HUK�HKKP[PVUHS�
information on how to resist, not only individually but as part of an 
V]LYZPNO[�VYNHUPaH[PVU��

CONCLUSION

It is necessary for social workers to become advocates for digital 
inclusion. Technology is only progressing and becoming a greater part 
VM�V\Y�L]LY`KH`�SP]LZ��*\YYLU[S �̀�[OL�(0�Z`Z[LTZ�ILPUN�KL]LSVWLK�YLÅLJ[�
[OL�OPZ[VYPJ�KPZJYPTPUH[PVU�VM�THYNPUHSPaLK�PUKP]PK\HSZ�IHZLK�VU�ZLUZP[P]L�
characteristics such as race, class, and gender. Well-versed in systemic 

RESIST, REGULATE, REIMAGINE, AND REINFORCE



COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  |   81   

oppression—its roots, causes, and manifestations—social workers 
T\Z[�IL�PU]VS]LK�PU�KPZTHU[SPUN�[OPZ�SH[LZ[�P[LYH[PVU!�JVKLK�PULX\HSP[`�
(Benjamin, 2019). Through resistance, regulation, reimagination, and 
reinforcement social workers in any position are able to advocate for 
those being harmed by an algorithm. 

REFERENCES

Algorithmic Justice League. (2020). Drag vs. AI. https://www.ajl.org/drag-vs-ai

Allen, M. (2019). Health insurers are vacuuming up details about you—and it could 
raise your rates. ProPublica. https://www.propublica.org/article/health-insurers-are-
vacuuming-up-details-about-you-and-it-could-raise-your-rates

Anguiano, K., Darkwa, E., & Patton, D. (2021). Recommendations to end 21st century 
online “stop and frisk” policing. SAFELab. https://safelab.socialwork.columbia.edu/
content/policy-0

Angwin, J. (2016). When algorithms decide what you pay. ProPublica. https://www.
propublica.org/article/breaking-the-black-box-when-algorithms-decide-what-you-pay

Anyoha, R. (2017, August 28). ;OL�OPZ[VY`�VM�HY[PÄJPHS�PU[LSSPNLUJL� Science in the News.
O[[WZ!��ZP[U�OTZ�OHY]HYK�LK\�ÅHZO������OPZ[VY`�HY[PÄJPHS�PU[LSSPNLUJL�

Associated Press. (2021). States push back against use of facial recognition by police. U.S. 
News. https://www.usnews.com/news/politics/articles/2021-05-05/states-push-back-
against-use-of-facial-recognition-by-police

Ausiello, G., & Petreschi, R. (2013). The power of algorithms: Inspiration and examples in 
everyday life��:WYPUNLY��O[[WZ!��KVP�VYN�������� ���������� �����

))*������ ��:LW[LTILY������/VUN�2VUN�WYV[LZ[Z!�>OH[�PZ�[OL�º\TIYLSSH�TV]LTLU[»&�BBC.
O[[WZ!��^^ �̂IIJ�JV�\R�UL^ZYV\UK�� ������

Bean, R. (2017). How big data is empowering AI and machine learning at scale. MIT Sloan 
Management Review. https://sloanreview.mit.edu/article/how-big-data-is-empowering-
ai-and-machine-learning-at-scale/

Benjamin, R. (2019). Assessing risk, automating racism. Science����������������������

Bock, C. (2016). Preserve personal freedom in networked societies. Nature, 537(9). 

)\P��8���������4H`������>PSS�`V\Y�QVI�IL�KVUL�I`�H�THJOPUL&�NPR. https://www.npr.org/
ZLJ[PVUZ�TVUL`����������������������^PSS�`V\Y�QVI�IL�KVUL�I`�H�THJOPUL

SARAH E. DILLARD



82  |  COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  

Buolamwini, J., & Gebru, T. (2018). Gender shades: Intersectional accuracy disparities in 
JVTTLYJPHS�NLUKLY�JSHZZPÄJH[PVU��Proceedings of Machine Learning Research, 81, 1-15. 

