COLUMBIA JOURNAL OF RACE AND LAW FORUM VOL. 11 JULY 2021 NO. 4 COMMENT CALCULATING THE SOULS OF BLACK FOLK: PREDICTIVE ANALYTICS IN THE NEW YORK CITY ADMINISTRATION FOR CHILDREN’S SERVICES J. Khadijah Abdurahman* In 1995, there were nearly 50,000 children removed from their families into the New York City Administration for Children’s Services’ (ACS) foster care system.1 The NYC ACS’ forcible transfer of children from a protected group into another group may amount to genocide under Article 2(e) of the Genocide Convention if formal review can demonstrate an “intent to destroy” the group “as such” or at least “in part.” Rather than pursuing a citizen’s tribunal, or truth and reconciliation committee to assess the historic transfer of Black children to other groups during this period by the child welfare system, ACS has focused on collecting data from currently targeted populations in order to “predict who needs * J. Khadijah Abdurahman, Abolitionist, Tech Impact Fellow, AI Now Institute, NYU; Assembling Voices Fellow, The American Assembly, Columbia University; Co-Founder of The Otherwise School, Human Computer Design and Engineering, University of Washington. It would not be possible to do this work without the village providing child care for me during these twin pandemics of COVID-19 and state violence. Thank you to Fatoumata Daou, my youngest son’s home attendant, Luz Leon, who provided community habilitation for my oldest son and my mother, Mary Schoen who continues to be the most supportive grandmother I could ask for. Thank you to Ayantu Tibeso, Rashida Richardson, Meredith Whittaker, Kurt Mundorff and many others who helped me transition a stream of consciousness into a coherent narrative. Finally, I’d like to acknowledge Joyce McMillan, Movement for Family Power, Stop LAPD Spying and Mijente for the radical grassroots campaigns that make abolition a tangible possibility. There are too many people and movements to fit in this space, but it’s the baby mothers refusing to open the door without a warrant and the power of the people that continues to inform my thinking and imagination. 1 Press Release, N.Y.C. Administration for Children’s Services, Fewer Than 8,000 Children in NYC’s Foster Care System—A Historic Low (Feb. 14, 2020), https://www1.nyc.gov/assets/acs/pdf/PressReleases/2020/FosterCare CensusRelease.pdf [https://perma.cc/ZUW9-97SJ]. 76 COLUM. J. RACE & L. F. [Vol. 11:75 prevention” services. This paper examines the Family First Prevention Act’s legislative mandate to calculate the “souls of Black folks” and the geographies of predictive analytics developed to serve this aim. Using an abolitionist lens grounded in the epistemology offered by W. E. B. Du Bois’ Souls of Black Folks, this argument moves beyond the Fairness, Accountability and Transparency (FAT) framework to propose strategies for dismantling the “new modes of surveillance and social control” manifested in NYC ACS’ preventive turn. I propose a Get Out mathematics drawing from Katherine McKittrick’s proposal to “count it out different” as the fugitive’s alternative to state sanctioned datafication. 2021] CALCULATING THE SOULS OF BLACK FOLK 77 I. Introduction ............................................................................... 78 A. Widening the Digital Net................................................ 78 B. Predicting Prevention (and its Discontents) ................. 81 C. Get Out Mathematics ..................................................... 84 II. Intent to Destroy ..................................................................... 87 III. Creating “Prevented” Populations ........................................ 90 A. The Federal Family First Prevention Act ..................... 92 B. What Are Prevention Services? ...................................... 94 C. When Predictive Analytics Became the Remedy for an Abusive System ........................................................ 97 IV. ACS’ Considerations in Building a Predictive Risk Model .................................................................................. 102 V. Geographies of Prediction ..................................................... 104 A. What is SafeMeasures? ................................................. 106 VI. Conclusion ............................................................................. 108 78 COLUM. J. RACE & L. F. [Vol. 11:75 I. INTRODUCTION A. Widening the Digital Net Figure 1. New York State, Office of Children and Family Services, Racial Categorization2 In July 2016, DJ Patil, the Chief Data Scientist for the Obama White House Office of Science and Technology Policy, asked Gladys Carrión, the 2013–2016 Commissioner of New York City’s vast child welfare administration, what she thought of using predictive analytics in her agency.3 “It scares the hell out of me,” Carrión said, inciting half nervous chuckles in the room.4 Carrión continued, “I think about how we are impacting and infringing on people’s civil liberties.”5 She added that she runs a system “that exclusively serves black and brown children and families . . . I am concerned about widening the net under the guise that we are going to help them. How can we use these tools to keep children and families in communities together?”6 Five 2 N.Y. OFF. CHILD. & FAM. SERV., 2020 MONITORING AND ANALYSIS PROFILES WITH SELECTED TREND DATA: 2016–2020, 7 (2020), https://ocfs.ny .gov/main/reports/maps/countiesz/New%20York%20State.pdf [https://perma .cc/MF3S-C233] 3 Megan Smith, The White House Names Dr. DJ Patil as the First U.S. Chief Data Scientist, THE WHITE HOUSE (Feb. 18, 2015, 4:48 PM), https://obamawhitehouse.archives.gov/blog/2015/02/18/white-house-names-dr- dj-patil-first-us-chief-data-scientist [https://perma.cc/7UQ2-RW6E]. 4 Devon Ziminski, A ‘Mind Blowing’ Experience at the White House Foster Care Hackathon, IMPRINT (May 31, 2016, 1:46 PM), https://imprintnews.org/featured/white-house-foster-care-hackathon/18511 [https://perma.cc/BFZ8-XQBB]. 5 Id. 6 Id. 2021] CALCULATING THE SOULS OF BLACK FOLK 79 months later, in the wake of the highly publicized murder of Zymere Perkins by his parents, and despite multiple contacts with the New York City Administration of Children’s Services (ACS), Carrión resigned.7 A public records request,8 submitted to ACS in December 2019 for all documentation associated with the procurement, training, use policy, and funding associated with their implementation of predictive analytics, highlights the degree to which the agency resists its own inspection. Consistent with previous reporting in ProPublica,9 the agency dragged its feet in response to the FOIA request, requested multiple extensions, and a full nine months later released three heavily redacted documents related to its contracts with New York University (NYU) and the City University of New York (CUNY). The grounds on which ACS specifically refused to provide the predictive analytics algorithm was New York State Public Officers Law § 87(2)(f), which provides that the requested records are exempt from disclosure because the disclosure of the records “could endanger the life or safety of any person.”10 Carrión’s concern about a widening digital net ensnaring Black and Brown11 families under the guise of helping them appears 7 Rich Calder et al., Zymere Perkins’ Sickening Death Exposes NYC’s Unforgivable Child Welfare Failures, N.Y. POST (Dec. 14, 2019), https://nypost.com/2019/12/14/zymere-perkins-sickening-death-exposes-nycs- unforgivable-child-welfare-failures [https://perma.cc/U9EW-GYQA]. 8 Beryl Lipton, Predictive Analytics at NYC Administration for Children’s Services, MUCKROCK (Dec. 1, 2019), https://www.muckrock.com/foi /new-york-city-17/predictive-analytics-at-nyc-administration-for-childrens- services-83900/ [https://perma.cc/4ZDW-G7WC]. 9 Joaquin Sapien, Foiled by FOIL: How One City Agency Has Dragged Out a Request for Public Records for Nearly a Year, PROPUBLICA (Apr. 21, 2016), https://www.propublica.org/article/how-city-agency-dragged-out-request-for- public-records-for-nearly-a-year [https://perma.cc/9Z2J-TDWK]. 10 See, e.g., Asian Am. Legal Def. & Educ. Fund v. NY City Police Dep’t, 5 N.Y.S.3d 13 (App. Div. 2015); Bellamy v. NY City Police Dep’t, 930 N.Y.S.2d 178 (App. Div. 2011). 11 N.Y. OFF. CHILD. & FAM. SERV., supra note 2, at 7. New York State classifies race and ethnicity in “into mutually exclusive categories. In the process, children identified as White/Hispanic or Black/Hispanic are counted under Latino and not in White or African American.” Id. The non-specific racial category “Brown” is being invoked by the author as we do not have truly representative data in New York State by race but the condition of mutually exclusive race/ethnicity categories suggests Black children are under counted. The table listed in Appendix A suggests those identified as Latino are primarily Puerto Rican ethnicity but the racial composition is unclear. 