In Algorithms, Discrimination and the Law, Professor Michael Selmi performs an excellent analysis of the many controversial issues related to an employer’s use of algorithms in making employment-related decisions.
The use of algorithms in the workplace has garnered substantial academic discussion in recent years, as this type of technology has become more readily accessible to employers. The widespread use and reliance on technology to formulate employment-related decisions has created a host of workplace-related issues. At the forefront of these concerns is that the use of algorithms will discriminate against minority workers and applicants. These concerns are well-founded, and additional empirical work is needed to explore the parameters of this form of discrimination and to examine this important and emerging topic more broadly.
In his paper, Professor Selmi weighs in on the topic of algorithm use in workplace decision-making. His paper highlights the ways in which algorithm use could have a positive impact in the employment setting by reducing the role of human involvement in the process, which can be tinged with express or implied bias. Professor Selmi finds the failure of legal scholars to “eager[ly] embrace” these types of algorithms surprising, “[g]iven all that we know about human decision making, particularly its biased nature.” (Pp. 614-15.) In taking this critical approach to the existing scholarship, Professor Selmi addresses the two primary criticisms that are typically advanced about algorithms use: the potential that algorithms may themselves be discriminatory and the inability of the law to eradicate any discrimination associated with algorithms.
Professor Selmi approaches the first issue by asking the broad question of whether it is possible to construct these algorithms in a way where the results are less discriminatory than the traditional use of human review (which involves well-known biases). Examining and comparing traditional decision-making to a more technological-based approach, Professor Selmi concludes that “between an algorithm and a human, the smart money [on which will avoid discrimination] is likely on the algorithm.” (P. 617.)
Professor Selmi’s primary takeaway is that human decision-making has resulted in both express and implicit bias over the years and discrimination continues to pervade the workforce. Thus, while algorithms may still produce discriminatory results if they are flawed in their construction or use flawed data as inputs, there is also the largely unacknowledged possibility that these programs offer the potential to substantially curb discrimination if properly constructed. Essentially, there is no doubt that human decision-making results in discrimination. The use of algorithms, then, while not perfect, offers a potential opportunity to curb existing bias.
Professor Selmi further addresses the concern over the inability of the law to properly address discrimination where it occurs in this setting. He examines the frequent critique of disparate impact law, which addresses facially neutral policies or practices that have a discriminatory effect. He notes the existing academic concerns that the “inscrutable nature of algorithms will make any legal challenge ineffective.” (P. 618.)
Proessor Selmi responds to these concerns by distinguishing between two specific types of algorithms. First, he notes that where this technology is based on algorithms that are easily understood, there should be no new concerns about the application of disparate impact theory which has addressed these types of policies/practices for years.
The more challenging area is the second type of algorithm based on “black-box” formulations. Professor Selmi challenges the existing view that such black-box algorithms result in more difficult discrimination claims. Indeed, Professor Selmi notes that under disparate impact law, employers are required to establish that the algorithm is job-related and consistent with business necessity after a disparate impact is established. Thus, this theory is well-grounded in its ability to require employers to establish the validity of the algorithm. Additionally, Professor Selmi looks at the final step of disparate impact law; the ability of a plaintiff to articulate a less discriminatory approach. He surmises that there will be easily identifiable alternatives to many black-box algorithms that have a discriminatory impact.
Professor Selmi’s work is exceptionally well-researched and written, and the topic could not be more important as we see so many companies moving toward this type of decision-making model. Disparate impact law itself is an underexplored area of the law, and this piece further pushes the academic boundaries of the unintentional discrimination provisions. Professor Selmi’s willingness to take on directly the concern in this area is helpful in considering how employers should approach this topic. His novel and more nuanced view that “the concerns about the law’s impotency seem overstated,” and that workplace algorithms could be created in a way that result in “less discriminat[ion] than human actors” is critical in balancing the ongoing debate in this area. (Pp. 618-19.)
This piece is extraordinary in reframing this debate. It reimagines the role of algorithms in the workplace and the potential use of this technology to lessen any discriminatory effect. The essay pushes the boundaries of disparate law, applying it to the context of modern technology. There is far too little work in this area, and Professor Selmi’s piece is critical in opening an important dialogue in this important field.