Campisi, N. (2021, June 18). The black homeownership gap is worse. Here’s what’s being 
done. Forbes. https://www.forbes.com/advisor/mortgages/black-homeownership-gap/

Campisi, N. (2021, February 26). From inherent racial bias to incorrect data—the problems 
with current credit scoring models. Forbes. https://www.forbes.com/advisor/credit-
cards/from-inherent-racial-bias-to-incorrect-data-the-problems-with-current-credit-
scoring-models/

*VZ[HUaH�*OVJR��:����������Design justice: Community-led practices to build the worlds 
we need. MIT Press. 

Cotter, K., & Reisdorf, B. (2020). Algorithmic knowledge gaps: A new dimension of (digital)
PULX\HSP[ �̀�International Journal of Communications��������������

Council, J. (2021). Facial recognition tool used by police faces civil lawsuit in California. 
Wall Street Journal. https://www.wsj.com/articles/facial-recognition-tool-used-by-
police-faces-civil-lawsuit-in-california-11615395179

Counts, L. (2018). Minority home buyers face widespread statistical lending discrimination, 
Z[\K`�ÄUKZ��Berkeley Haas School of Business. 

*YLHTLY��1����������0ULX\HSP[PLZ�WLYZPZ[�KLZWP[L�KLJSPUL�PU�WV]LY[`�MVY�HSS�THQVY�YHJL�HUK�
Hispanic origin groups. U.S. Census. https://www.census.gov/library/stories/2020/09/
poverty-rates-for-blacks-and-hispanics-reached-historic-lows-in-2019.html

DeMatteo, M. (2020). How this entrepreneur is working to help Black women build 
generational wealth through homeownership. CNBC. https://www.cnbc.com/select/
what-is-generational-wealth/

,JROV\ZL��3���3\T��2���*VU[P�*VVR��*���
�*PJJVSPUP��1������ ���3H`LYZ�VM�IPHZ!�(�\UPÄLK�
approach for understanding problems with risk assessment. Criminal Justice and 
Behavior,�������������� �

Foggo, V., & Villasenor, J. (2020). Algorithms, housing discrimination, and the new 
disparate impact rule. Science and Technology Law Review, 22(1), 1-62.

Forman, J. (2017). Locking up our own: Crime and punishment in Black America. Farrar, 
Straus, and Giroux. 

Hao, K. (2020). The coming war on the hidden algorithms that trap people in poverty. 
MIT Technology Review��O[[WZ!��^^ �̂[LJOUVSVN`YL]PL �̂JVT��������������������
HSNVYP[OTZ�JYLH[L�H�WV]LY[`�[YHW�SH^`LYZ�ÄNO[�IHJR�

RESIST, REGULATE, REIMAGINE, AND REINFORCE



COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  |   83   

Heilweil, R. (2019, December 12). (Y[PÄJPHS�PU[LSSPNLUJL�^PSS�OLSW�KL[LYTPUL�PM�`V\�NL[�`V\Y�
next job.�=V_�]V_�JVT�YLJVKL���� ���������  �����HY[PÄJPHS�PU[LSSPNLUJL�HP�QVI�ZJYLLU

Hurley, M., & Adebayo, J. (2016). Credit scoring in the era of big data. Yale Journal of Law 
and Technology 18������

1VOUZVU��2���������(\N\Z[�����This new way to train AI could curb online harassment. 
Wired. https://www.wired.com/story/new-way-train-ai-curb-online-harassment/

Klowsowski, T. (2020). Facial recognition is everywhere. Here’s what we can do about it. 
The New York Times. https://www.nytimes.com/wirecutter/blog/how-facial-recognition-
works/ 

Lecher, C. (2018, March 21). A healthcare algorithm started cutting care, and no one knew 
why.�;OL�=LYNL��O[[WZ!��^^ �̂[OL]LYNL�JVT��������������������OLHS[OJHYL�TLKPJHPK�
algorithm-arkansas-cerebral-palsy

3LYULY��4���������1\S`������6UL�OVTL��H�SPML[PTL�VM�PTWHJ[��The Washington Post. https://
www.washingtonpost.com/business/2020/07/23/black-homeownership-gap/

4PYVU��4���;VSHU��:���.VTLa��,���
�*HZ[PSSV��*����������,]HS\H[PUN�JH\ZLZ�VM�HSNVYP[OTPJ�IPHZ�
in juvenile criminal recidivism. (Y[PÄJPHS�0U[LSSPNLUJL�HUK�3H^��� ��������������

