Each year as summertime arrives neighborhood lemonade stands begin to magically appear, staffed by entrepreneurial children learning the ropes of running their own miniature business. Just as predictably, those child-operated stands start generating police complaints about everything from noise complaints to being unlicensed commercial food vendors. Laws rarely explicitly exempt child-run food establishments run off a folding card table on a front lawn, leaving it to law enforcement, health departments and lawmakers to intervene when the local news crew shows up to cover armed police being dispatched to shut down a first grader’s bake sale. This raises the question of what will happen as AI gradually takes over the law enforcement and judicial systems and replaces human judgment with the cold algorithmic efficiency of enforcing laws verbatim as they are written on paper.
Enforcing the nation’s laws is an extremely human process. While we often think of our legal system as a relatively cut and dry application of exceptionally detailed written rules, the reality is that the reason we have judges and juries and laws that give a degree of discretion to police officers is that laws are inherently incomplete. Ambiguousness in the wording of our laws is both intentional to give the judicial system discretion or to allow the law to evolve over time and unintentional as unexpected scenarios arise that the law was not designed to address. This places considerable responsibility in the hands of the humans that execute our laws, allowing them to use “common sense” but also raising issues of bias as laws are enforced unevenly.
Take the classic children’s lemonade stand, a staple of summer entrepreneurial spirit for decades. Most municipalities have laws regulating small businesses in their area and require a variety of permits, certifications and specialized equipment and food handling processes for businesses that prepare or sell unpackaged food. Not all cities and states explicitly exempt child-run temporary fundraisers from their health inspection laws, leading to situations like last month’s “Lemonadegate” when a New York state health inspector shuttered a 7-year-old's lemonade stand.
In such situations, lawmakers and law enforcers are typically eager to show their human side and allow “common sense” to rule. The Governor of New York ended up weighing in on Lemonadegate, ordering the state’s Department of Health to permit the stand to continue to operate, even as the department noted that the law did not appear to permit it. Or a case in Illinois where police officers were called to another child’s lemonade stand and instead of shuttering it, the department posted on its Facebook page about enjoying the “illegal” lemonade.
On the one hand, it might seem like “common sense” to exempt a 7-year-old’s lemonade stand from Department of Health inspections and permit requirements. On the other hand, those laws exist for a reason. What would happen if the lemonade container had accidentally come into contact with peanuts at one point and someone with a peanut allergy died from consuming it? The same people ridiculing the health inspectors would instead be demanding resignations for failing to shut down an unlicensed food vendor. Similarly, rules against barbequing outside of designated zones typically exist to address fire safety concerns. While it might seem overkill for someone to call the police about someone grilling in a no-grilling zone in a public park, if that innocent grilling sparked nearby underbrush and lead to a massive fire that consumed nearby houses, the public would demand to know why fire regulations were not enforced.
More importantly, these situations leave considerable latitude to enforcement officers to decide on their own whether to apply the law, leaving the very real potential for bias in deciding which cases to ignore and which to enforce.
What does all of this have to do with the future of AI policing? The fact that in all of these cases, human judgment trumped the literal interpretation of the law.
Former Estonian President Toomas Hendrik Ilves is fond of quipping “you can’t bribe a computer” when describing how Estonia’s computerized government has eliminated the human touch from the majority of government-citizen interactions, removing the human in the loop element that is a key source of corruption in other governments. In Estonia’s case government automation extends only to routine administrative tasks, not AI-powered policing or judiciary, but the idea of computers enforcing the absolute letter of the law as written, without the power of human judgement to bend the rules for good or bad is both a tremendous leap forward in eliminating corruption and also a potential source of problems when laws are written under the assumption that humans will enforce them loosely rather than literally.
What happens as AI increasingly takes over law enforcement? An AI algorithm could certainly be programmed to allow for a certain level of discretion, but that flexibility itself would be subject to the same biases that exist today, though at least they could be quantified, documented and potentially mitigated. Allowing such flexibility without inadvertently missing legitimate cases would be far harder.
It is also unclear how society would react to an AI system that allowed some law breakers to go free while charging others for the exact same offense. What happens when an AI algorithm decides that one child’s lemonade stand should be allowed to operate due to estimates it makes about the health standards of the neighborhood, while another in a different neighborhood is shuttered due to the algorithm’s estimates of lower hygienic standards or poorer water quality there? It is not too hard at all to envision such scenarios, with AI taking into account a myriad factors and informational streams not available to their human counterparts.
The greatest question, however, is how the shift to more mechanical and codified law enforcement would impact the way in which we write and enforce our laws. If computerized enforcement requires exquisitely detailed laws that cover every case and exemption, with no room for compromise, our laws will become increasingly brittle and require constant renovation to ensure the algorithms have the necessary flexibility to accommodate society’s continually evolving standards.
At the same time, exhaustively codifying the exceptions to our laws, rather than leaving them to ad-hoc political interference, would avoid reactive situations and ensure evenhanded application. If the law states that anyone under the age of 18 running a food or beverage stand is exempted from all health and business permit and inspection requirements, then concerns over bias are largely eliminated as law enforcement is no longer required to decide which stands to leave open and which to close and politicians no longer intervene in some cases but not others. It would also force lawmakers to be far more explicit when writing laws and to explicitly decide up front what exemptions to permit and which to disallow.
In the end, the shift towards the more brittle and codified world of AI law enforcement would make our laws more brittle and in need of constant updating, but also eliminate many sources of bias in the enforcement of basic laws, removing the fallible human element. Whether our legal system could cope with a world that required explicitly enumerating every single exemption and removing common sense and human judgment from the equation is a different story.
