Algorithms as Legal Entities

September 22, 2017

Anything you can do, they can do better. The past decade has marked a growing concern over the displacement of white collar jobs by increasingly sophisticated software. Some believe that software will only ever supplement white collar work. Others predict a change as dramatic as the industrial revolution. The validation of one side over the other may depend, in some part, on future developments in contract law.
The law already treats non-human entities as “persons” for limited purposes. For example, corporations do not physically exist. They do not breathe, eat, or think. While the rest of us put our pants on one leg at a time, corporations don’t even wear pants. As far is the law is concerned, pants are optional. Corporations can enter into contracts as non-natural entities, and those contracts bind corporations just as they would bind natural persons. Parties in these contracts breathe easy with the knowledge that even if the non-human entity fails to perform, one or more natural humans will remain at least partially liable for damages. If the great Wizard of Oz welshes on a deal, there will always be someone behind the corporate curtain accountable for pulling the levers.
But what if the machine pulls its own levers? Back in 2014, the Hong Kong firm, Deep Knowledge Ventures, appointed an algorithm to its board of directors. Named VITAL, the algorithm sifted through market data to highlight potential business opportunities. As a member of the board, VITAL ostensibly had the power to vote on management decisions, including whether to enter into contracts. VITAL’s vote carried no real power, however, and it was largely considered to be “a gimmick.” VITAL only offered advice. Humans made the real decisions.
Using an LLC, however, it is possible to empower an algorithm to contractually bind parties to business decisions. In its articles, an LLC can bind itself to enter into contracts based on decisions made by an algorithm. In this case, the LLC would enter into contracts as a non-human person, and one or more humans would act as the LLC’s agents. Given that the LLC only acts at the behest of an algorithm, the human agents would exist only as formalities. Here, the tables have turned. The humans have become the gimmick.
Several advantages could drive investors towards algorithm-based entities. An entity’s managers and shareholders have an inherently adversarial relationship. They share the same pie. The bigger the slice for one, the smaller the slice for the other. If, as in stock exchanges, software effectively replaces highly trained and expensive professionals, algorithms could represent a cheap alternative to managers and directors. A single director acting at the behest of an algorithm could cost significantly less than a team of directors and managers each with their own slice of the pie. Further, shareholders take great pains to monitor their managers and directors to prevent corporate fraud. If, as has been proposed in some fields of research, algorithmic managers publish their code, fraud would become effectively impossible. In short, investors could have cheap, tireless, and wholly honest managers for their investments.
Risks would also accompany these investments. Even meticulously-designed algorithms can fail in practice. Just last year, experts attributed the “flash crash” of the British pound to rogue trading algorithms. By conducting business at rapid speed, algorithms can have huge effects – positive or negative – in little time. Built-in latency or “circuit breakers” can slow or halt algorithms when they begin to spiral out of control, but a human likely would not have spiraled out of control in the first place. Because algorithms currently lack the common sense or basic wisdom of human directors and managers, they may represent a riskier prospect for investors.

As algorithms improve, they will become increasingly attractive to potential shareholders as replacements for traditionally human directors and managers.

If there comes a time when the human managers of a corporate entity act solely at the direction of an algorithm, the law will be faced with a fundamental question: do those puppet humans assent to the contracts they enter, or does the algorithm itself act as an agent of the corporate entity? Currently, some governments dislike the prospect of any non-natural person serving as a manager or director in a corporation. They further confuse an already tangled web of liability. If one or more humans voluntarily assume accountability for an algorithm’s actions, however, it may convince courts to permit the existence of algorithm-driven organizations.