MA SJC Ruling on Bail Instructive Re: Algorithms and Criminal Justice

Cyberlaw Clinic » Blog 2017-09-05

One track of the Berkman Klein Center’s work on artificial intelligence ethics and governance concerns the use of algorithms, machine learning, and related technologies in ways that impact social and criminal justice. Among other things, this research examines technologies employed by courts in their disposition of criminal cases. Increasingly, judicial determinations are informed by software that helps judges perform “risk assessments” of defendants or otherwise process and weigh factors relevant to decisions about sentencing, parole, and the like. The Center (along with collaborators at the MIT Media Lab) is undertaking a number of efforts to evaluate ways in which these kinds of technologies might mitigate or exacerbate bias. The initiative has both technical and legal components, and a significant amount of our work to date has involved technologists and lawyers working together to correlate technical concepts with legal standards (and vice-versa). In the context of these efforts, the Massachusetts Supreme Judicial Court’s recent ruling in Brangan v. Commonwealth — which has nothing to do with algorithms but concerns, broadly, the process of making bail determinations in Massachusetts — is of significant interest.

In assessing the use of technical tools in the context of bail, it is important to go back to first principles and understand the purposes behind bail and the forces that animate bail decisions in individual cases. In making a bail determination, a court is faced with a decision whether and on what terms (monetary and otherwise) to release a criminal defendant pending full adjudication of that defendant’s rights. Such decisions are complex and depend upon on analysis of factors that are often in direct tension:

  • To state the obvious, incarceration has significant adverse impacts on those who are incarcerated. Thus, any decision to deprive someone of her liberty (either prior to trial or following a finding of guilt) must take into consideration the defendant’s right to due process and the extraordinary damage that a jail term can have on one’s livelihood, family relations, and the like. At the bail stage, that decision must recognize the presumption that criminal defendants are innocent unless and until the state proves otherwise.
  • At the same time, lenient standards for pre-trial release might increase the likelihood that a criminal defendant will fail to return to court for a full adjudication of underlying charges. In some cases, releasing a criminal defendant might put victims or others members of the community at risk.

Against this backdrop, the Brangan case deals with a situation in which a defendant is “unable to give the necessary security for his appearance at trial because of his indigence” — a situation in which “the purpose of bail is frustrated.” The opinion (issued August 25, 2017 and authored by Justice Geraldine S. Hines) concerns a criminal defendant, Jahmal Brangan, held in Hampden County, Massachusetts jail since January 17, 2014 following his arrest for armed robbery. During the intervening three-and-a-half years, Mr. Brangan has been held subject to bail set at amounts ranging from $20,000 to $50,000 in cash (or $200,000 to $500,000 in surety). Mr. Brangan was unable to pay any of these amounts, and he thus remains incarcerated despite never having received a full and final adjudication of the criminal charges against him.1 On appeal to the Massachusetts Supreme Judicial Court, Mr. Brangan argued that, in setting bail, the lower court had “violated his right to due process because the judge failed to give adequate consideration to his financial resources, and set bail in an amount so far beyond his financial means that it resulted in his long-term detention pending resolution of his case.”

In its opinion, the SJC notes that the relevant statute — Massachusetts General Laws c. 276, § 57 — does not expressly identify ability to pay as among the factors a judge must consider when setting bail. The statute indicates that one imposing bail in the Commonwealth “shall take into consideration” the following factors (links added):

the nature and circumstances of the offense charged, the potential penalty the person faces, the person’s family ties, employment record and history of mental illness, the person’s reputation, the risk that the person will obstruct or attempt to obstruct justice or threaten, injure or intimidate or attempt to threaten, injure or intimidate a prospective witness or juror, the person’s record of convictions, if any, any illegal drug distribution or present drug dependency, whether the person is on bail pending adjudication of a prior charge, whether the acts alleged involve abuse, as defined in said section 1 of said chapter 209A, a violation of a temporary or permanent order issued pursuant to said sections 18 or 34B of said chapter 208, said section 32 of said chapter 209, said sections 34 or 5 of said chapter 209A or said sections 15 or 20 of said chapter 209C, whether the person has any history of issuance of such orders pursuant to the aforesaid sections, whether the person is on probation, parole or other release pending completion of sentence for any conviction and whether the person is on release pending sentence or appeal for any conviction.