80 COLUM. J. RACE & L. F. [Vol. 11:75 warranted, based on a review of the public record in search of the answers that the child welfare agency refused to provide. Examining the series of changes enacted by ACS following Gladys Carrión’s resignation provides insight into ACS’ implementation of, and vision for, predictive analytics. In February of 2017, New York City Mayor Bill DeBlasio recruited David Hansell from KPMG12 to be the new Commissioner of ACS. The appointment came in the wake of a series of high profile child deaths,13 multiple reports14 from the Department of Investigation (DOI) criticizing the agency’s systemic failures and violations of the law, and a class action lawsuit filed on behalf of all children in New York City foster care claiming, “ACS and OCFS [New York State Office of Children and Family Services]are causing irreparable harm to children in custody by failing to protect children from maltreatment, failing to ensure services provided are effective and of acceptable quality, and failing to ensure appropriate placements.”15 It’s in this context that Hansell declared that he would conduct his own “top-to- bottom review of ACS’s protective and preventive functions to strengthen what’s working and to change what isn’t.”16 This review does not seek to reckon with the agency’s role in shattering bonds and regulating17 families. Rather, it sought to 12 Hansell was the former Head of the Global Human & Social Services Center of Excellence at KPMG. There, he led a team working with ACS since 2012 to consolidate their administrative data. They had also worked with the agency to develop a new RFP process so it’s unsurprising they were immediate beneficiaries of the system they helped to create once Hansell took office. 13 Jeff Mays, 10 Kids Known to ACS Died in Three Months This Summer, Stringer Audit Says, DNA INFO (Dec. 22, 2016, 3:15 PM), https://www.dnainfo.com/new-york/20161222/hamilton-heights/stringer-audit- acs-zymere-perkins-jaden-jordan-de-blasio [https://perma.cc/F29H-HKRC]. 14 MARK G. PETERS, DOI INVESTIGATION OF CHILD FATALITIES AND OTHER INJURIES REVEALS LEGAL AND PRACTICE VIOLATIONS BY THE CITY ADMINISTRATION FOR CHILDREN’S SERVICES (2016), https://www1.nyc.gov/assets /doi/reports/pdf/2016/2016-05-03-14-ACS-Report-FINAL.pdf [https://perma.cc/TH8K-G4RD]. 15 LETITIA JAMES, A BETTER CHILDHOOD, FACT SHEET: ELISA W. V. CITY OF NEW YORK (2015), https://www.clearinghouse.net/chDocs/public/CW- NY-0009-0002.pdf [https://perma.cc/7QWA-J8SN]. 16 Nikita Stewart & Jacey Fortin, De Blasio Picks New Commissioner for Troubled Child Welfare Agency, N.Y. TIMES (Feb. 21, 2017), https://www.nytimes.com/2017/02/21/nyregion/david-hansell-new-york- administration-for-childrens-services.html [https://perma.cc/WC5X-HSGJ]. 17 Dorothy Roberts, Abolishing Policing Also Means Abolishing Family Regulation, IMPRINT (June 16, 2020, 5:26 AM), https://imprintnews.org/child- 2021] CALCULATING THE SOULS OF BLACK FOLK 81 quantify and manage the Black and Brown communities it deems at risk. B. Predicting Prevention (and its Discontents) Figure 2. Chapin Hall, Family First Readiness, Array of Data Analytics for Family First Provisions18 The family regulation system employs two separate layers of prediction. The first is composed of algorithms whose developers claim they predict the future prospects of a growing, racialized “underclass.” The second can be seen in the transition to a prevention services model—in which families deemed “at risk of imminent removal” (meaning that their children would be taken and placed into foster care) are enrolled instead into the child welfare roster under the guise of prevention, and subject to surveillance, monitoring, and other forms of coercive control. Predictive analytics is not just “coding over the cracks”19 in the broken foundation of the child welfare system. The differential response programs that child welfare departments in a number of states have adopted— sorting parents into risk welfare-2/abolishing-policing-also-means-abolishing-family-regulation/44480 [https://perma.cc/N26N-QC8C]. 18 CHAPIN HALL, AT UNIVERSITY OF CHICAGO, BRIDGING THE GAP BETWEEN WHAT WE KNOW AND WHAT WE DO (2016), https://www .chapinhall.org/wp-content/uploads/Chapin-Hall-Family-First-Framework.pdf [https://perma.cc/4Q6L-A5UA]. 19 Stephanie K. Glaberson, Coding Over the Cracks: Predictive Analytics and Child Protection, 46 FORDHAM URB. L.J. 307, 310 (2019). 82 COLUM. J. RACE & L. F. [Vol. 11:75 categories— animates child welfare agencies’ use of predictive analytics. Providing in-home services to those identified as low risk of maltreating their children and reserving more coercive forms of surveillance and foster care for parents rated high risk might seem like progressive policy on the surface. But as Dorothy Roberts points out: [T]hese dual track systems based on risk assessments can have precisely the opposite effect, sweeping into the carceral net low-risk individuals who previously would not have been on the government’s punitive radar at all. Struggling parents who are targeted by automated models become subject to agency monitoring and therefore more vulnerable to losing custody of their children even though they are unlikely to harm them. 20 Public sector adoption of automated decision systems (ADS) has been accompanied by concern with the technical complexity and opacity inherent to Black Boxes. Civil Rights attorney, Rashida Richardson’s seminal report, Confronting Black Boxes: A Shadow Report of the New York City’s Automated Decision System Task Force21 deploys22 the term as a double entendre, referring to both “the opaque nature of government processes and data-driven technologies.”23 Foregrounding opacity has informed policies demanding greater transparency of algorithmic systems and the data collection it requires. Counter- intuitively, this discursive turn has made it more difficult for the public to discern the “new modes of state surveillance and 20 Dorothy E. Roberts, Digitizing the Carceral State, 132 HARV. L. REV. 1695, 1723 (2019) (book review). 21 AI NOW INSTITUTE, CONFRONTING BLACK BOXES: A SHADOW REPORT OF THE NEW YORK CITY AUTOMATED DECISION SYSTEM TASK FORCE (Rashida Richardson ed., 2019), https://ainowinstitute.org/ads-shadowreport-2019.pdf [https://perma.cc/888D-XDCX]. Unlike the government task force, the report makes specific recommendations for accountability of city agencies including that of the Administration for Children’s Services (ACS). 22 Elizabeth R. Petrick, Building the Black Box: Cyberneticians and Complex Systems, 45 SCI., TECH. & HUM. VALUES 575 (2020). This report reflects the most commonly understood definition of Black Box. In cybernetics and in electrical engineering, these are described as models where inputs and outputs are known but the internal processes producing given outputs from given inputs are unknown. 23 Id. at 2. 2021] CALCULATING THE SOULS OF BLACK FOLK 83 control”24 constructed through ADS deployment. Demands for transparency do not just “privileg[e] a type of accountability that needs to look inside . . . [rather than] . . . across . . . sociotechnical systems,”25 they elide holistic analysis of how governance is inextricably linked to policing and fundamentally desires to “produce [the] truth of the social world.”26 Transparency doesn’t contest the state’s authority to do so, rather, it “reinforce[s] . . . [dominant] narratives about . . . what it mean[s] to maintain order in a society.”27 The garbage in, garbage out debate makes sense, within a narrow frame. Racist police departments can’t help but produce “dirty data.”28 This flawed data creates a feedback loop, codifying the historic legacy of America into predictive policing systems, which in turn directs the police to be in the neighborhoods they already occupy. There’s an ease in which we can make out the automating of inequality, but it’s much harder to see the infrastructure or the digital poorhouse29 being built through an iterative process of classifying and predicting the futures of Black, Indigenous30 and poor people. ADS accelerates and scales institutional racism while simultaneously building new forms of containment and discipline. Both must be understood together, if we are to map the implications of the turn to automated systems and the possibilities for resistance and refusal. 24 Roberts, supra note 20, at 1699. 25 Mike Ananny & Kate Crawford, Seeing Without Knowing: Limitations of the Transparency Ideal and Its Application to Algorithmic Accountability, 20 NEW MEDIA & SOC’Y 973, 974 (2016) (emphasis omitted). 26 We Be Imagining Podcast, Tired of the Future, Time for the Now (with Josh Scannell), AM. ASSEMBLY (May 29, 2020), https://americanassembly .org/wbi-podcast/tired-of-the-future-time-for-the-now [https://perma.cc/YW9F- JNEA]. See also Ingrid Burrington, Policing is an Information Business, URB. OMNIBUS (June 29, 2018), https://urbanomnibus.net/2018/06/policing-is-an- information-business/ [https://perma.cc/3Y9G-TAXB]. 27 We Be Imagining Podcast, supra note 23. 28 Rashida Richardson et al., Dirty Data, Bad Predictions: How Civil Rights Violations Impact Police Data, Predictive Policing Systems, and Justice, 94 N.Y.U. L. Rev. 192 (2019). 29 VIRGINIA EUBANKS, AUTOMATING INEQUALITY 184 (2017). 30 N.Y. OFF. CHILD. & FAM. SERV., supra note 2, at 7. Less than 0.3% of children in New York State Foster Care were classified as Native American/Alaska Native but this is not reflective of national trends where children from Indigenous nations are removed from their families at disproportionate rates. 