Najibi, A. (2020). Racial discrimination in face recognition technology. Science in the 
News��O[[WZ!��ZP[U�OTZ�OHY]HYK�LK\�ÅHZO������YHJPHS�KPZJYPTPUH[PVU�PU�MHJL�YLJVNUP[PVU�
technology/

NASW. (n.d.) Code of ethics. National Association of Social Workers. https://www.
socialworkers.org/About/Ethics/Code-of-Ethics/Code-of-Ethics-English

6ILYTL`LY��A���7V^LYZ��)���=VNLSS��*���
�4\SSHPUH[[OHU��:������ ���+PZZLJ[PUN�YHJPHS�IPHZ�PU�
an algorithm used to manage the health of populations. Science 366�����������������

6_MVYK�(K]HUJLK�3LHYULY»Z�+PJ[PVUHY �̀��U�K���+LÄUP[PVU�VM�HSNVYP[OT�UV\U��Oxford University 
Press��O[[WZ!��^^ �̂V_MVYKSLHYULYZKPJ[PVUHYPLZ�JVT�\Z�KLÄUP[PVU�LUNSPZO�HSNVYP[OT

6aLY��5���9\HUL��2���
�*HNSL��4����������.YHZZYVV[Z�HJ[P]PZ[Z�HYL�SLHKPUN�[OL�ÄNO[�[V�Z[VW�
face recognition. It’s time for Congress to step up, too. American Civil Liberties Union.
https://www.aclu.org/news/privacy-technology/grassroots-activists-are-leading-the-
ÄNO[�[V�Z[VW�MHJL�YLJVNUP[PVU�P[Z�[PTL�MVY�JVUNYLZZ�[V�Z[LW�\W�[VV�

Patton, D., Frey, W. R., McGregor, K. A., Lee, F. T., Mckeown, K., & Moss, E. (2020).
Contextual analysis of social media: The promise and challenge of eliciting context 
in social media posts with natural language processing. Association for Computing 
Machinery�����������

SARAH E. DILLARD



84  |  COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  

7HaaHULZL��*����������,[OPJHS�JVUJLYUZ�TV\U[�HZ�(0�[HRLZ�IPNNLY�KLJPZPVU�THRPUN�YVSL�
Havard Gazette��O[[WZ!��UL^Z�OHY]HYK�LK\�NHaL[[L�Z[VY`���������L[OPJHS�JVUJLYUZ�
mount-as-ai-takes-bigger-decision-making-role/

Pew. (2021). Social media fact sheet. Pew Research Center. https://www.pewresearch.org/
internet/fact-sheet/social-media/

Singletary, M. (2020). Credit scores are supposed to be race-neutral. But that’s impossible.
Washington Post. https://www.washingtonpost.com/business/2020/10/16/how-race-
HɈLJ[Z�`V\Y�JYLKP[�ZJVYL�

:UV �̂�1����������(THaVU»Z�MHJL�YLJVNUP[PVU�MHSZLS`�TH[JOLK����TLTILYZ�VM�*VUNYLZZ�
with mugshots. American Civil Liberties Union. https://www.aclu.org/blog/privacy 
[LJOUVSVN`�Z\Y]LPSSHUJL�[LJOUVSVNPLZ�HTHaVUZ�MHJL�YLJVNUP[PVU�MHSZLS`�TH[JOLK���

Trainor, S. (2015, July 22). The long, twisted history of your credit score. Time. https://time.
com/3961676/history-credit-scores/

=PUJLU[��1���������6J[VILY������Amazon reportedly scraps AI recruitment tool that was 
biased against women.�;OL�=LYNL��O[[WZ!��^^ �̂[OL]LYNL�JVT�������������� ������HP�
YLJY\P[PUN�[VVS�IPHZ�HTHaVU�YLWVY[

Wells, S. (2020, May 22). A.I. can now correctly predict something we think about privately. 
Inverse. https://www.inverse.com/innovation/ai-photo-judgment

RESIST, REGULATE, REIMAGINE, AND REINFORCE



COLUMBIA SOCIAL WORK REVIEW, VOL. XIX  |   85   

SARAH E. DILLARD