Nevertheless, the Court identifies common law support for the proposition that “[a] Superior Court judge . . . must still consider a defendant’s financial resources when setting bail.” The Court specifically points to the SJC’s prior decisions in Commonwealth v. Torres, 441 Mass. 499, 504 (2004) and Querubin v. Commonwealth, 440 Mass. 108, 115 n. 6 (2003), which included “financial resources” as among the factors a judge should consider when conducting a bail hearing:

the factors that a judge is to consider when conducting a bail hearing are “(1) the nature and circumstances of the offense charged, (2) the accused’s family ties, (3) his financial resources, (4) his length of residence in the community, (5) his character and mental condition, (6) his record of convictions and appearances at court proceedings or of any previous flight to avoid prosecution or (7) any failure to appear at any court proceedings.”

Beyond statutory language and common law, the SJC looks to the federal and state constitutions and notes that both the Eighth Amendment to the United States Constitution and Article 26 of Massachusetts Declaration of Rights (i.e., the state constitution in the Commonwealth) expressly prohibit excessive bail. Such prohibition also finds support in the broader constitutional principles of due process and equal protection. Avoiding imposition of excessive bail and complying with constitutional mandates regarding due process and equal protection necessarily requires individualized, case-by-case determinations that include consideration of defendants’ ability to pay

The Court does not go so far as to hold that imposition of unaffordable bail constitutes a due process violation per se. On the contrary, the Brangan opinion serves to establish a framework for bail-setting without pre-determining outcomes in any particular case:

Although the judge must take a defendant’s financial resources into account in setting bail, that is only one of the factors to be considered, and it should not override all the others. Bail that is beyond a defendant’s reach is not prohibited. Where, based on the judge’s consideration of all the relevant circumstances, neither alternative nonfinancial conditions nor an amount the defendant can afford will adequately assure his appearance for trial, it is permissible to set bail at a higher amount, but no higher than necessary to ensure the defendant’s appearance.

Indeed, the Court notes that — in some cases — it may be permissible to detain a defendant prior to trial without bail as long as fundamental requirements of both substantive and procedural process have been met in reaching that result.

Applying this legal framework to the facts at hand, the Court expresses three sets of concerns in support of its decision that a new bail determination for Mr. Brangan is required:

  • First, the Court notes that a defendant’s “dangerousness” is separate from whether or not he poses a “flight risk.” Under the applicable statute, the former may not be taken into consideration in assessing the amount of bail and may only be considered in the context of other conditions of pretrial release. The SJC finds that arguments presented in the Commonwealth’s briefs below conflated dangerousness and risk of flight and holds that the lower court’s bail determination did not adequately disentangle the two.
  • Second, and perhaps most importantly, the Court holds (footnotes and citations omitted):

[W]here, based on a defendant’s credible representations and any other evidence before the judge, it appears that the defendant lacks the financial resources to post the amount of bail set by the judge, such that it will likely result in the defendant’s long-term pretrial detention, the judge must provide findings of fact and a statement of reasons for the bail decision, either in writing or orally on the record. The statement must confirm the judge’s consideration of the defendant’s financial resources, explain how the bail amount was calculated, and state why, notwithstanding the fact that the bail amount will likely result in the defendant’s detention, the defendant’s risk of flight is so great that no alternative, less restrictive financial or nonfinancial conditions will suffice to assure his or her presence at future court proceedings.

Applying this standard, the Court finds that the bail order in question was deficient insofar as it did not evince consideration of Mr. Brangan’s financial resources and offered no explanation of “how the judge calculated the bail amount, how Brangan’s criminal history demonstrated that he posed a serious flight risk, or why that risk was so great that it necessitated a bail amount beyond his means.” (The Court also notes that the order failed to explain why the judge “rejected Brangan’s alternative proposal that he be released on $5,000 cash with the condition that he wear a GPS tracking bracelet.”)