84 COLUM. J. RACE & L. F. [Vol. 11:75 The 2018 federal Family First Prevention Services Act (FFPA) financially incentivizes state child welfare agencies to transition from a primarily “removal” or separation-based system to a prevention services model. Again, while the Dunantist31 humanitarian language of “prevention” and “support for the poor” sounds progressive, beneath the rhetoric, we see a very different material reality. And we witness how a model labeled “prevention services” actually functions to extend the scope of the carceral state. In the literal sense, preventing family separations is a noble commitment. However, we have to ask why the US Immigration and Customs Enforcement (ICE) and municipal child welfare agencies separate families to begin with. Is it because they have not had the good fortune to be enrolled into the supervision of agencies that operate the foster care system? Or is it something else? In answering this question, we must recognize something that is not immediately apparent in the banal language of the bill: that expanding data collection, risk assessments and predictive analytics is central32 to the project of “predicting who needs prevention” and memos guiding implementation of the Family First Prevention Act. C. Get Out Mathematics33 While this disconcerting realization might appear as cause for despair, it points to a bracing and urgently necessary revelation: it is impossible to calculate the souls of Black folk. Indeed, the actuarial approach to human decision making34 continuously reproduces a mythology35 of Black inferiority, in part by codifying a transposition of demographic traits that 31 Stuart Gordon & Antonio Donini, Romancing Principles and Human Rights: Are Humanitarian Principles Salvageable?, 97 INT’L REV. RED CROSS 77 (2016), https://international-review.icrc.org/sites/default/files/irc_97_1-2-5.pdf [https://perma.cc/X59Y-RBZK]. 32 CHAPIN HALL, supra note 18. 33 Katherine McKittrick, Mathematics Black Life, 44 BLACK SCHOLAR 16 (2014), https://ageingcompanions.constantvzw.org/books/Mathematics_Black _Life.pdf [https://perma.cc/B6K3-96AX]. Here, I’m animated by McKittrick’s question: “What if we trust the lies—she says she was born free—and begin to count it out differently?” in tandem with Jordan Peele’s film GET OUT (Universal Pictures 2017). 34 Robyn Dawes et al., Clinical Versus Actuarial Judgment, 243 SCIENCE 1668 (1989). 35 Megan J. Wolff, The Myth of the Actuary: Life Insurance and Frederick L. Hoffman’s Race Traits and Tendencies of the American Negro, 121 PUBLIC HEALTH REPORTS 84 (2006), https://www.ncbi.nlm.nih.gov/pmc/articles /PMC1497788/ [https://perma.cc/JC43-M26A]. 2021] CALCULATING THE SOULS OF BLACK FOLK 85 correlate to poor social outcomes as the cause of poor social outcomes. The objectivist rhetoric surrounding big data36 directs us to adjudicate the downstream social impact of emerging technologies, asking questions like: Does the Allegheny Family Screening Tool (AFST) produce fair outcomes? Who do we hold accountable if it doesn’t? Is there enough transparency about the algorithm and its implementation for us to determine the social impact? However, I argue that the fairness, accountability and transparency framework (FAT), while helpful as a narrow set of methods for adjudicating downstream impact, does not address the core structural issues at work; it only allows us to name and manage the immediate predicament we find ourselves in. Instead, we must consider “[t]he epistemological economy in which Souls [of Black Folks] operates . . . [and] calls for thinking [of] a world other than the familiar world master named as objects and things.”37 Pragmatists may find this call for epistemic correction frivolous. But allowing the carceral state to define the limit of our imagination will ensure that our efforts to dismantle the “digital poorhouse” will only produce an apparatus to regulate and manage its expansion. Accepting the terms on which these technologies are developed or deployed, limits redress to tinkering with the parts that compose them. The remedy sought by those harmed by stochastic recognition is not increased accuracy. In fact, historically marginalized people are better38 positioned to perceive how these systems are mediated by race and undermine human autonomy39 by design. What dominant technical perspectives interpret as failure or intractable second order effects40 of automation may also be a site of fugitive 36 Dominik Balazka & Dario Rodighiero, Big Data and the Little Big Bang: An Epistemological (R)evolution, FRONTIERS IN BIG DATA (Sept. 18, 2020), https://doi.org/10.3389/fdata.2020.00031 [https://perma.cc/JFC4-ZCY2]. 37 Kevin Thomas Miles, Place Post-Paradise: Poetic Epistemology in the Souls of Black Folk, 15 NEW CENTENNIAL REV. 83, 86 (2015). 38 CARLA SHEDD, UNEQUAL CITY: RACE, SCHOOLS AND PERCEPTIONS OF INJUSTICE 59 (2015). 39 Christopher Poulin et al., Response to: Autonomous Vehicles, INGENIA (Mar., 2015) https://www.ingenia.org.uk/Ingenia/Issue-62/Letters [https://perma.cc/WW5N-NEAW]. 40 Andrew J. Hawkins, Deadly Boeing Crashes Raise Questions About Airplane Automation, VERGE (Mar. 15, 2019), https://www.theverge.com/2019/3 /15/18267365/boeing-737-max-8-crash-autopilot-automation 86 COLUM. J. RACE & L. F. [Vol. 11:75 possibility.41 Stefano Harvey and Fred Moten assert42 in The Undercommons, “Knowledge of freedom is (in) the invention of escape, stealing away in the confines, in the form of a break.” In other words, undermining the fiction that carceral governance43 produces (and relies on) about the value (and devaluation) of human life is the only path to imagining, and charting, a way out and beyond. Even absent full transparency from the child welfare system, public records alone document the rapidly expanding reach of the “digital poorhouse.” Roberts’ insight that, “racism is central to the carceral state’s reliance on prediction” guides us to see how assumptions of Black pathology are rearticulated as risk management. It is important to remember that the official statistics may offer a partial tally of the violence (enacted onto families in the name of “care”) but it does not record Black agency. The indeterminacy of Black life—classified as a negative externality for predictive risk modeling by the state (and capital)—is the basis of a “Get Out mathematics” which recognizes escape’s omission from the ledger. 41 FRED MOTEN & STEFANO HARNEY, THE UNDERCOMMONS: FUGITIVE PLANNING & BLACK STUDY 51 (2013). 42 Id. 43 CARLA SHEDD, UNEQUAL CITY: RACE, SCHOOLS AND PERCEPTIONS OF INJUSTICE (2015). 2021] CALCULATING THE SOULS OF BLACK FOLK 87 II. INTENT TO DESTROY Figure 3. New York State, Office of Children Family Services44 Every time a familial bond is shattered, it’s a tragedy. This loss is under-appreciated by those unaffected because it happens outside of public view with the highest rates of removal clustered in a handful of neighborhoods. Caseworkers operate like police officers under another name45. Their routine scope of 44 New York State Office of Children Family Services (@NYSOCFS) (Nov. 29, 2020, 10:10 A.M.), https://twitter.com/NYSOCFS/status /1333063167891554311 [https://perma.cc/LJ2G-BQLM]. 45 Tarek Z. Ismail, The Consent of the Compelled: Child Protective Agents as Law Enforcement Officers (July 7, 2021) (unpublished manuscript) (on file with author). 88 COLUM. J. RACE & L. F. [Vol. 11:75 work46 includes intruding into the intimate spaces of Black families, strip searching all children in the home, demanding entry without a warrant, without a reading of one’s Miranda rights, nor providing an attorney at the onset of an investigation. This image of officers of the state intruding into the home— claiming their presence as care and separating Black families for their own safety—is not simply analogous to plantation logics; it is their evolution. 47 The passage of the 1997 Adoption and Safe Family Act (ASFA) expedited the termination of parental rights for children removed into foster care, producing generations of legal orphans. Black children are most likely to have their parents’ rights terminated, most likely to languish in care without adoption once the rights are terminated, most likely to be living in congregate care settings, and most likely to experience multiple foster boarding home placements. The circulation of classifieds advertising children in foster care— rendered legal orphans by ASFA—for adoption, embodies the simultaneous discarding and commodifying of children who come into care. In order to understand the impact that family separation continues to have from its height to the present, it’s important to remember that the 50,000 children who were in foster care in 1996 are now between 25 and 46 years old. And it’s important to recognize that the number one predictive factor for a child being investigated by child protective services is that their mother had a previous child welfare history. This means that many survivors of the system, whose family bonds were shattered as children, find themselves ageing out of foster care only to experience their own children being taken by the same system. The child welfare system is not biased, or “unfair”––it is racist. As Ruth Wilson Gilmore clearly explains: racism is the state-sanctioned and/or extralegal production and exploitation of group-differentiated vulnerability to premature death.48 Working with Gilmore’s insight, we see the outlines of child welfare’s “cradle to the grave” system, how it produces vulnerability first through collective 46 Victoria Copeland, “It’s the Only System We’ve Got”: Exploring Emergency Response Decision-Making in Child Welfare, 11 COLUM. J. RACE & L.F. 43 (2021). 47 Katherine McKittrick, On Plantations, Prisons, and a Black Sense of Place, 12 SOC. & CULTURAL GEOGRAPHY 947 (2011). 48 GILMORE RUTH WILSON, GOLDEN GULAG: PRISONS, SURPLUS, CRISIS, AND OPPOSITION IN GLOBALIZING CALIFORNIA (2007). 