Perfect law enforcement is a frequently used key element in fictional dystopias. Usually it is depicted as a classic crowd-sourced surveillance state, where people watch and report on each other, or as a technologically advanced authoritarian state, where cameras/sensors penetrate all aspects of society and some advanced computer algorithm scans all the data for unlawful activities. Some (such as shown in the movie Minority Report) go even further, and have crime prediction capabilities brought in to prevent crime before it happens.

The morals of these stories typically are that any such perfect law enforcement system would be too powerful, and some small group of people will use it for their own gains. Since we cannot prevent people from abusing it (i.e. can’t fix human nature), we probably should not have the all-powerful system built in the first place. But, at least in principle, perfect law enforcement is usually seen as desirable. This has more or less been my own belief for years. In fact, I’ve always fantasized about a scenario where somehow we can build such a law enforcement system, and hand over control of it to some guaranteed-to-be benevolent AI, then society would be better off.

That is, until I ran into this classic piece on Moxie’s blog recently. In it, he tackles the concept of perfect law enforcement at the foundational level, and claims that it is not desirable, even in principle. Even if we can guarantee that such a powerful enforcement system will not be abused, society would still be much worse off with such a system in place. His main argument is that the ability to break laws without consequences (i.e. getting away with crime) is one of the main ways for laws to evolve/change over time, and is therefore absolutely critical to the overall progress of society.

I think this is absolutely true. Oregon, Colorado, Washington, and Canada did not just one day decide to legalize marihuana out of the blue. It was a very gradual and long process, over many years, that involved people using marihuana illegally. Such processes always start with small groups of risk takers, which over time expand to larger and larger groups, gaining more and more social acceptance. And by the end, the actual legalization of the activity is more just a formality, because the activity, which may still be technically illegal, is already well accepted by society at large that it’s essentially practically legal.

There are of course other ways for laws to change. But I think this natural, gradual process is one of the best (because it comes directly from the people, going in the bottom-up direction), most commonly occurring, and most powerful ways. Having a perfect law enforcement system would completely destroy this process of social change.

Okay, now that we recognize this, what about implementing a mechanism in our otherwise perfect system to allow this? Suppose that in our guaranteed-to-be benevolent AI (that we are feeding all our surveillance data into), we programmed in a small random probability for the system to completely ignore an enforcement event. This would make it possible for anyone to get away with any crime, albeit with a small, random probability. It would be a “programmed probabilistic near-perfect law enforcement system”.

The first problem with an implementation like this is how to determine the random probability of “getting away with crime”. We probably don’t want the probability to be uniformly distributed across all types of crime. For extremely severe crimes, like murder, the probability should be 0 (or nearly 0), and for something more trivial, like mild vandalism, the probability can be higher. Determining the optimal probability distribution in this system for all crime is not a simple problem, and may actually be as difficult as building this law enforcement system itself, if not more.

Another problem with this system of law enforcement is the distribution of the random chance of crime evasion among the population. In the current world, with our imperfectly enforced legal system, the probability to get away with a crime is not uniformly distributed. More prominent individuals, who are in the public spotlight and hold a lot of influence, usually find it more difficult to get away with certain crimes than a non-celebrity (and perhaps easier to get away with other crimes due to the power they hold). This creates a complex dynamic, where each individual in society is weighing the risks and gains in his/her particular situation, when deciding to commit or crime or not. Similar illegal activities usually begin within groups of people that share a social circle for this reason, as they all share a higher-than-usual chance of getting away with the crime. This would be nearly impossible simulate in the purely random system, where everyone has exactly the same probability of evading the same crime. Under such a system, criminal activities of all kinds would be much more discouraged, resulting in more stagnation of the legal system’s evolution.

With these problems in mind (and many others that I had not considered), it seems that our current far-from-perfect law enforcement system (in the West) is actually quite good. It deters and punishes most criminal behavior reasonably well, without infringing on the privacy of the people too much, all while still allowing some to get away with crime to facilitate the natural evolution of laws over time. I agree with Moxie’s blog title, that “we should all have something to hide”. In the game of compromise between privacy and security, I think favoring security is often a short-sighted mistake, one that we are perhaps making too many times these days.