Algorithm vs. Algorithm

Government entities have begun to adopt machine learning algorithms for adjudicatory and enforcement actions to enhance public administration. Critics of these applications have suggested that machine learning algorithms would produce entirely new problems—that is, that these algorithms are distinctively complex, inscrutable, difficult to scrub of bias, and lacking in accountability. But a realistic assessment of these digital algorithms must acknowledge that government already rests on “algorithms” of arguably greater complexity and potential for abuse: humans. Currently, government functions depend on human decision-making, which can be highly complex, inscrutable, prone to bias, and unaccountable. On an individual level, humans are subject to a multitude of personal interests, physical limitations, and cognitive biases: memory, fatigue, hindsight bias, confirmation bias, and racial and gender biases, among others. On an organizational level, humans are prone to groupthink and social loafing, along with other group phenomena. In many cases, human decision-making may be even more prone to suboptimal decisions than their machine learning counterparts. Digital algorithms offer the promise of increased accuracy, consistency, speed, and capacity. These characteristics are particularly suited to the needs of public administration, given the desire for uniformity in state policies, the high volume of data, and the constraints in time, finances, and personnel. Government agencies will increasingly confront a choice between the human status quo and a significantly more promising machine learning application. This paper presents a framework for public officials in deciding when to deploy automated decision tools. It considers the likelihood that a new use case for digital algorithms would satisfy the preconditions for successful deployment. It also emphasizes validation that a machine learning system would indeed make an improvement over the status quo in terms of the principal objectives, legal issues, or other new types of problems. Government officials must ensure adequate planning, careful procurement of any private contractor services, and appropriate opportunities for public participation in design, development, and ongoing oversight. Ultimately, when evaluating the use of machine learning in governmental settings, any anticipated shortcomings of machine learning must be placed in proper perspective—with human algorithms and digital algorithms side-by-side.