2021] CALCULATING THE SOULS OF BLACK FOLK 89 surveillance, separation, and abandonment, and then imprisonment, disability and death. To encounter this system is to see what Sharpe defines as the wake: Living in/the wake of slavery is living ‘the afterlife of property’ and living the afterlife of partus sequitur ventrem (that which is brought forth follows the womb), in which the Black child inherits the non/status, the non/being of the mother. That inheritance of a non/status is everywhere apparent now in the ongoing criminalization of Black women and children.49 The scale at which the New York City Administration for Children’s Services (ACS) has forcibly removed and continues to transfer Black children into “care” is forecast by “the weather, [where] antiblackness is pervasive as climate.”50 It also may meet the criteria of cultural genocide.51 Article 2(e) of the Genocide Convention lists “[f]orcibly transferring children of the group to another group”52 as one of the acts committed with intent to destroy, in whole or in part, a national, ethnical, racial or religious group to assess whether genocide has been committed. Notably, the current legal mandate requires a consideration of the “the best interest of the child”.53 This individualist conception of safety is at odds with the Genocide Convention’s commitment to preserving the group. As legal scholar Kurt Mundorff’s substantive research on genocide indicates, it is clear that a citizen’s tribunal is needed to assess54 why so many Black 49 CHRISTINA SHARPE, IN THE WAKE: ON BLACKNESS AND BEING 15 (2016). 50 Id. at 106. 51 KURT MUNDORFF, A CULTURAL INTERPRETATION OF THE GENOCIDE CONVENTION (2020). 52 Convention on the Prevention and Punishment of the Crime of Genocide, art. 2(e), Dec. 9, 1948, S. Exec. Doc. O, 81-1 (1949), 78 U.N.T.S. 277. 53 N.Y. FAM. CT. § 1027 (2016). 54 CIVIL RIGHTS CONGRESS, WE CHARGE GENOCIDE: THE HISTORIC PETITION TO THE UNITED NATIONS FOR RELIEF FROM A CRIME OF THE UNITED STATES GOVERNMENT AGAINST THE NEGRO PEOPLE (1951). It’s worth noting that the Civil Rights Commission submitted a complaint to the United Nations General Assembly charging that the United States Government’s intentional destruction of Negro people amounts to genocide. W. E. B. Du Bois was among the leaders presenting the complaint to the UN in 1951 but the UN refused to acknowledge receipt of the petition. This refusal is frequently attributed to the United States’ outsized influence over the UN and longstanding reservations to the Genocide Convention. https://codes.findlaw.com/ny/family-court-act/fct-sect-1027.html%20N.Y.%20Fam.%20Ct.%20%C2%A7%201027 90 COLUM. J. RACE & L. F. [Vol. 11:75 children have been removed from their families in New York City and to what degree they have been transferred into placements outside their culture. III. CREATING “PREVENTED” POPULATIONS Power is relational.55 If building up a new model of child “welfare” produces inequity for some, it also produces benefits for others. To understand how “prevented” populations are created, we must examine how the incentive structures that animated previous models of child welfare have shifted, and what incentives are being constructed in their place. The “epic struggle to change the child welfare system”56 did not begin with the development of new technologies. In fact, the history of endless reforms in the child welfare system may make it hard to appreciate what is qualitatively new about the “digital poorhouse” and the narrative move to codify “prevention” as the primary disposition for families who encounter it. In this section, we will review the implications of the Family First Prevention Act, explain what preventive services are and how ACS in particular has responded to the legislation. Similar to subfelony enforcement—where a substantial number of actions terminate in a disposition that involves no jail time or criminal conviction—57the majority of child maltreatment investigations today do not result in removal. This fact is not offered to minimize those that do, nor the plight of the more than 7600 children currently in New York City foster care, the majority of whom are Black and from neighborhoods with the highest rates of arrest and COVID-19 related deaths. Rather, understanding the vast amount of investigations that do not result in removal helps us heed Hausmann’s guidance on the value of understanding these less visible consequences of policing: [E]xclusively focusing on the historically and internationally unprecedented numbers of people the United States puts into prison and jails cells understates the reach of the criminal justice 55 RUHA BENJAMIN, RACE AFTER TECHNOLOGY: ABOLITIONIST TOOLS FOR THE NEW JIM CODE 53 (2019). 56 NINA BERNSTEIN, THE LOST CHILDREN OF WILDER: THE EPIC STRUGGLE TO CHANGE FOSTER CARE (2001) 57 ISSA KOHLER-HAUSMANN, MISDEMEANORLAND: CRIMINAL COURTS AND SOCIAL CONTROL IN AN AGE OF BROKEN WINDOWS POLICING 2 (2018). 2021] CALCULATING THE SOULS OF BLACK FOLK 91 system and, in some sense, misrepresents the modal criminal justice encounter . . . . [I]f we want to understand the precise ways the criminal justice system functions as a form of social control, we need to look beyond custodial sentences and even criminal convictions.58 While over 84,000 children were investigated59 by NYC ACS in 2019, only 4% of those children were removed from their families and placed into foster care. We must document the reaches of this carceral system, which enrolls subjects, even if it does not remove children from parents. Exclusively focusing on removal to assess the function of child welfare produces an incomplete picture of how the system operates. If we more closely examine ACS data for 2019, over 45,000 children were enrolled in preventative services.60 There are almost as many children enrolled in preventive services as there were children removed into foster care during the peak of family separations in the late 1990s. The prevention services roster in 2019 reflects more than 22% of all NYC families living in poverty.61 These are the only numbers available without having access to data disaggregated by race—given the concentration of preventive service enrollment in neighborhoods that are “majority minority” the percentage of families who are poor and Black or Latinx enrolled in prevention is likely much higher. Joyce McMillan, impacted parent and Executive Director of the Parent Legal Action Network (PLAN), is vocal about communities needing “support not surveillance.” The public conversation around Allegheny’s use of predictive analytics has raised alarms for McMillan, as well as other advocates and 58 Id. 59 NEW YORK CITY ADMINISTRATION FOR CHILDREN’S SERVICES, ABUSE/NEGLECT INVESTIGATIONS BY COMMUNITY DISTRICT, 2015–2019 (2019), https://www1.nyc.gov/assets/acs/pdf/data-analysis/abuseneglectreport15to19 .pdf [https://perma.cc/543R-97QL]. 60 NEW YORK CITY ADMINISTRATION FOR CHILDREN’S SERVICES, CHILDREN SERVED BY CHILD WELFARE PREVENTION SERVICES BY HOME BOROUGH/CD (2019), https://www1.nyc.gov/assets/acs/pdf/child_welfare/2020 /ChildrenReceivingPreventiveServicesByCDCY2019.pdf [https://perma.cc/QB9J-HRX6]. 61 THE ANNIE E. CASEY FOUNDATION: KIDS COUNT DATA CENTER, CHILDREN IN POVERTY (100 PERCENT POVERTY) IN NEW YORK CITY, https://datacenter.kidscount.org/data/tables/43-children-in-poverty-100- percent-poverty [https://perma.cc/YZ2Z-5DQE]. https://www1.nyc.gov/assets/acs/pdf/data-analysis/abuseneglectreport15to19.pdf 92 COLUM. J. RACE & L. F. [Vol. 11:75 attorneys, but how exactly is predictive analytics being implemented by ACS in relation to preventive services? What happens to the “prevented” populations and where do we look to find out? A. The Federal Family First Prevention Act A history of child welfare reform—in particular the piloting of services other than foster care by state child welfare agencies—predates the 2018 passage of the Family First Prevention Act. However, it’s worth focusing on the act as it defines what prevention services are, while displacing the previous incentive structure and corresponding motivations that propelled state child welfare agencies. Introduced to the House as bill H.R. 253 in January of 2017 by Florida’s Republican Representative, Vern Buchanan, the purpose of the act is framed around the overhaul of child welfare financing: To amend parts B and E of title IV of the Social Security Act to invest in funding prevention and family services to help keep children safe and supported at home, to ensure that children in foster care are placed in the least restrictive, most family-like, and appropriate settings, and for other purposes.62 Understanding the implications of changes enacted by child welfare’s new guard doesn’t require insight into the psychology of system administrators. Rather, we must survey the “prevention-focused infrastructure” of legislation finances.63 There are 8 parts of FFPA. Part 1: Prevention Activities Under Title IV–E: States Have the Option to Use Title IV-E to Prevent Children’s Entry into Foster Care, is arguably the most significant component of the legislation. This part is what provides, in theory, an opportunity for states to receive unlimited federal matching funds for children and their families. Prior to the passage of FFPA, federal funding for foster care was based on family income and the state’s Federal Medicaid Assistance Program (FMAP) eligibility rate. In some states, the FMAP rate 62 Family First Prevention Services Act of 2017, H.R. 253, 115th Cong. § (2017). 63 Family First Prevention Services Act, NAT’L CONF. STATE LEGISLATURES (Apr. 1, 2020), https://www.ncsl.org/research/human-services /family-first-prevention-services-act-ffpsa.aspx [https://perma.cc/FJG8-TSQ5]. 