Algorithmic risk assessments, offered as a means to improve decision-making by standardizing the prediction of an individual’s future behavior, present myriad challenges in criminal administration. The tools are inscrutable. The tools are discriminatory. What is to be done? One popular solution is public participation in the design and adoption of actuarial risk assessments. Though not a panacea, jurisdictions across the country – from New York City to Sacramento – are passing or considering laws that require public oversight in the adoption of actuarial risk assessments.
In her article, The Democratizing Potential of Algorithms?, forthcoming in the Connecticut Law Review, Ngozi Okidegbe challenges the assumption that these kinds of initiatives can resolve one of the most deep-seated critiques of pretrial algorithms – their racialized effect on marginalized people disproportionately subject to the carceral state. To the contrary, she argues that such efforts threaten to exacerbate the problem. Because her article questions the compatibility of the algorithmic project with racial justice in a novel way, it is a must read for scholars interested in criminal legal reforms.
While actuarial risk assessments are proliferating throughout criminal administration, particular enthusiasm and momentum exists to adopt these tools in the pretrial bail context. Here, the tools tend to predict the likelihood of an individual failing to appear for court or engaging in crime in the future based on statistical analyses of large datasets. This information ostensibly indicates for judges which defendants should be released before trial and which ones should be subject to more intense forms of state surveillance, whether in jail or not. As states and localities confront the shortcomings of cash bail practices, law and policymakers are shifting toward “pretrial algorithmic governance” by institutionalizing algorithmic risk assessments as a key part of the process.
Yet these tools have been plagued with critiques of their racialized effects, a point that Okidegbe unpacks as layered. She identifies three levels of exclusion specific to marginalized black and brown communities that occur with the shift toward pretrial algorithmic governance. First, marginalized communities most impacted by pretrial detention are largely excluded from the algorithmic construction process. This can lead to the creation of particularly harmful algorithms for that community. Second, pretrial algorithmic governance entrenches marginalized communities’ exclusion from pretrial governance just as effective bottom-up strategies to combat the harms of cash bail detention are spreading. The expansion of actuarial risk assessments in lieu of that flawed practice repositions the most impacted marginalized communities as outside the scope of political influence over pretrial decision-making. Finally, pretrial algorithmic governance perpetuates the adverse effects of the criminal legal system on the ability of system-involved people to realize full participation in a democratic society. By reproducing exclusion in governance, pretrial algorithms threaten to exacerbate the political, social, and economic costs of unnecessary carceral supervision experienced by the most marginalized.
In light of the particular harm pretrial algorithmic governance poses for marginalized communities, Okidegbe considers popular public participation “solutions” to this governance problem. These include focus groups, public hearings, and appointed citizen boards, all of which could provide “ex-post input” but do not retain power over first order questions like whether to adopt a tool at all. Such interventions, she argues, do not resolve the particular harm to marginalized populations because they are not “power-shifting” approaches by design. Indeed, those in power can use such symbolic gestures to make pretrial algorithmic governance appear legitimate without redistributing power to marginalized communities, which creates another kind of harm.
Yet power-shifting approaches are feasible, and here is where Okidegbe’s work shines. She argues that localities could create bail reform commissions that intentionally pursue communal involvement. These commissions would need power to make key decisions in the adoption, implementation, and oversight of pretrial algorithms. The commissions would have to be populated with representatives from the most impacted communities, meaning Black individuals with some direct connection to the experience of incarceration or crime. Moreover, the commission would have to be designed to prevent power differentials between technocrats and “marginalized community commissioners.” If these design features were incorporated into pretrial algorithmic governance, then perhaps the algorithms would improve by “mitigating negative externalities associated with the imposition of incarceration on low income communities.” Regardless, it would transform pretrial governance into a space of deep democratization. This would redress the exclusion and political estrangement facilitated by the expansion of the carceral state.
Okidegbe anticipates great resistance to her proposal, so she spends a fair amount of time engaging with various objections. Critics of the democratizing criminal law scholarship may reject her proposal because such layperson involvement reflects “penal populism” that actuarial risk assessments are imagined to combat. Algorithmic reformers may resist these governance changes because it could produce algorithms not tightly tethered to their technical conception of accuracy. Judges may resist using tools built by community members. Marginalized communities may reject her proposal because it contradicts tenets of individual sentencing.
The resistance really makes the point of the paper. After all, if so many people would reject such deep democratization, then algorithmic governance is not compatible with a racial justice agenda. Now, many would say they never committed to a racial justice agenda by embracing actuarial risk assessments. That may well be true. But if that is the case, we must release ourselves of the assumption that this way has to be the way to address racialized mass incarceration. There are other ways to reduce incarceration, many of which those most affected by the carceral state have already begun to imagine. The real question Okidegbe’s contribution raises, then, does not concern whether algorithms can be democratized. It’s why we as a society are not willing to embrace those other possibilities.