Designing Legal Systems for an Algorithm Society

We are rapidly moving from the age of the Internet to the Algorithmic Society, and soon we will look back on the digital age as the precursor to the Algorithmic Society. What do I mean by the Algorithmic Society? I mean a society organized around social and economic decision-making by algorithms, robots, and AI agents, who not only make the decisions but also, in some cases, carry them out. (Balkin 2017).

Law can be looked at in many ways, but in every incarnation, information is a central component. As one lawyer recently wrote, “from the moment we lawyers enter our offices, until we turn off the lights at night, we deal with information.” Information is the fundamental building block that is present and is the focus of attention at almost every stage of the legal process. Legal doctrine, for example, is information that is stored, organized, processed, and transmitted. Legal judgments are actions that involve obtaining information, evaluating it, storing it, and communicating it. Lawyers have expertise in and have control over a body of legal information. … Indeed, one way of understanding the legal process is to view information as being at its core and to see much of the work of participants as involving communication. In this process, information is always moving—from client to lawyer, from lawyer to jury, from judge to the public, from the public to the government, and so on. (Katsh 1995).

The embodiment of law in printed form did not simply replace law in [handwritten] form. Rather, it replaced a system of dispute resolution that had often involved written law and oral tradition working together. Printed law, therefore, emphasized decision making according to rules more than some previous systems because attention was focused on rules in a way that had not occurred earlier. In this respect, it is important to recognize that print also brought about a narrowing of the judge’s focus. Judges in earlier periods had been able at times to escape rule-oriented decisions because their attention was not bound to a printed or even a written text. They had options that were not available to later generation of judges whose decision had to be made “by the book.” (Katsh 1989)

Government decision-making is increasingly automated. Cities use machine-learning algorithms to track gunshots, determine where to send police on patrol, and fire ineffective teachers. State agencies use algorithms to predict criminal behavior, interpret DNA evidence, and allocate Medicaid benefits. Courts decide, using “decision-support” tools, whether a suspect poses a risk, eligibility for pretrial release, and how harsh a sentence to impose. The federal government uses algorithms to put individuals on immigrant and terrorist watchlists, make policy decisions about whether and how to change Social Security, and catch tax evaders. … But increasing automation may also make government less participatory and open to public oversight and input. (Bloch-Wehba 2020) (citations omitted).

Beagle is an A.I. tool for contract review that is primarily targeted at non-lawyers. Beagle is designed for users who need to review and manage contracts, but lack the expertise to do it themselves or the money to hire an attorney. First, users upload their contracts to the platform. Then, the natural language processing system identifies key clauses for review. This is done by identifying which clauses are used most often (for the type of contract at hand) and analyzing how this contract deviates from the norm. It also has a built-in communication system where users can interact with each other and discuss their documents. In addition to the system’s ability to learn from the individuals who use the tool, the system is able to learn the personal preferences of different users, and incorporate those preferences in future documents. (Semmler and Rose 2017)

First, the new algorithmic enforcement tools hold important implications for the accountability of agency enforcement activities. It remains unclear whether the new tools will degrade or enhance legal and political accountability relative to the status quo. On the one hand, the technical opacity and “black box” nature of the more sophisticated AI-based tools may erode overall accountability by rendering agency enforcement decisions even more inscrutable than the human judgments of dispersed agency enforcement staff. But the opposite might also prove true: formalizing and making explicit agency priorities could render an agency’s enforcement decision making relatively more tractable compared to pools of agency enforcement staff. (Engstrom and Ho 2020)