2021] CALCULATING THE SOULS OF BLACK FOLK 93 has meant that as few as 38% of children placed into foster care are determined eligible for Federal matching funds. FFPA removes all means testing, providing federal matching to states for 100% of families enrolled into preventive services. While the criteria for removing a child from their family is arguably subjective, the criteria for determining whether a child is “a candidate for foster care” is even more amorphous.64 FFPA and the Program Instructions released by the Administration for Children and Families (ACF) allow state65 and Native American Nations66 (referred to as tribal agencies in the legislation) full discretion in determining candidacy. The bill describes candidacy as a child who is at risk for imminent removal but can stay safely in the home if given “evidence-based prevention services.”67 Child welfare agencies that have zero incentive to classify families as anything other than at risk, given the funding this legislation put on the table, are the decision point for approving enrollment. To understand the scale of this financing overhaul—in addition to the removal of means testing and broadening of criteria of who can be enrolled into child welfare supervision—we must understand how the act enables entire households, including the adult members, to be enrolled into prevention services with federally matched funds. The act clearly changes the financial incentive structure68 in order to motivate child welfare agencies to transition to a primarily prevention services model. The impact of the legislation is not only a reallocation of 64 FAMILY FIRST PREVENTION SERVICES ACT: CANDIDACY BY JURISDICTION, CHAPIN HALL & CASEY FAMILY PROGRAMS (Mar. 12, 2020), https://www.chapinhall.org/wp-content/uploads/PDF/Jurisdictional-Candidacy- 3.12.20.pdf [https://perma.cc/8ZZ6-X33M]. 65 U.S. DEP’T HEALTH & HUM. SERV., ADMIN. CHILD., YOUTH, & FAM., ACYF-CB-PI-19-06 (July 18, 2019), https://www.acf.hhs.gov /sites/default/files/documents/cb/pi1906.pdf [https://perma.cc/NN6S-44GA]. 66 U.S. DEP’T HEALTH & HUM. SERV., ADMIN. CHILD., YOUTH, & FAM., ACYF-CB-PI-18-10 (Nov. 30, 2018), https://www.acf.hhs.gov/sites /default/files/documents/cb/pi1810.pdf [https://perma.cc/2Z8B-P4YU]. 67 Id. 68 National Conference of State Legislators, The Family First Prevention Services Act: Reforming Foster Care, FACEBOOK (Aug. 5, 2019), https://www.facebook.com/NCSLorg/videos/709946162790123 [https://perma.cc /R7QK-J4BU]. The bill has some detractors, including Wisconsin State Republican Representative Joan Ballweg, who claimed FFPA is too restrictive and Wisconsin would have preferred Block grants. Block grants provide states greater autonomy in determining how funding is implemented. 94 COLUM. J. RACE & L. F. [Vol. 11:75 federal funding, we also have to attend to the formalizing of prevention services’ definition. B. What Are Prevention Services? The four types of programs eligible for Title IV-E funding under the new act are: (1) mental health prevention and treatment services; (2) substance abuse prevention and treatment services provided by a qualified clinician; (3) in-home parent training/counseling; and (4) kinship navigator programs.69 ACF and the US Department of Health and Human Services have established a Title IV-E Prevention Services (Family First) Clearinghouse to rate and review applicant programs. The Clearinghouse rates programs as well supported, supported, promising or not supported based on a systematic literature review. The three contributing factors to ratings are: (1) risk of harm; (2) duration of effects after treatment is ended; and (3) a requirement that the “favorable evidence for a program or service . . . [that was] obtained . . . [in a] usual care or practice setting.”70 An odd quirk of the Clearinghouse, is that by its own admission, all of the highest rated programs list more reviews demonstrating71 that they are ineffective or harmful than studies demonstrating “Favorable Effects.” In other words, prevention services are extremely criteria driven irrespective of whether they’re helpful or harmful as per their stated mission. The evidence based medicine (EBM) movement—that the act uses to define and evaluate preventive services—advocates for a focus on greater objectivity in medical decision making.72 69 CHILDREN’S DEFENSE FUND ET. AL, IMPLEMENTING THE FAMILY FIRST PREVENTION SERVICES ACT: A TECHNICAL GUIDE FOR AGENCIES, POLICYMAKERS AND OTHER STAKEHOLDERS (2020), https://www .childrensdefense.org/wp-content/uploads/2020/07/FFPSA-Guide.pdf [https://perma.cc/T76V-9DY6]. 70 TITLE IV-E PREVENTION SERVICES CLEARINGHOUSE, HANDBOOK OF STANDARDS OF PROCEDURE, VERSION 1.0 (2019), https://preventionservices .abtsites.com/themes/ffc_theme/pdf/psc_handbook_v1_final_508_compliant.pdf [https://perma.cc/9EDD-Z76C]. 71 TITLE IV-E PREVENTION SERVICES CLEARINGHOUSE, FIND A PROGRAM OR SERVICE https://preventionservices.abtsites.com /program?combine_1=&prograting%5B1%5D=1 [https://perma.cc/M5AA-U923] (last visited Aug. 20, 2021). 72 AMERICAN PSYCHOLOGICAL ASSOCIATION, Template for Developing Guidelines: Interventions for Mental Disorders and Psychosocial Aspects of Physical Disorders (1995). This movement began in the 1970s and was revived in 1995 by the APA white paper. 2021] CALCULATING THE SOULS OF BLACK FOLK 95 The methodological73 and epistemic74 limitations of EBM are well explored in the literature: “[f]ailure to acknowledge the limitations of clinical trials and systematic reviews has limited their applicability to individual patients’ circumstances;”75 overreliance on mechanistic forms of knowledge production and cultural biases about what qualifies as evidence. On one hand, the most interesting aspect of FFPA is that preventive services are all behavioral health programs typically covered by Medicaid.76 This exemplifies the “therapeutic spirit of neoliberalism,” which seeks power during profound socioeconomic inequity through management of citizen’s subjectivity.77 It is also reminiscent of protest psychosis,78 “a condition in which the rhetoric of the Black Power movement drove ‘Negro men’ to insanity.”79 In The Protest Psychosis: How Schizophrenia Became a Black Disease, Jonathan Metzl argues that because racism is “historically embedded into the very DNA of healthcare . . . . [F]ocusing on the individual obscures the impact of the structural, while putting undue pressure on even well-intentioned patients or doctors to solve problems in ten- minute office visits that have taken decades or even centuries to evolve.”80 73 Glenn D. Shean, Some Methodological and Epistemic Limitations of Evidence-based Therapies, 32 PSYCHOANALYTIC PSYCH. 500, 500–16 (2015). 74 R E Ashcroft, Current Epistemological Problems in Evidence Based Medicine, 30 J. MED. ETHICS 131, 131–35 (2004). 75 Desmond J. Sheridan & Desmond G. Julian, Achievements and Limitations of Evidence-based Medicine, 68 J. AM. COLL. CARDIOLOGY 204, 204– 13 (2016). 76 NEW YORK STATE KINSHIP NAVIGATOR, https://www .nysnavigator.org/?page_id=542Z [https://perma.cc/39C3-D6FB] (last visited Mar. 5, 2021). The exception is “kinship navigator programs” which are state based information hotlines for kinship caregivers. 77 Roger Foster, The Therapeutic Spirit of Neoliberalism, 44 POL. THEORY 82, 82–105 (2016). 78 This term originated from New York Psychiatrists Walter Bromberg and Franck Simon. Walter Bromberg & Franck Simon, The “Protest” Psychosis: A Special Type of Reactive Psychosis, 19 ARCH GEN PSYCHIATRY 155 (1968). 79 JONATHAN M. METZL, THE PROTEST PSYCHOSIS: HOW SCHIZOPHRENIA BECAME A BLACK DISEASE 100 (2009). 80 Id. at 202. 96 COLUM. J. RACE & L. F. [Vol. 11:75 Figure 4. Johnathan M. Metzl, Protest Psychosis: How Schizophrenia Became a Black Disease81 However, the most critical aspect of the legislation, hidden in the footnotes, is that Title IV-E funds are the payor of last resort for prevention services. This means that states who enroll families into prevention services must first attempt to fund the services through Medicaid, Temporary Assistance for Needy Families (TANF) or any other possible source prior to being approved for Title IV-E funds. If you take away the services funded by Medicaid from the core section of the Family First Prevention Act, all that remains of the “prevention-focused infrastructure” is data collection and predictive risk modeling. Therefore, “the prevented populations” are people enrolled into community surveillance programs and a site of discipline by the therapeutic state. 81 Id. at xiv. 2021] CALCULATING THE SOULS OF BLACK FOLK 97 C. When Predictive Analytics Became the Remedy for an Abusive System In 2016, Zymere Perkins was murdered by his stepfather and mother despite the family having been investigated by ACS multiple times. This was followed by what Richard Wexler describes as a foster care panic or a surge in family separations by local child welfare agencies as a knee-jerk reaction to a high- profile death of a child.82 However, as Wexler points out, these media narratives omit that a child is more likely to be seriously injured or murdered while in foster care than with their parents—who most often come to the attention of child welfare agencies due to poverty, and its pressures on domestic life, not due to inherent inability as parents. This double violence: first separation from the family, and then placement into “care,” where children in custody of ACS frequently encounter new forms of abuse and abandon/ment—was the basis for the class action lawsuit filed in 2015, Elisa W., et al. v. The City of New York.83 The State agreed to 7-year settlement84 in response to the class action lawsuit while ACS refused. A central feature of the settlement agreement, was OCFS would designate an independent monitor to assess ACS’s compliance with the guidance set forth in a consent decree.85 The same month Hansell was appointed as Commissioner of ACS, Kroll—a global investigative and risk management corporation—was selected by OCFS to serve as the independent monitor.86 Eight months later, ACS hailed Kroll’s newly-published report in a press release: “INDEPENDENT MONITOR’S REPORT SHOWS “THE STEPS WE’VE BEEN TAKING ARE THE RIGHT ONES.”87 But an 82 Richard Wexler, Panic in Foster Care, L.A. TIMES (Sept. 16, 2009, 12:00AM), https://www.latimes.com/archives/la-xpm-2009-sep-16-oe-wexler16- story.html [https://perma.cc/L2GP-S8XJ]. 