Ideal enforcement


Safety. Mobility. Technology. Data. Equity. Autonomy. Community. Justice. Power. Trust. These themes have resonated over the last few years across a wide range of specific issues—some that are new and others that have only recently received the broader attention that they demand. The implications of vehicle automation for enforcement capture all these issues: They meet—in the language of physics, they interfere, whether constructively or destructively—to affect both perception and reality. And so our topic, particularly its focus on automation and nongovernmental actors, is both important in its own right and useful for exploring the larger challenges and opportunities that these themes present.

These challenges are numerous. Far too many people die on our roads—more now than since the pandemic began and an increasing share of whom were just walking or biking. Insights from countries with dramatically better traffic safety records—particularly the "safe systems approach" that originated in Sweden and is better known as "vision zero" in the United States—have motivated a growing recognition that every traffic death is the culmination of a series of failures and that roadway design is often foremost among these failures. Meanwhile, the tens of millions of traffic stops each year—and "traffic" includes pedestrians as well as motorists—have come under greater scrutiny in connection with prominent police shootings that range from merely preventable to outright criminal.

These high-profile incidents, the statistical rarity of which does not diminish their tragedy, represent the tip of an iceberg of larger concerns about racial and other inequities in both treatment by and access to law enforcement. This has led to calls for more just enforcement and to a discussion of the relationship between justice and safety—whether a "safe systems approach" and a "just systems approach" are antagonistic, symbiotic, or inextricably intertwined such that neither safety nor justice is possible without the other. Even more broadly, particularly in a car-dominated country, injustice also manifests in a mobility infrastructure that is simultaneously difficult to access and impossible to escape: Black households are less to have access to a car, less likely to have adequate sidewalks, more likely to breathe hazardous air, and more likely to be exposed to pervasive noise pollution.

Technology generally—and automated driving specifically—is often offered as a panacea that can eliminate human error and mitigate social inequity. The reality, of course, is that technologies both exacerbate and ameliorate existing problems while creating both new problems and new opportunities. The internal combustion engine, for example, was the environmentally friendly improvement over the horse—a transport mode that produces some 25 pounds of manure a day. In this way, innovation is about replacing one set of problems with a new set of problems and hoping that the new set is in aggregate less than the old. Policy, which is much the same, layers on top of innovation. Another dependent function—or, to return to physics—another set of waves capable of interference. Law in turn implements, and sometimes constrains, those policy choices.

Depending on the trajectories of both innovation and policy, increasing vehicle automation could revolutionize enforcement in both a narrow sense and a broad sense. In the world of thought experiments, where every motor vehicle on every road is fully automated, every motor vehicle could also be programmed not to speed, thereby wholly negating—or, somewhat ominously, perhaps merely displacing—traditional traffic tickets along with the law enforcement infrastructure that they require and the semirandom police-public contacts that, for better or worse, they facilitate. More realistically, even a smaller proportion of automated vehicles that predictably follow a lawful speed could have a calming effect on human drivers. Even today, speed governors are mandated on trucks in Ontario (the rule having survived the argument that it infringes on drivers' fundamental freedoms), and intelligent speed control is available on some European cars. Automated vehicles might also form a fleet of mobile speed cameras that generate evidence to ticket other motorists for speeding. And yet the more conventional speed camera, which has been available for decades, has experienced radically different receptions around the world, with some US states restricting their meaningful use.

As the safe systems approach would remind us, other forms of infrastructure can also constitute not just enforcement but a more effective form of enforcement: Designing roads as streets rather than speedways through technologies such as speed cushions, chokers, and pavement treatments (or the lack thereof) can slow vehicles far more consistently than the occasional police officer. And so, while automated vehicles might eventually offer new enforcement tools, they could also distract from the low-hanging fruit available today.

Some governments have already recognized some arguably low-hanging fruit: Many states already require drivers previously convicted of intoxicated operation to use an ignition interlock device that prevents them from operating their vehicle when they are drunk. Congress has now directed the US National Highway Traffic Safety Administration (NHTSA) to, within six years, mandate impaired driver detection in new vehicles. In a broader sense, impairment could also include drowsiness, distraction, and the influence of drugs as well as alcohol. And some advanced driver assistance systems already incorporate limited—and not necessarily effective—driver monitoring to discourage inattention. Two private organizations that rate vehicle safety recently announced that effective driver monitoring will figure prominently in their ratings.

Companies also play a role—likely a changing role—in enforcing laws as well as norms. This is especially evident in ongoing public debates about right-to-repair laws (including in the vehicle sector) and about the responsibilities of Internet platforms and other intermediaries. Automotive manufacturers, telematics providers, fleet operators, so-called transportation network companies such as Uber and Lyft, and in some cases insurers and even employers also have considerable insight into and control—in both the practical and legal senses—over their systems. In theory, some of these companies could already prevent speeding (by implementing intelligent speed control), report speeding (by notifying law enforcement), or penalize speeding (by banning offending users).

