Justice Algorithm
The amount of information available for processing and analysis using computers is growing like a snowball. Data from CCTV cameras, GPS-trackers, mobile phone sensors, records of financial transactions, the history of visits to pages on the Internet are having an increasing impact on decision-making. And the more of this data, the more you have to rely on their automatic interpretation. An inevitable consequence of this is the emergence of “computer justice” systems, which, without human intervention, reveal violations of laws and regulations. Fines for speeding, which are automatically calculated based on data from cameras and radars or a content analysis system on Youtube that searches for copyright infringement, are already everyday reality.
A group of American scientists, combining lawyers, linguists and programmers, conductedAn interesting experiment in this area. During the experiment, 52 student programmers had to create a program that would analyze data from a GPS tracker installed in the car and write out fines for violating the high-speed mode in accordance with the New York State traffic rules. This turned out to be a very difficult task - even in the most law-abiding states, laws are never carried out literally and at 100%. Some violations go unnoticed, some too insignificant for law enforcement officers to pay attention to them. Computers do not forget anything and do not miss anything. The thoughtless application of rules and algorithms leads to unnecessarily harsh penalties and ridiculous mistakes like blocking videos with wind noise for violating copyright.
The experimental data consisted of a GPS track obtained during a real trip by car, information about the speed regime of different sections of the route, and the actual set of traffic rules. The participants in the experiment were divided into three groups. The first group wrote the program, strictly and literally following the text of the rules. The second group was to follow the spirit, not the letter of the law, and its participants independently decided in which cases speeding deserved a fine. The programmers from the third group were given a specification compiled by the experimenters, where the permissible limits of violations and sensor errors were detailed.
According to the graph of the vehicle’s speed, the driver generally complied with the rules - insignificant peaks that went beyond the official limits were most likely associated with overtaking on the highway or changing terrain. At the same time, the programs that followed the letter of the law during the trip on average wrote absurd 498 fines. In the group that followed the spirit of the law, an average violation was recorded 1.5 times. The group that wrote the code according to the specification drawn up by the experts found no violations at all.
During the writing of the programs, many factors were revealed that the law does not say a word about, but which have a huge impact on decision-making. What is the error of the sensor? Is speeding 5 miles per hour considered a violation? And at 10? If it lasted only a few seconds? Or minutes? Or repeated several times over the course of an hour? The legislator leaves these and other questions to the discretion of the performers. A living policeman is unlikely to fine the driver, who exceeded the speed by several seconds during overtaking (at least if he is interested in real observance of the rules, and not in the execution of the plan for fines). The program does not have common sense, so the programmer needs to take into account all the nuances and special cases.
Interestingly, students from different groups reacted very differently to the possibility that their program would analyze their own ride. In the group that followed the letter of the law, only one student agreed with this, but only on condition that there was a “backdoor” in the system that would allow him to avoid punishment. The results of those who wrote the program according to a strictly specified specification were equally low, and this despite the complete absence of fines in their version of the program. Only among those who decided how the program should work, 37.5% were ready to obey its decision.
The authors of the experiment believe that justice algorithms have a great future, and the practice of using computers can have a strong influence on the laws themselves. At the intersection of jurisprudence and programming, a new, highly demanded profession may arise. Computers and the Internet have irreversibly changed many areas of human activity. The education revolution is already in full swing, the Watson supercomputer makes diagnoses no worse than professional doctors, and there may be big changes in jurisprudence.
A group of American scientists, combining lawyers, linguists and programmers, conductedAn interesting experiment in this area. During the experiment, 52 student programmers had to create a program that would analyze data from a GPS tracker installed in the car and write out fines for violating the high-speed mode in accordance with the New York State traffic rules. This turned out to be a very difficult task - even in the most law-abiding states, laws are never carried out literally and at 100%. Some violations go unnoticed, some too insignificant for law enforcement officers to pay attention to them. Computers do not forget anything and do not miss anything. The thoughtless application of rules and algorithms leads to unnecessarily harsh penalties and ridiculous mistakes like blocking videos with wind noise for violating copyright.
The experimental data consisted of a GPS track obtained during a real trip by car, information about the speed regime of different sections of the route, and the actual set of traffic rules. The participants in the experiment were divided into three groups. The first group wrote the program, strictly and literally following the text of the rules. The second group was to follow the spirit, not the letter of the law, and its participants independently decided in which cases speeding deserved a fine. The programmers from the third group were given a specification compiled by the experimenters, where the permissible limits of violations and sensor errors were detailed.
According to the graph of the vehicle’s speed, the driver generally complied with the rules - insignificant peaks that went beyond the official limits were most likely associated with overtaking on the highway or changing terrain. At the same time, the programs that followed the letter of the law during the trip on average wrote absurd 498 fines. In the group that followed the spirit of the law, an average violation was recorded 1.5 times. The group that wrote the code according to the specification drawn up by the experts found no violations at all.
During the writing of the programs, many factors were revealed that the law does not say a word about, but which have a huge impact on decision-making. What is the error of the sensor? Is speeding 5 miles per hour considered a violation? And at 10? If it lasted only a few seconds? Or minutes? Or repeated several times over the course of an hour? The legislator leaves these and other questions to the discretion of the performers. A living policeman is unlikely to fine the driver, who exceeded the speed by several seconds during overtaking (at least if he is interested in real observance of the rules, and not in the execution of the plan for fines). The program does not have common sense, so the programmer needs to take into account all the nuances and special cases.
Interestingly, students from different groups reacted very differently to the possibility that their program would analyze their own ride. In the group that followed the letter of the law, only one student agreed with this, but only on condition that there was a “backdoor” in the system that would allow him to avoid punishment. The results of those who wrote the program according to a strictly specified specification were equally low, and this despite the complete absence of fines in their version of the program. Only among those who decided how the program should work, 37.5% were ready to obey its decision.
The authors of the experiment believe that justice algorithms have a great future, and the practice of using computers can have a strong influence on the laws themselves. At the intersection of jurisprudence and programming, a new, highly demanded profession may arise. Computers and the Internet have irreversibly changed many areas of human activity. The education revolution is already in full swing, the Watson supercomputer makes diagnoses no worse than professional doctors, and there may be big changes in jurisprudence.