83 Elisa W. v. City of N.Y., No. 15 CV 5273-LTS-HBP, 2016 U.S. Dist. LEXIS 123332 (S.D.N.Y. Sep. 12, 2016). 84 Consent Decree, Elisa W. v. City of N.Y., at ¶3.2. 85 Id. 86 Melissa Russo, Exclusive: NYC Pays Significant Sum to Firm Tasked With Reforming City’s Embattled Administration for Children’s Services, NBC NEW YORK (May 4, 2017), https://www.nbcnewyork.com/news/local /administration-for-childrens-services-scandal-cases-abuse-consultant-pay-cost- new-york-city-mayor-bill-de-blasio/121675/ [https://perma.cc/F3P2-EDEV]. 87 Press Release, N.Y.C. Administration for Children’s Services, ACS: Independent Monitor’s Report Shows “The Steps We’ve Been Taking Are the 98 COLUM. J. RACE & L. F. [Vol. 11:75 important question was left unanswered: whose risk is Kroll mitigating, responding to, and remediating? The report makes no mentions of the extreme racial disproportionality, delayed permanency in the form of adoption or reunification and complex developmental trauma it causes for children and families within the New York City child welfare system––the claims that formed the legal pretext for their appointment as independent monitor. The Kroll report does dedicate a section to praising ACS for using predictive analytics: ACS is currently in the process of implementing Safe Measures, an innovative web-based data dashboard developed by the NCCD [the National Council on Crime and Delinquency],88 a non-profit research organization. Safe Measures will provide ACS a platform for using predictive analytic models recently developed by ACS and its academic partners at the University of Chicago, City University of New York and New York University. Each of these predictive models uses approximately 170 different data points, such as: the number of prior ACS cases involving a family; the number of times the family received preventive services; the ages of the parents and children; the number of siblings; the types of allegations and priority codes, etc., to determine what variables correlate to adverse outcomes and what combinations of variables are associated with the greater likelihood of such outcomes occurring.89 Kroll recommends improving this practice by changing interagency data sharing protocols for example between ACS and the New York City Housing Authority or the New York Police Department (NYPD) because “[t]hese [additional] data points Right Ones” (Dec. 12, 2017), https://www1.nyc.gov/assets/acs/pdf/PressReleases /2017/IndependantReport.pdf [https://perma.cc/3G4B-AUC6]. 88 Evident Change NCCD was the acronym for the National Council on Crime and Delinquency. In November 2020, the National Council on Crime and Delinquency and the Children’s Research Center became Evident Change. 89 KROLL, KROLL REPORT: THE NEW YORK CITY ADMINISTRATION FOR CHILDREN’S SERVICES 19 (2017) (emphasis added), https://ocfs.ny.gov/main /reports/2017-NYC-ACS-Report-Kroll.pdf [https://perma.cc/9JM4-WKNG]. https://www.evidentchange.org/ 2021] CALCULATING THE SOULS OF BLACK FOLK 99 could be used by ACS, in combination with CONNECTIONS90 data, to develop additional predictive analytic models that would further identify factors that place families at risk. This would afford both ACS and the respective agency the opportunity to take corrective action proactively.”91 This narrative assumes that the dangers to children and their families are located within them and their communities, rather than in the carceral continuum—of which ACS is a part of—which produces conditions of unsafety through separation, surveillance, and investigation. It seeks to access data collected, and created, by the NYPD, but not data reflecting the frequency of which the NYPD murders and brutalizes Black children. It seeks data collected, and created, by the New York City Department of Homeless Services on the behavior of their family “clients,” but it does not seek to integrate data on the squalid conditions of DHS family shelters which jeopardize the safety of children.92 As Christina Sharpe writes: Put another way, living in the wake means living in and with terror in that much of what passes for public discourse about terror we, Black people, become the carriers of terror, terror’s 90 N.Y. STATE OFF. CHILD. SERV., CONNECTIONS, https://ocfs.ny.gov /connect/ [https://perma.cc/M5HW-GZ27] (last visited Mar. 13, 2021). The New York State Office of Children and Family Services (OCFS) is responsible for overseeing New York’s child welfare system and services. Child welfare services are provided through 58 Local Departments of Social Services (LDSS) and over 350 voluntary and preventive services agencies. Caseworkers and other staff from local districts and agencies use a case management computer system called CONNECTIONS to document child welfare activities. The CONNECTIONS system is part of a Federal initiative called the Statewide Automated Child Welfare Information System (SACWIS). In 1993, the Federal government provided financial incentives for states to develop statewide automated child welfare processes in an effort to provide more efficient and effective administration of programs and Federal reporting. Many other states have similar systems with other names. By law, a SACWIS program is required to support the reporting data to the Adoption and Foster Care Analysis Reporting System (AFCARS) and the National Child Abuse and Neglect Data System (NCANDS). CONNECTIONS was designed to become a single, statewide, integrated system for the collection and recording of child protective, preventive, foster care and adoption service information. 91 KROLL, supra note 89, at 20 (emphasis added). 92 SCOTT M. STRINGER, N.Y.C., OFF. COMPTROLLER, AUDIT REPORT ON THE SAFETY AND WELLBEING OF INFANTS RESIDING IN SAMPLED DEPARTMENT OF HOMELESS SERVICES SHELTERS (2020), https://comptroller.nyc.gov/wp- content/uploads/documents/MG19_110A.pdf [https://perma.cc/2ZPV-UJ5R]. 100 COLUM. J. RACE & L. F. [Vol. 11:75 embodiment, and not the primary object of terror’s multiple enactments; the ground of terror’s possibility globally. This is everywhere clear as we think about those Black people in the United States who can “weaponize sidewalks” (Trayvon Martin) and shoot themselves while handcuffed (Victor White III, Chavis Carter, Jesus Huerta, and more), those Black people transmigrating the African continent toward the Mediterranean and then to Europe who are imagined as insects, swarms, vectors of disease, familiar narratives of danger and disaster that attach to our always weaponized Black bodies (the weapon is blackness).93 It’s clear that ACS’ interest in identifying adverse childhood experiences (ACEs)94 does not include the state sanctioned racist violence that produces the most defining traumatic conditions for families on the receiving end of these surveillance predictions. Then what is the goal of their predictive analytic model? The Kroll report continues: The goal of these predictive models is to provide enhanced services to high-need families rather than to identify children for removal from homes. This approach differentiates ACS’s predictive analytic models from models used by some other child welfare agencies, which tend to use modeling as a means of screening cases and identifying children who may need foster care services, inevitably creating difficult questions about the “false positive” rates the models produce.95 93 SHARPE, supra note 36, at 15–16. 94 Vincent J. Felitti et al., Relationship of Childhood Abuse and Household Dysfunction to Many of the Leading Causes of Death in Adults: The Adverse Childhood Experiences (ACE) Study, 14 AM. J. PREVENTIVE MED. 245, 245–58 (1998). The original Kaiser Permanente Study, identifying a correlation between ACEs and leading causes of death, enrolled 8,506 people, of which 79.4% were white and 43% were college graduates. Despite frequent invocation of ACE as justification for preventive services by child welfare administrators, the demographic of who is enrolled into “prevention” is completely at odds with the original public health data. 95 KROLL, supra note 89, at 19. 