Flores v. Stanford

1

CARLOS FLORES, LAWRENCE BARTLEY, DEMETRIUS BENNETT, L'MANI DELFMA, EDGARDO LEBRON, ANTONIO ROMAN, DONTAE QUINONES and SHAROD LOGAN, on behalf of themselves and all others similarly situated, Plaintiffs, v. TINA M. STANFORD, as Chairwoman of the New York State Board of Parole; WALTER W. SMITH, As Commissioner of the New York State Board of Parole; SALLY THOMPSON, as Commissioner of the New York State Board of Parole; JOSEPH P. CRANGLE, as Commissioner of the New York State Board of Parole; ELLEN E. ALEXANDER, as Commissioner of the New York State Board of Parole; MARC COPPOLA, as Commissioner of the New York State Board of Parole; EDWARD SHAKEY, as Commissioner of the New York State Board of Parole; TANA AGOSTINI, as Commissioner of the New York State Board of Parole; CHARLES DAVIS, as Commissioner of the New York State Board of Parole; CAROL SHAPIRO, As Commissioner of the New York State Board of Parole; ERIK BERLINER, as Commissioner of the New York State Board of Parole; OTIS CRUSE, as Commissioner of the New York State Board of Parole; TYCEE DRAKE, as Commissioner of the New York State Board of Parole; and CARYNE DEMOSTHENES, as Commissioner of the New York State Board of Parole, Defendants.

No. 18 Civ. 02468 (VB)(JCM)

United States District Court, S.D. New York

September 28, 2021

OPINION & ORDER

JUDITH C. McCARTHY United States Magistrate Judge

Before the Court is an application from non-party Northpointe, Inc. ("Northpointe") seeking an order preventing disclosure of certain materials produced by Northpointe to Plaintiffs Carols Flores, Lawrence Bartley, Demetrius Bennett, L'Mani Delima, Edgardo Lebron, Antonio Roman, Dontae Quinones and Sharod Logan's (collectively "Plaintiffs") expert, Dr. Cynthia Rudin ("Dr Rudin"). (Docket No. 211). Plaintiffs opposed the request. (Docket No. 207). The Court heard oral argument on September 16, 2021. (Docket No. 219). For the reasons that follow, Northpointe's request is denied.

I. BACKGROUND

This dispute stems from the Court's previous order, dated February 12, 2021 (the "February 12, 2021 Order"), compelling Northpointe to produce to Plaintiffs certain proprietary information pertaining to Northpointe's Correctional Offender Management Profiling for Alternative Sanctions ("COMPAS") tool. (See Docket Nos. 181, 186, 209-3; February 12, 2021 Minute Entry).

As alleged in the Second Amended Complaint (the "SAC"), COMPAS is a commercial product sold by Northpointe containing "secret algorithms" that is used as a risk assessment tool by defendants Tina Stanford, as Chairwoman of the New York State Board of Parole (the "BOP"), and the individual BOP Commissioners (the "Commissioners") (collectively, the "Defendants"), when deciding whether to grant offenders sentenced to life in prison discretionary parole. (Docket No. 110 ¶¶ 10, 209, 211, 213, 215). The SAC further alleges that Defendants rely on COMPAS "without knowing how or whether COMPAS considers the diminished culpability of juveniles and the hallmark features of youth" because Northpointe "considers COMPAS a proprietary instrument and a trade secret." (Id. ¶¶ 213, 215). Moreover, COMPAS "sometimes treats youth as an aggravating factor" in predicting risk of felony violence. (Id. ¶¶ 211-12). For example, Plaintiff L'Mani Delima ("Mr. Delima"), is twenty-six years old, has been denied parole three times, and has a high "Risk of Felony Violence" COMPAS score even though he was convicted when he was only thirteen. (See Id. ¶¶ 47, 49-53, 212). According to Plaintiffs, Defendants' reliance on this algorithm of which they lack knowledge or understanding deprives juvenile offenders who received life sentences of the individualized parole assessments to which they are entitled under the Eighth and Fourteenth Amendments. (See Id. ¶¶ 2-10, 136-46, 222-35).

The Court's February 12, 2021 Order required Northpointe to produce to Plaintiffs (1) the normative dataset used to create and normalize COMPAS (the "Norm Group Data"); and (2) the regression models for two COMPAS "scales": (a) the General Recidivism Risk Scale, and (b) the Violent Recidivism Risk Scale (the "Regression Models" and collectively, the "Compelled Materials"). (See February 12, 2021 Minute Entry; Docket No. 209-3; see also Docket Nos. 181, 186). The Norm Group Data is a repository of offender information from several jurisdictions that Northpointe uses to generate the Regression Models, which are sets of inputs used to predict the likelihood of new offenses and new violent offenses after an offender's COMPAS assessment date. (Docket Nos. 181 at 2; 209-3 at 7:9-8:9; 219 at 20:6-12). Northpointe also uses the Norm Group Data to translate recidivism risk scores into data presented to individual Defendants before parole hearings. (Docket No. 209-3 at 7:22-8:20; see also Docket No. 181 at 2). To address Northpointe's concerns that the Compelled Materials constitute trade secrets and/or other proprietary information, (see Docket No. 186), the Court ordered that the Compelled Materials be produced pursuant to a protective order and be designated "attorney's and expert's eyes only." (See February 12, 2021 Minute Entry; Docket No. 209-3 at 37:5-15).