As more inward-facing sensors and computers are added to advanced vehicles, companies may face similar enforcement choices with respect to aggressive, intoxicated, inattentive, and drowsy operation. And even when automation replaces human driving, this control may still reach other behaviors inside a vehicle that might be unlawful (such as failing to wear a seatbelt or using illicit drugs) or, in the company's view, undesirable (such as dirtying the interior or having sex).

The powerful external sensors on advanced vehicles may also extend a company's digital awareness to far outside those vehicles. The digital version of Jane Jacobs's "eyes on the street" could identify all manner of major crimes and minor violations (even those, such as solicitation or loitering, that have little to do with road traffic). Imagine, for example, that a pedestrian impedes, whether lawfully or unlawfully, the movement of an automated vehicle operated by a large Internet company. Using facial recognition or cellphone data, that company may be able to identify the pedestrian with some level of confidence. Might the company then penalize that pedestrian by temporarily restricting their use of its automated driving service, its shopping site, or its social media platform?

These possibilities demand a thoughtful analysis of and approach for enforcement in service to any goal—and here the devil is in the details. Take the important goal of traffic safety. First, "traffic safety" requires a workable definition. Is it only harms from crashes or also harms from the road pollution that costs at least as many lives each year as those crashes? Is it only physical safety or also feelings of safety, including fears of police misconduct? Does it encompass or exclude equity and autonomy? Second, analysis of that safety requires an acceptable analytical scope. Some potential approaches to optimizing traffic safety—such as discouraging single-occupant vehicle trips and discouraging sprawl—might not be directly related to driving itself. And some interventions could have unintended consequences—such as depressing activity, encouraging overreliance, or displacing responsibility—that could ultimately have adverse safety impacts.

Safety is not the only ambiguous concept that underlies our discussion of enforcement. Others include behavior, rules, regulation, norms, behavior, compliance, and enforcement itself.

Behavior is the conduct—including both actions and omissions—of actors including individuals, governments, companies, organizations, and even machines. Behaviors are shaped by rules, by the enforcement of those rules, and by norms. Behaviors in turn shape those norms.

Rules are public or private requirements, and norms are social or business expectations. Rules and norms can be identical, complementary, or in conflict. For example, traveling just above the speed limit is a norm that is both in conflict with a rule (the speed limit) and arguably complementary to that rule (since it emphasizes near-compliance with that rule). This relationship is even more complex: Many nonstatutory speed limits are set so that 15 percent of drivers will violate them (and 85 percent will not)—an approach that has generated deserved criticism.

Regulation broadly refers to ways that the behavior of actors is checked and influenced. Regulation can be public (such as speed limits) or private (such as risk-dependent insurance premiums). And it can be prospective (such as a driver's license requirement) or retrospective (such as civil liability for negligently causing a crash). Rules are one form of regulation.

Compliance is the conscious or unconscious conformance of behavior to rules. It is a challenging concept for several reasons. First, compliance might instead be used with reference to norms (such as polite and respectful behavior) that are not themselves rules and that may (as in the case of speeding) conflict with rules. Second, compliance with rules might not be sufficient for safety; it has been suggested that perfect compliance with existing rules of the road could reduce fatalities by just over half—a gain that would be as massive as it would be insufficient. Third, it is not clear how compliance should be measured. Does a situation in which most drivers slightly speed most of the time represent mass compliance or mass noncompliance? Even the unit of noncompliance—in other words, each ticketable offense—becomes less clear as enforcement becomes more routine: Has a motorist who consistently speeds for two separate periods of 20 minutes within a single hour failed to comply with the speed limit once or twice or forty times?

In some contexts, states recognize that perfect—that is, complete and absolute—compliance may not be ideal. Some laws expressly provide flexibility: New Jersey recognizes exceptions to a prohibition on crossing the double-yellow line, for example, and other states have an enforcement threshold for speeding. Some implicitly subordinate one legal requirement to another: The duty to exercise due care to avoid striking a pedestrian would supersede a requirement to avoid crossing a double-yellow line. At least in the criminal context, defendants could also assert defenses of necessity, protection of self, or protection of others. And in practice, much of this flexibility is provided by the lack of detection when no police officer is present and by the exercise of discretion when a police officer is present.

The relationship between noncompliance and safety likely depends on the circumstances. In a first case, nominal noncompliance clearly increases safety or is necessary to avoid a greater legal violation, as when a motorist crosses a double-yellow line to avoid striking a bicyclist who falls. In a second case, noncompliance arguably increases safety, as when a motorist crosses a double-yellow line to give extra space for a bicyclist or to get around a parked truck. In these circumstances, the motorist could in theory follow the bicyclist until reaching a passing zone or wait behind the truck until it moves, but these decisions could introduce their own safety risks, such as being hit from behind or prompting another driver to attempt an aggressive passing maneuver. In a third case, noncompliance has an ambivalent or ambiguous relationship to safety, as when a motorist speeds on a freeway to travel at the prevailing speed or exceeds the speed limit to pass a tractor on a rural road. Not speeding on a freeway, for example, could have a negative short-term impact on safety but, in the aggregate, could have a positive long-term impact on safety by contributing to a norm of more appropriate speeds. In the fourth and easiest case, noncompliance clearly decreases safety.