2021] CALCULATING THE SOULS OF BLACK FOLK 101 In this passage, Kroll argues that the agency can mitigate the public backlash generated by false positive rates through articulating the purpose of the model away from identifying “whose child should be removed” and towards framing the output as the identification of high needs families for the provision of voluntary services. In highlighting this approach, Kroll is not claiming ACS’ use of SafeMeasures as more accurate in predicting the likelihood of child maltreatment than software like the Allegheny Family Screening Tool (AFST). Kroll is also not claiming that ACS’ predictive risk modeling seeks to answer a fundamentally different question from AFST. Kroll’s primary concern appears to be ensuring that ACS will be able to evade the same intensity of public scrutiny or “difficult questions about ‘false positive rates’ the models produce.”96 And it also suggests that such avoidance can be accomplished by tying the algorithm’s output to the provision of preventive services, while decoupling it from family separations. This focus on preventive services is a narrative feat in its own right, the rhetorical equivalent to cigarette manufacturer claiming benevolence for operating a hospital dedicated to lung cancer. With the narrative focus on prevention, child welfare agencies are going a step further, claiming to prevent the violence they enact—through new services only they can provide. Decreasing their rate of child separations in exchange for massive expansion of community level surveillance, ACS predicts which families they will prevent from being torn apart by themselves. This Kroll report is a signpost directing us towards ACS’s implementation of new modes of surveillance. To understand how they function within preventive services as a method of social control, we must now examine the legal and socio-historical logics that animate their use. In particular, we must expand the scope of our investigation beyond investigations that result in removal. 96 KROLL, supra note 89, at 19. 102 COLUM. J. RACE & L. F. [Vol. 11:75 IV. ACS’ CONSIDERATIONS IN BUILDING A PREDICTIVE RISK MODEL In November of 2017, data science researcher Ravi Shroff wrote a paper97 focusing on lessons he learned while developing a predictive analytics model for ACS in partnership with researchers at the CUNY. In order to do the kind of predictive risk modeling ACS requested Shroff’ assistance with, ACS turned to machine learning which requires large amounts of data. Conveniently for the aims of this project, the State Central Register (SCR) hotline—the central intake in New York State for child maltreatment allegations—does not expunge indicated case records until 10 years after the youngest named child in the report turns 18 years old.98 This method of data collection, creation, and retention generates a uniquely intergenerational data set. However, even these hundreds of thousands of records that were provided to Shroff by ACS, in service of training a predictive machine learning model, were deemed insufficient, giving a sense of the data-hungry nature of machine learning overall. He explains further in his paper: “The scale of available data is also relatively small; for example, each year, ACS conducts roughly 55,000 investigations of alleged abuse or neglect, and New York City receives several million 911 calls (Google or Facebook, on the other hand, analyze datasets that are many orders of magnitude larger).”99 Shroff does not make recommendations in the report about how ACS can increase the scale of available data it has access to. However, we know part of the solution is taking up Kroll’s recommendation to modify memorandums of understanding (MOUs) to improve interagency data sharing in local government and programmatically requiring contracted preventive service agencies to collect data. Shroff also highlights 97 Ravi Shroff, Predictive Analytics for City Agencies: Lessons from Children’s Services, 5 BIG DATA 189, 189–96 (2017). Diane DePanfilis and Maria Rodriguez are co-researchers in the CUNY/NYU Team. The NYC-ACS Predictive Analytics Technical Team as of January 2018 includes a team from Chapin Hall/University of Chicago and ACS. The Chapin Hall Team consisted of Dana Weiner, Brian Chor, Mike Stiehl and Kiljoong Kim. The ACS team consisted of Allon Yaroni, Teresa De Candia, Leila Pree and Synia Wong. 98 N.Y. Soc. Serv. Law § 422 (2021). Legally sealed, unfounded reports are supposed to be expunged 10 years after the report was made. However, in accordance with §422-5(a) reports may be unsealed if a new report is made during this 10-year period. 99 Shroff, supra note 97, at 190. 2021] CALCULATING THE SOULS OF BLACK FOLK 103 methodological limitations of using ACS records for machine learning training poses: Non-representative samples of data are also problematic for applying predictive methods. It can be misleading to apply an algorithm trained on one population to a population with a different distribution of attributes. For example, ACS leadership has been clear in stating that the repeat reports model described in this article will only be used to predict the likelihood that a child already in their system will have another investigation of abuse or neglect. In particular, even if required data were available, the model will not be used to make predictions on the general population of children in New York City. It is important to note that in general, if data collection procedures are strongly biased in some systematic way, the only option may be to devise strategies to improve those procedures.100 Shroff raises concerns about the ability for a model trained on racialized groups below the poverty line to accurately predict the likelihood of child maltreatment and abuse on the rest of the city population. This articulation of the problem alongside his concluding thought—that we devise strategies to improve data collection—perniciously circumvents any mention of the structural inequalities that produce the data sample they do have while simultaneously laying the groundwork to request we collect even more. While ACS is promising that the model will not be implemented on the general population of New York City children, there’s currently no legal framework or mechanism by which we would be notified or able to seek remedy if they did. Additionally, the report does not state whether the children and families who compose ACS’ dataset consented to having their data used to train a machine learning model designed to predict “repeat reports”. This highlights how Black and Brown children—already overrepresented in child welfare data— become a site of experimentation for city agencies developing automated decision-making systems. If the families represented in the data set were to discover their data was used without their 100 Id. at 192. 104 COLUM. J. RACE & L. F. [Vol. 11:75 consent and want recourse, too bad: there is currently none available. Shroff mentions ACS’ reliance on an external ethics advisory board that “will provide valuable advice on how to balance the short- and long-term costs and benefits of the myriad decisions made when implementing machine learning models.”101 However, it appears the iterative software development process—in which a random forest model was selected, the choice of features was made, and the purpose of the model were decided—was informed solely by technologists and ACS leadership, without any consultation with the ethics board. Further, ACS does not appear to have included the advisory board in any of the upstream decisions, and there is currently no publicly available documentation describing their comment. If this ethics advisory board was to form an opinion in opposition to the decisions made by ACS leadership and consulting technologists, there are currently no mechanisms to enforce their position in the face of agency opposition. There is also no mandate for the actions of the board to be communicated to the broader public. V. GEOGRAPHIES OF PREDICTION The digital poorhouse is not a metaphor. It is a layered web of digital infrastructure transmitting petabytes of personal data, through brick-and-mortar data centers which are often metaphorically referred to as “the cloud.”102 The interpretation of these data sets by frontline caseworkers, supervisors or other agency staff is mediated by the user interface of software, along with other intervening computational infrastructures. Technical infrastructure is not inherently bad, the moral value of technology is determined by the ideologies, values, and people who govern its use and creation. Predictive analytics is a socio- technical system (STS) “that considers requirements spanning hardware, software, personal, and community aspects. It applies an understanding of the social structures, roles, and rights (the social sciences) to inform the design of systems that involve communities of people and technology.”103 You cannot abolish or 101 Id. at 195. 102 EUBANKS, supra note 29, at 184. 103 Social-Technical Systems, INTERACTION DESIGN FOUNDATION, https://www.interaction-design.org/literature/topics/socio-technical-systems. [https://perma.cc/X2EY-NCJX] (last visited Mar. 13, 2021). 