Northpointe and Plaintiffs entered into a Second Supplemental Stipulation of Confidentiality and Proposed Protective Order regarding the Compelled Materials on February 26, 2021 (the "Second Protective Order"). (Docket No. 198). The Second Protective Order requires that the Compelled Materials be designated attorneys' eyes only and "Highly Confidential" as described in the parties' first Stipulation of Confidentiality and Protective Order (the "First Protective Order") and the supplement thereto concerning Northpointe (the "Supplemental Protective Order") (collectively, the "Protective Orders"). (Docket No. 198 ¶¶ 2-3; see also Docket Nos. 151, 154). Under these earlier Protective Orders, materials designated as such cannot be "provide[d, ] show[n] [or] . . . discuss[ed] . . . with Plaintiffs" without Northpointe's written consent or a court order. (Docket No. 154 ¶ 4). They "shall not be released or disclosed in any manner to any other person," with some limited exceptions, and "shall be used by the receiving party solely for the purposes of [the instant] action and solely to the extent necessary for [its] litigation . . ., including any appeals thereof." (Docket No. 151 ¶¶ 6, 21). At the conclusion of this action, all such materials and any copies thereof must either be destroyed or returned to adverse counsel. (Id. ¶ 22). The Second Protective Order further provides that the Compelled Materials "may not be shared with, shown to, or discussed with a party or a party's representative;" "can be reviewed by" only eight specific attorneys for Plaintiffs and Defendants; and cannot be disclosed to any expert unless Northpointe provides "express written consent," or any objections to such disclosure are overruled. (Docket No. 198 ¶ 3). Furthermore, neither the Compelled Materials nor their contents can be filed publicly. (Id. ¶¶ 4-5).

Notwithstanding these protections and Plaintiffs' efforts to meet and confer, (see Docket No. 207 at 3), Northpointe will not consent to allow Dr. Rudin to access the Compelled Materials. (See generally Docket No. 211). According to Northpointe, these documents contain Northpointe's "most proprietary information" and "permitting Dr. Rudin access to [it] . . . would threaten [Northpointe's] very existence." (Id. at 1). This is because Dr. Rudin, a professor of computer science, electrical and computer engineering, and statistical science at Duke University, is an outspoken critic of Northpointe and the COMPAS tool. (See Docket No. 211 at 1-2; see also Docket Nos. 207 at 3; 211-1). Indeed, Dr. Rudin has published several articles and given lectures questioning the government's reliance on COMPAS to evaluate recidivism risk, when neither the public nor the government have access to the data or formulas on which COMPAS is based.^[1]According to Dr. Rudin, this lack of transparency is particularly unfair when it is possible to use publicly available algorithms to predict recidivism risk with the same accuracy as COMPAS. In making this argument, Dr. Rudin has indicated that she and her coauthors have attempted, without success, to uncover the COMPAS algorithm(s). See, e.g., NC State Research, DSI RED Talk with Cynthia Rudin Secrecy, Criminal Justice and Variable Importance, YouTube (Nov. 6, 2019) https://www.voutube.com/watch?v=QWcH1U-5Ftw, at 27:03 to 27:45 ("[W]e've been working on this analysis that tries to uncover some of COMPAS' secret sauce"; "we've been trying to figure out what's actually inside COMPAS"; "[I]f COMPAS is gonna sell people this model and say that they're better than my interpretable machine learning models I want to know what the heck's in there, right?"; "Our goal is to identify bounds for how much COMPAS relies on different covariate subsets . . . ."). Dr. Rudin has also opined that Northpointe should be "put. . . out of business" and "there should not be a business model" for the COMPAS tool. Nat'l Academy of Sciences, supra n.l at 1:48-1:52.

Apart from the above projects, Dr. Rudin focuses on the socially responsible use of machine learning to solve real-world problems in a variety of contexts, such as healthcare, energy use and criminology. (See Docket Nos. 207 at 2; 209-1 at 2). For example, she has used her expertise to help Con Edison establish an underground electrical distribution network in New York City; create optimal scoring systems to predict seizures in ICU patients; and develop a code for detecting crime in cities, which was adapted by the New York City Police Department ("NYPD") in 2016. (Docket No. 209-1 at 2). Dr. Rudin also collaborated with the Cambridge Police Department...