  • Third, the Court holds that the long delay in bringing Mr. Brangan to trial must be considered in determining whether continuing pre-trial detention is justified.

The Brangan case is important on its merits and serves as a roadmap for judges, prosecutors, and defendants. For those approaching bail from a technical or design perspective, the case may be of interest for a slightly different set of reasons:

  • Bail may appear to present a paradigm for the type of straightforward judicial decisionmaking framework in which algorithms can play a role in processing information and balancing competing interests. But, the SJC’s opinion in Brangan underscores the complexities of bail and multivariate nature of a bail determination in any given case.
  • For those interested in “interpretability” in the context of algorithms,2 the opinion highlights the need for decisionmaking models that are transparent, trustworthy, replicable, and explainable. It also might serve to inform development of audit mechanisms for risk assessment tools in the context of bail, offering guidance regarding how judges must explain their decisionmaking processes as we seek to ensure the ability to review algorithmic decisions.
  • The decision underscores the need to draw from multiple data sources in order to arrive at reasonable and supportable bail determinations. Decisions may be informed by general principles (e.g., about the circumstances in which risk of flight is likely). But, individualized determinations based on specific facts about specific defendants’ financial circumstances are necessary to satisfy constitutional requirements.
  • On that note, if we are going to allow algorithms to play a role in certain aspects of the bail determination process, it is vitally important that we understand and make clear: (a) which questions those algorithms are helping to answer; (b) on what data sets those algorithms draw in reaching their conclusions; and (c) at what stages of a given legal analysis the outcomes prescribed by those algorithms may be relevant. Brangan‘s distinction between dangerousness and flight risk in the context of bail determinations made under 276 M.G.L. § 57 is instructive in this regard. The case makes clear that a bail determination might involve separate evaluations of two distinct factors: (i) whether a defendant, if released on bail, will pose a danger to a community, generally (or to particular past or potential future victims); and (ii) whether a defendant, if released on bail, is likely to leave the jurisdiction and/or otherwise fail to return for future court appearances. The case also makes clear that an analysis of each factor may be relevant at some stages of a monetary bail assessment and may be forbidden at other stages. If an algorithmic tool is used in making a legal determination (such as a decision about the conditions for pre-trial release), it is vital that a judge understand what that tool is assessing and in what ways that assessment may be relevant (or, conversely, in what circumstances consideration of that assessment may be prohibited) during the various stages of that legal determination.
  • In holding that the lower court erred by failing to consider Mr. Brangan’s alternative proposal – regarding a lower bail amount plus GPS tracking – the SJC demands not just an absolute assessment of whether to permit bail and at what amount but a relative assessment of the merits of alternative proposals and the ways in which those alternative proposals might, in different ways, help achieve the underlying goals of bail. This need to consider multiple options may inform the ways in which algorithmic risk assessment tools might be used in making bail or other similar determinations.
  • Finally, the Brangan decision highlights the importance of constitutional considerations in guiding bail determinations and defining both the universe of acceptable outcomes and the universe acceptable processes one might employ to achieve those outcomes in any given case. This is true regardless of the statutory architecture in place in a particular state.  Those designing tools to support judges must do so in ways that go beyond merely comporting with state statutes and must ensure such tools operate cognizant of and consistent with overarching considerations about due process and equal protection under law.

1 Mr. Brangan was tried, but the trial judge declared a mistrial due to improper conduct by the prosecution. Various appeals and other procedural consequences followed, but a court-ordered retrial had yet to commence by the time the SJC reached its decision.

See, e.g.Finale Doshi-Velez, Been Kim, “Towards A Rigorous Science of Interpretable Machine Learning, v. 2” arXiv:1702.08608v2 (last revised March 2, 2017)Zachary C. Lipton, “The Mythos of Model Interpretability, v. 3” arXiv:1606.03490v3 (last revised March 6, 2017)Jesse Johnson, “Goals of Interpretability,” The Shape of Data (November 17, 2016).

Image, “John Adams Courthouse – Suffolk County Courthouse – Boston, MA – DSC04718.JPG,” courtesy of Wikimedia Commons, user Daderot, made available under Creative Commons CC0 1.0 Universal Public Domain Dedication.