Sunstein2019CorrectingBiases

Cass R. Sunstein, "Algorithms, Correcting Biases"

Bibliographic info

Sunstein, Cass R., Algorithms, Correcting Biases (December 12, 2018). Forthcoming, Social Research, Available at SSRN: https://ssrn.com/abstract=3300171

Commentary

The author explores the question of whether algorithms are biased, specifically focusing on the use of algorithms in the context of public policy and law. Two main claims are made. The first claim is that algorithms can help overcome harmful cognitive biases that often affect human decision-making in social prediction problems. Algorithms act as a form of statistical prediction and can outperform human judgments by avoiding biases. The second claim is that algorithms can be designed to avoid discrimination, particularly racial discrimination, while also presenting difficult trade-offs in balancing different social values.

A research conducted by Jon Kleinberg and his colleagues is used to illustrate the effectiveness of algorithms in the context of bail decisions. The algorithm they developed outperformed human judges in making bail decisions, leading to potentially fewer crimes and more accurate judgments. A potential weakness of the research could be that there is a focus on discrimination in the context of disparate impact and racial balance. It could potentially overlook other forms of discrimination that algorithms might perpetuate, such as gender, age, or socioeconomic biases.

Excerpts & Key Quotes

Narrow view on judges' decisions

"The principal research on which I will focus comes from Jon Kleinberg, Himabinku Lakkaraju, Jure Leskovec, Jens Lutwig, and Sendhil Mullainathan, who explore judges’ decision whether to release criminal defendants pending trial."

Comment:

The research by Jon Kleinberg and his colleagues is valuable in shedding light on the potential advantages of using algorithms in bail decisions. However, it is essential to recognize that focusing solely on one study may not provide a comprehensive view of the broader implications of algorithmic decision-making in the criminal justice system. Further research and replication studies from diverse contexts are needed to draw robust conclusions about the efficacy and fairness of algorithmic solutions in this domain.

Limitations of human intuition

"The use of algorithms is often motivated by an appreciation of the limitations of human intuition. In the private and public sectors, people are often asked to make predictions under conditions of uncertainty, and their intuitions can lead them astray."

Comment:

While recognizing the limitations of human intuition, it is essential to understand that algorithms do not offer a one-size-fits-all solution to decision-making challenges. Relying too heavily on algorithms, without human oversight, may result in "automation bias," where decision-makers unquestioningly accept algorithmic outputs without considering their accuracy or the unique contextual factors involved. To foster responsible and ethical decision-making, a balanced approach that combines human expertise with algorithmic assistance is vital.

Tradeoffs

"The most important point here may not involve the particular numbers, but instead the clarity of the tradeoffs. The algorithm would permit any number of choices with respect to the racial composition of the population of defendants denied bail. It would also make explicit the consequences of those choices for the crime rate."

Comment:

The author talks about how algorithms can show us the choices we have regarding the racial makeup of people who are denied bail and how it affects crime rates. It suggests that algorithms can help us understand the tradeoffs involved in making these decisions.

However, there is a potential problem with this idea. Algorithms might unintentionally continue or worsen existing racial biases because they rely on data from the past, which could be biased itself. So, even with transparency, the algorithm's outcomes may not be fair. Deciding on the racial makeup of those denied bail should not only be based on crime rates. It is important to consider other factors and address underlying issues to avoid perpetuating racial inequality.