In 2019, Facebook announced that it had begun using machine-learning
algorithms to preemptively screen uploads for nonconsensual pornography.
Although the use of screening algorithms has become commonplace, this seemingly
minor move from reactive to preemptive legal analysis–based prevention—this
Article argues—is part of a groundbreaking shift in the meaning and effect of
algorithmic screening, with potentially far-reaching implications for legal discourse
and development.

For more than a decade, employment discrimination causation law has been a confusing, often overly restrictive quagmire that has contributed substantially to the paltry success rate of plaintiffs in employment discrimination cases. Most of these cases are dismissed pretrial, all too often based on a failure of causation. A key reason traces back to loose and misleading language—centered on a single word—in a 2009 Supreme Court opinion involving the but-for causation standard that applies in most discrimination cases. The Court said that the discriminatory motive must be “the” but-for cause of the employer’s action when it should have said “a” but-for cause. This language incorrectly implies that the discriminatory motive must be the sole cause—“the” cause—of the employer’s action, and though the sole-causation standard is demonstrably wrong, many courts nevertheless have required such a showing.

Civil rights advocates have long viewed litigation as a vital path to social change. In many ways, it is. But in key respects that remain underexplored in legal scholarship, even successful litigation can hinder remedial projects. This perverse effect stems from civil rights doctrines that incentivize litigants (or their attorneys)to foreground community plight—such as academic underachievement or overincarceration. Rational plaintiffs, responding in kind, deploy legal narratives that tend to track racial stereotypes and regressive theories of inequality. When this occurs, even successful lawsuits can harden the structural and behavioral forces that produce and perpetuate racial inequality.

Technologies such as social media, autonomous vehicles, and “big data” analytics generate enormous benefits for society, but they also create substantial harms. Many of these effects take the form of externalities—external benefits and harms that a decisionmaker (such as an inventor) imposes on third parties without charge or compensation. Considering negative externalities, for example, social media networks spread misinformation throughout the electorate, autonomous vehicles threaten the jobs of millions of professional drivers, and predictive policing based on big data can lead to unreasonable searches and seizures. Externalities can cause inefficient resource allocation, and the classic remedy is to “internalize” externalities by ensuring that decisionmakers consider the external benefits and costs of their actions. Patents, which confer exclusive rights on new inventions, enable inventors to internalize a share of the positive externalities from technology, thus shoring up incentives to invent. However, inventions also produce harms, and how patents treat negative externalities from new technologies has been largely overlooked. This Article is the first to extensively examine this issue. It argues that while patents internalize positive externalities associated with innovation, they do surprisingly little to internalize negative externalities. This Article refers to this underappreciated dynamic as patent law’s externality asymmetry.