Even in this easiest case, enforcement is not straightforward. Enforcement is the provision of information and incentives for compliance. It is another form of regulation. It can include communicating rules (such as posting speed limit signs), penalizing noncompliance with those rules (such as issuing a ticket for speeding), attributing responsibility for harms (such as through a prosecution for vehicular manslaughter), and providing education aimed at creating or reinforcing norms that complement rules (such as through driver education and safety campaigns).

As with compliance, it is important to distinguish perfect enforcement from ideal enforcement. Perfect enforcement is complete and absolute, whereas ideal enforcement is socially optimal. Routine speeding—indeed, the social opprobrium that can at times attach to the failure to speed—is the most obvious example of a difficult truth in the United States: Many of our many laws are regularly flouted with little consequence—or at least with disparate consequence. The perfect enforcement of these laws could profoundly and unpredictably change our streets and our society. One likely consequence would be safer roadways with lower vehicle capacities, as motorists yielded to pedestrians at all crosswalks and maintained required following distances on all freeways. Another might be an increase in the quotidian burdens on those at the margins of society who drive unregistered cars or take unlicensed shuttles from their illegal apartments to their unlawful jobs—violations that the combination of vehicle sensors and machine intelligence might someday be able to readily identify. In short: How do we achieve ideal enforcement as perfect enforcement (or at least something approaching it) becomes technologically conceivable?

And so, if the objective of enforcement is to achieve a socially desirable level of compliance, then it is necessary to consider the costs as well as the benefits of that enforcement. Costs may include the monetary expenses and disruptions (including to safety) of specific enforcement activities. They may also include unintended consequences of enforcement generally, including risk homeostasis (in which people trade one risk for another), the abdication of personal responsibility, and the overreliance on technology. And they may include the psychic burdens on individuals and communities of pervasive and permanent surveillance, the perception of oppression, and the very real opportunity for that oppression, whether through abuse by design or abuse of individual discretion.

The promise and peril of human discretion characterizes several of these concerns. Human discretion is tricky: Its exercise can prevent some harms and contribute to others. Its importance vis-à-vis technology features prominently in science fiction as well as in historical fact: Stanislav Petrov is credited with saving the world by trusting his intuition rather than the Soviet Union's launch detection system. Similarly, criminal and often civil law emphasize the related concept of culpability.

Culpability on its own, however, is generally insufficient for criminal or civil liability, which tend to also require conduct or conduct plus harm. And so, in many areas of the law, noncompliance in the abstract is also treated differently than noncompliance in practice. For example, the Uniform Law Commission's model law for automated driving does not penalize companies for designing automated vehicles that might deviate from a rule of the road but does penalize them when their vehicles do in fact deviate.

As we have discussed, enforcement is broad—certainly much broader than just a police officer with a radar gun. Traffic enforcement specifically can be carried out by a wide variety of actors, including police officers, regulators, companies, and victims. And the subjects of enforcement can—and, importantly, should—include more than just individual road users: We must also consider other individuals and entities, including governments and companies, that make design decisions further upstream the safety stream, including in the design of roads, vehicles, and regulations. Enforcement can be effected through negative consequences intended to deter bad choices, through positive consequences intended to encourage good choices, and through the design of systems to limit choices to those that are good: Driving at 120 miles per hour is simply not possible in a car that does not go that fast on a road that cannot physically accommodate the acceleration necessary to reach such a speed. Indeed, this systematic enforcement epitomizes the principles of effective enforcement: swift, sure, and universal.

We can already see some evolution in the specific case of traffic enforcement. In the traditional narrow model, governments set traffic rules, rely on the basic lawfulness of drivers, and occasionally enforce those rules through semirandom police stops. In the safe systems model that has heavily influenced parts of Europe, enforcement moves upstream to reach design decisions and expands to include infrastructure itself, including physical measures to calm vehicle traffic. And in the future, ubiquitous digital technologies might open new models for public enforcement, for public-private enforcement, for private enforcement, for enforcement by design, and for enforcement inside or through a vehicle. These new models might in turn justify or facilitate changes to the underlying substantive rules—by, for example, defining "reasonably safe speed" with mathematical precision.

If we were to summarize these opportunities in a grand unified theory of ideal enforcement, it might have five key principles. First, define goals broadly. Second, recognize costs and benefits, including those that are abstract or uncertain. Third, focus on upstream systematic decisions by regulators and designers. Fourth, make desirable behaviors easy and undesirable behaviors hard. And fifth, reassess in light of evolving norms, values, and technologies.