Artificial Administration in Action: Criminal Sentencing
Sentencing of criminal defendants requires an individualised analysis of the offender, the offence and the public interest, with a view to fixing an appropriate punishment. It is for a judge to fix the sentence, but she must do so based on a reasoned analysis of the relevant circumstances. It is thus an excellent example of the moral judgment model of administrative justice.
Sentencing discretion has long been a target of reformers in the United States. Discretion has been cabined, for instance, by the development of detailed federal sentencing guidelines. More recently, judges have come to rely on evidence-based risk assessment tools to aid in the exercise of their sentencing discretion. Sentencing discretion perhaps falls more naturally in the field of criminal justice, but it has clear affinities to administrative justice and, moreover, raises similar concerns to the use of artificial administration in parole decisions, which are more classically administrative in nature.
Detailed consideration was given to one of these sentencing tools – COMPAS – in the Wisconsin Supreme Court in State v Loomis:
COMPAS is a risk-need assessment tool designed by Northpointe, Inc. to provide decisional support for the Department of Corrections when making placement decisions, managing offenders, and planning treatment.The COMPAS risk assessment is based upon information gathered from the defendant’s criminal file and an interview with the defendant.
A COMPAS report consists of a risk assessment designed to predict recidivism and a separate needs assessment for identifying program needs in areas such as employment, housing and substance abuse.The risk assessment portion of COMPAS generates risk scores displayed in the form of a bar chart, with three bars that represent pretrial recidivism risk, general recidivism risk, and violent recidivism risk. Each bar indicates a defendant’s level of risk on a scale of one to ten.
As the PSI explains, risk scores are intended to predict the general likelihood that those with a similar history of offending are either less likely or more likely to commit another crime following release from custody. However, the COMPAS risk assessment does not predict the specific likelihood that an individual offender will reoffend. Instead, it provides a prediction based on a comparison of information about the individual to a similar data group.
The promise here is of more accurate decision-making, which is fairer to the individual concerned and members of the public, as well as greater equality in sentences applied to similarly situated individuals. It evokes a utopian vision. As the Supreme Court of Indiana had previously put it:
Such assessment instruments enable a sentencing judge to more effectively evaluate and weigh several express statutory sentencing considerations such as criminal history, the likelihood of affirmative response to probation or short term imprisonment, and the character and attitudes indicating that a defendant is unlikely to commit another crime.
From Loomis’s perspective, however, dystopia loomed large, in the form of a breach of his due process rights under the U.S. Constitution. He argued that assessing his sentencing by reference to the COMPAS tool failed to give him individualised treatment, because it relied on “group data”, not his individual circumstances. He also argued that he had not received adequate disclosure about the underlying basis for the tool’s calculations because the private company that developed the tool considers it to be “a proprietary instrument and a trade secret” and refuses to disclose “how the risk scores are determined or how the factors are weighed”.
The Supreme Court concluded that Loomis had not been deprived of any constitutional rights, but it purported to limit the scope for the use of the COMPAS tool. First, any sentencing judge using the tool is obliged to take into account:
…the following cautions regarding a COMPAS risk assessment’s accuracy: (1) the proprietary nature of COMPAS has been invoked to prevent disclosure of information relating to how factors are weighed or how risk scores are to be determined; (2) risk assessment compares defendants to a national sample, but no cross-validation study for a Wisconsin population has yet been completed; (3) some studies of COMPAS risk assessment scores have raised questions about whether they disproportionately classify minority offenders as having a higher risk of recidivism; and (4) risk assessment tools must be constantly monitored and re-normed for accuracy due to changing populations and subpopulations. Providing information to sentencing courts on the limitations and cautions attendant with the use of COMPAS risk assessments will enable courts to better assess the accuracy of the assessment and the appropriate weight to be given to the risk score.
Second, the COMPAS risk assessment cannot be the “determinative factor considered at sentencing”. Indeed, a sentencing judge “must explain the factors in addition to a COMPAS risk assessment that independently support the sentence imposed”. Roggensack CJ went further in her concurrence; she required that any sentencing judge “must evaluate on the record the strengths, weaknesses, and relevance to the individualized sentence being rendered of the evidence-based tool (or, more precisely, the research-based or data-based tool)”.
The point here is that over-reliance on tools such as COMPAS (which could, quite easily, become more sophisticated over time) replaces moral judgment with bureaucratic rationality. Pushed to its logical and technological limits, a tool such as COMPAS would cause the individualised assessment of an offender’s circumstances and the public interest to be replaced by a sentence calculated by reference to a generalised group of offenders. One way of reading the Supreme Court’s decision in Loomis is that it insists on moral judgment informed by bureaucratic rationality. There is thus a striking contrast with the Australian robo-debt scandal. Rather than the introduction of information technology replacing one model of administrative justice with another, here the Supreme Court sought to retain moral judgment as the operative model, enhanced by artificial administration but certainly not replaced by it. Perhaps this compromise position will not provide durable, especially because of the risk of automation bias; but for present purposes, it is significant that the Supreme Court sought compromise, not the triumph of either technological or legal norms.
Fitting artificial administration into existing structures is a significant challenge but one which presents significant possibilities. One major issue will be the contest between competing value systems, the norms of administrative justice (and administrative law) on one side and those of technology on the other. The potential for tension is evident, as the examples discussed in this paper highlight. Some lessons can be learned: the robo-debt debacle is a straightforward example of a failure to pay proper attention to the norms of administrative justice in governmental decision-making. By contrast, the Wisconsin Supreme Court attempted in Loomis to mould the bureaucratic rationality of artificial administration to the moral judgment of the sentencing process.
The answer to the question posed at the outset about whose norms (the technologists’ or the jurists’) will shape the future of public administration may well be inconclusive: artificial administration will not herald the triumph of technology, nor will it be harnessed to lawyerly norms, but it is more likely to be integrated into existing decision processes in a manner respectful of existing models of administrative justice.
 Renton and Brown’s Criminal Procedure, 6th. at 22-18.See also Criminal Justice Act 2003, s.142(1):
Any court dealing with an offender in respect of his offence must have regard to the following purposes of sentencing—(a) the punishment of offenders, (b) the reduction of crime (including its reduction by deterrence), (c) the reform and rehabilitation of offenders, (d) the protection of the public, and (e) the making of reparation by offenders to persons affected by their offences
 See generally Mistretta v. United States, 488 U.S. 361 (1989).
 See e.g. Ralph Serin and Renée Gobeil, “Analysis of the Use of the Structured Decision-making Framework in Three States” (January 2014).
 371 Wis.2d 235 (2016).
 371 Wis.2d 235, at pp. 245-246.
 Malenchik v. State, 928 N.E.2d 564, at p. 574 (2010).
 371 Wis.2d 235, at p. 265
 371 Wis.2d 235, at p. 258.
 371 Wis.2d 235, at p. 264.
 371 Wis.2d 235, at p. 265.
 371 Wis.2d 235, at p. 275.
 371 Wis.2d 235, at p. 287.
This content has been updated on April 4, 2019 at 14:38.