2021] CALCULATING THE SOULS OF BLACK FOLK 105 dismantle a system you cannot identify, so, before we proceed, we must map the geographies and infrastructures of prediction. In New York City, children placed into foster care or enrolled into preventive services are under the aegis of non-profit agencies which the city pays for provision of services. Each of these contracted agencies is required to maintain case and program data for all child welfare cases in CONNECTIONS, the OCFS system of record. Agencies contracted for prevention must input their data into both ACS’ system for tracking provision of preventive services, the Preventive Organization Management Information System, and use the Safe Measures dashboard to enter the data of enrolled families. A central change instituted by Commissioner David Hansell upon appointment—in tandem with a return to Child Stat—104was the formalizing of a Quality Assurance department in ACS and set of procedures governing their operation. The non-profit agencies contracted out to by the City are instructed to rely on data in their decision making, and to take a “significant role in building capacity to use data and dashboards to guide planning and case practice within agencies, as well as participate in the Quality Assurance/Quality Improvement learning collaborative.”105 A detailed account of how the “digital poorhouse” operates as a financialized computational structure is beyond the scope of this paper. However, it is worth mentioning that the expansive and costly technical capacity necessary to carry out data collection/creation and predictive risk modeling is frequently financed by Pay for Success or Social Impact Bonds. These mechanisms enable private investors to fund public sector programming—receiving a profit if certain performance-based metrics are met. Some have raised concerns that these “innovative financing structures” are a trojan horse for 104 ANNIE E. CASEY FOUND., IMPLEMENTING CHILDSTAT, A “HOW-TO” GUIDE FOR CHILD WELFARE AND CLIENT-SERVING SYSTEMS, (Feb. 2015), https://assets.aecforg/m/resourcedoc/aecf-ImplementingChildStat-2015.pdf [https://perma.cc/NBY6-XPET]. Child Stat was developed by ACS in 2006 as modified version of CompStat which began in 1994 under NYPD Commissioner William Bratton in service of broken windows policing. 105 N.Y.C. ADMINISTRATION FOR CHILDREN’S SERVICES, ACS PREVENTION SERVICES: REQUEST FOR PROPOSALS 40 (2020), http://sachspolicy com/wpcontent/uploads/2019/07/ACS-Prevention-Services-RFP.pdf [https://perma.cc/R8ZQ-ZZY5]. 106 COLUM. J. RACE & L. F. [Vol. 11:75 privatization.106 The digitization of a poverty management system does not merely shift capital and funding to new actors, it renegotiates107 where and how data and power are located. For example, collected data is stored on financialized cloud computing, which maximizes infrastructure utilization and revenue through customizing contractual software agreements.108 The calculative practices of risk assessment central to predictive analytics in the child welfare system—and their assumptions of a high level of control over risk—also “play a crucial role in the overall process of financialization.”109 In her examination of Moody’s, a major credit rating agency, Natalia Besedovsky describes how the fundamental shift in credit rating practices does not just “simply alter the rating methods or models. More importantly, they entail an entirely different set of epistemological assumptions about the calculability and predictability of the future, representing a fundamental paradigm shift in calculating and defining credit risk.”110 Further research is required to explore this intersection in greater depth. A. What is SafeMeasures? SafeMeasures is a trademarked subscription cloud-based service developed by the non-profit, National Council on Crime and Delinquency (NCCD) and originally commissioned in 2011 by the Alameda County for the California Department of Social Services. In November of 2020, the company changed its name to Evident Change, stating: “Our new name honors what’s at the heart of our work—we Inform Systems to Transform Lives.”111 According to their marketing, as of March 2019, the estimated 106 Rachel M. Cohen, Why “Pay for Success” Financing Could Cost Taxpayers More Than They Bargained For, IN THESE TIMES (May 30, 2017), https://inthesetimes.com/article/pay-for-success-private-investment-education [https://perma.cc/BR6W-E48S]. 107 Martha Poon, From New Deal Institutions to Capital Markets: Commercial Consumer Risk Scores and the Making of Subprime Mortgage Finance, 34 ACCOUNTING, ORG. & SOC’Y 654 (2009). 108 David Irwin et al., The Financialization of Cloud Computing: Opportunities and Challenges, 26th International Conference on Computer Communication Networks (July 31 through August 3, 2017). 109 Natalia Besedovsky, Financialization as Calculative Practice: The Rise of Structured Finance and the Cultural and Calculative Transformation of Credit Rating Agencies, 16 SOCIO-ECON. REV. 61, 61–62 (2017). 110 Id. at 61. 111 WE ARE NOW EVIDENT CHANGE!, EVIDENT CHANGE, (Nov. 30, 2020) https://www.evidentchange.org/blog/we-are-now-evident-change [https://perma .cc/52GW-8UTW]. 2021] CALCULATING THE SOULS OF BLACK FOLK 107 costs for year 1 implementation begins at $100,000. The cost for annual hosting and consultation services is $400,000, with increases based on the number of data sources and organization structure. In April of 2017, ACS procured through sole source agreement, the SafeMeasures dashboard, agreeing to pay the company $1.4 million dollars. SafeMeasures claims it provides a “Better Value” because their one-price subscription gives access to the software for all agency staff and unlimited access to Evident Change’s analysis and support teams, who also conduct the training for agency staff. The outsourcing of civic decision making to a non- governmental organization without public comment is alarming in and of itself. By definition, the children whose lives are reflected in the data inputted into the SafeMeasures dashboard are already vulnerable and have limited access to justice through the courts. This is exacerbated when the government’s socio- technical system’s design is outsourced. Those who know all too well the prejudice of the frontline caseworker, juvenile detention center and family court employees might be incredulous at the suggestion that we ought to defend racist institutions from having their decision-making capacity eroded by the introduction of these obscure and obscuring computational infrastructures, and the massive quantity of surveillance data they rely on. There are limitations to procedural liberalism in holding government institutions accountable, but procedures and protections exist, providing some adjacency or access to accountability. In contrast, packaging inequitable decision making in code and software, and turning this process over to an outside entity not subject to these accountability measures, further separates the systems and people at the core of decision making from the processes intended to provide justice. The organization’s assessment of their own ability to affect harm is questionable. In 2014, NCCD released a report about the Los Angeles County Delinquency Prevention Pilot (DPP). DPP was a program of the LA Child Welfare agency using SafeMeasures to predict which children were most likely to become criminals. The pilot combined pre-existing county data with incoming social services data and continuously provided a screening assessment which generated alerts to the assigned caseworker when a child was predicted likely to commit a 108 COLUM. J. RACE & L. F. [Vol. 11:75 juvenile offense. The pilot concluded there was insufficient data to determine whether the intervention was successful in decreasing the rate of arrests among the experimental group but NCCD not only, “suggests that LA County’s strategy was on the right path, but also that more government agencies should consider testing similar programs all over the country.” VI. CONCLUSION “To the real question, How does it feel to be a problem? I answer seldom a word.” —The Souls of Black Folk, W.E.B. Du Bois112 Between me and the other world, the question of how it feels to be a problem no longer goes unasked or finds itself proffered with delicacy. This is not to say that it—or the color line—has lost relevance in the reckoning of post-racial progress or the fascist white supremacy laying on the other side of the coin. It’s to say modernity’s impulse to classify and quantify the socioeconomic and anti-Black consequences of racial capitalism, seeks to account for the suffering without a column in the ledger for tallying complicity. This impulse restricts the accounting of violence and poverty within the people who are affected, freezing them in an eternal snapshot as a data point. Sites of carceral enclosure, whether the prison, the residential treatment center or decentralized modes of containment in the digital poorhouse, assume the danger to be managed is within the indeterminacy of the underclass. This lie is both the tragedy and the salve for the situation in which we find ourselves. Out of the fairness, accountability, and transparency framework—it is accountability that provides the through-line from present struggles and policies that aim to mitigate harm and an imagining of the world otherwise. Authoritarianism is brutal and violent in visible and named ways, which is why there is resistance bred in even the harshest conditions. The invisibility of the digital poorhouse and these emergent infrastructures of control present unique challenges to those seeking to build the grassroots resistance that abolition requires. We must get it how we live or rely on the tools, people and institutions we have in order to collectively fortify ourselves 112 W. E. B. DU BOIS, THE SOULS OF BLACK FOLK 1 (100th Anniversary Ed., Routledge 2016) (1903). 2021] CALCULATING THE SOULS OF BLACK FOLK 109 against new forms of hegemonic control. However, “our Spiritual Strivings” and not the ambitions of reformism, must guide the tempo of our demands for accountability. We must be guided by a vision of a world that is “not readily available for viewing by those with eyes to see or available for hearing by those with ears to hear.”113 Otherwise, we will find ourselves building an apparatus that manages and expands the digital poorhouse in lieu of dismantling it. 113 Kevin Thomas Miles, “Place Post-Paradise: Poetic Epistemology in The Souls of Black Folk, 15 CR: NEW CENTENNIAL REV. 83 (2015). 110 COLUM. J. RACE & L. F. [Vol. 11:75