Legal challenges of administration and AI
The theme of administration and AI has been under discussion since around 2017, but recently it has become a hot topic again, as many examples of AI uses have been seen around us. For administrative law, which considers the relationship between administration and law, it is a problem in a double sense. These are government regulations on AI and government’s utilization of AI.
The former is about how administrations should respond to social changes brought about by the spread of AI. Can traditional regulations adapt to the changes? From the process of generating new regulations to the process of enforcement, there are a number of issues.
The latter is also related to the digitalization of public administration. In order to cope with the shortage of workers due to the declining population and the social changes due to the aging of society, it is essential to use AI to reduce the burden on administrative services (and thus avoid the collapse of administrations). In that case, the issue is also how to control information.
Up to now, the administrative legal system has developed while learning from foreign legal systems. However, the various legal systems are not well organised in information situations, such as privacy and personal information norms.
In the first place, administrative law is sometimes thought to function as a prevention of conflicts. This suggests that there are disputes that cannot be resolved by ex post facto damages and penalties under civil and criminal law.
Each citizen has the freedom to engage in activities he/she wishes to. However, if an activity can pose a danger to society, the government may have to restrict such activities in advance, in order to reduce risks to a reasonable extent.
The main administrative method to prevent conflicts is permission, which verifies whether certain requirements are met. On the other hand, as a reactive measure, we may use our investigation authority, request reports from business operators, and issue business suspension orders or product recall orders in some cases.
For example, traffic rules on roads are stipulated by the Road Traffic Act, and driving is regulated by a permit system that focuses on people, such as driver’s licenses. In addition, there are regulations focusing on businesses such as taxis and trucking (the Road Transport Act, etc.), regulations focusing on things such as the Vehicle Inspection System (the Road Transport Vehicle Act, etc.), and regulations regarding places such as highway standards (the Road Act, etc.).
These administrative powers must have a legal basis because they restrict the rights and freedoms of citizens. Enforceable administrative actions can only be carried out by laws and ordinances created by the legislature representing the people. This is called legal reservation in the principle of administration by law and is an aspect of the rule of law. This has been achieved over a certain level in modern Japanese society.
Even so, now that AI is emerging as something that influences human behavior and thinking, there will be a number of issues to discuss, reviewing the principle of the rule of law.
An AI society requires a policy mix
Let us take a look at government regulations on AI. For example, suppose a car equipped with self-driving technology caused a traffic accident. The government is responsible for road maintenance and construction, and is also involved in the certification of vehicles. Whether they were appropriate or not will be questioned in court under the State Redress Act or other laws in some cases.
However, when AI is involved here, the possibility of making changes or alterations to products after they are shipped becomes an important issue, which is not a prerequisite for conventional administrative regulations. From the viewpoint of protecting users, administrative regulations focusing on things are based on the premise that the conditions of commercial goods are kept as they are at the time of shipment. When maintenance is necessary or a defect that could lead to injury or death has been discovered, regulations such as voluntary recall and recall orders are to be put into practice.
So far, however, this has not been the practice for software, including AI. Rather, users are also required to ensure safety by conducting security updates. In particular, the nature of machine learning, in which programs learn data to create and modify algorithms for making decisions and inferences, is a major issue surrounding safety and regulations.
First, data collection itself has an impact on users’ privacy. In order to enhance accuracy and applicability, AI is expected to collect large amounts of information as learning data and then process and analyze it. However, privacy violations in the process of collecting data, which is the basis of AI, have already surfaced in various areas.
Second, there is the question of who is responsible if any problems arise from the post-learning program. For example, if an AI-equipped device works in a way that was not expected at the time of shipment and leads to an accident, it is necessary to have a system that verifies the determining process.
Third, regulations introduced by the government can have a negative impact on AI development. In fact, in the past, the national government has drawn up guidelines for the development of AI, but there have been quite a few rejections in public comments.
However, in these reactions, there seems to have been a kind of misunderstanding and mutual distrust that the guidelines include only a ban or permission for developers or a verification system. From a legal standpoint, in principle, companies have the freedom to engage in commercial activities, including development, so they will not be bound with regulations as they fear. In other words, the purpose of the development guidelines was to show the direction to be followed and lead society in that direction, instead of simply responding with a ban, etc. It is very unfortunate that there was an overreaction, which could be called a kind of regulation allergy.
The idea that there should be guidelines to be followed has been spread to some extent, and various guidelines have now been issued by public and private organizations. Alternatively, instead of imposing strict regulations, the government can have developers and manufacturers submit reports voluntarily by sharing a vision that emphasizes openness of information. It is also possible for each company to make their information open and for the government to compile it into a report.
Of course, even these methods have their limits. Simply giving administrative guidance and saying, “we ask for your voluntary cooperation” is not a very effective way to handle companies which do not follow this approach. In such cases, a classical regulatory framework is required again … only an administration can do certain things after all. The content of regulations and procedures in this case can also be handled in ways that focus on information regulation, as I mentioned. Regarding voluntary reports I mentioned, making reports itself mandatory is one way of compromising.
Given the current situation in which global companies provide services to every corner of your life in Japan, we should not hesitate to use the regulatory framework itself. It is also important to establish laws and ordinances when necessary, and to have the government exercise its regulatory authority and show determination to protect safety. The fact that regulations are properly enforced in Japan also leads to the trust that the Japanese legal system is trying to protect its citizens, even from a foreign perspective. Now that domestic and overseas data distribution has become commonplace, gaining such trust will also have importance in the medium to long term.
This approach of combining various methods and considering appropriate regulations is a policy mix concept commonly used in environmental policy.
The point is that a multi-layered system designed by individual laws to ensure safety and civil rules is ideal. In order to ensure safety and transparency, I believe that both public and private sectors should find best practices by combining their approaches through trial and error, rather than focusing solely on regulations or relying entirely on self-regulation without regulatory systems.
Legislation that balances safety and convenience
In this way, while there are various issues regarding regulation on AI, the government’s utilization of AI is in progress simultaneously. As the labor force is constrained by the declining birthrate and aging population, automation and manpower reduction are essential in national and local governments.
An example of a use that has already been introduced in some municipalities is nursery school matching by AI. There have been several successful cases where the use of AI has significantly reduced the time required for services that have been performed by local government employees, such as selecting and allocating a large number of applicants for a nursery school and sorting out various requirements.
On the other hand, from the viewpoint of administrative law, there is a debate over the decision of the administration on whether to accept applicants to nursery schools or not. In other words, when applicants are dissatisfied with a notification from the government office that their child was not accepted to the nursery school of their choice, the government is responsible for explaining how the decision was made. The government staff cannot get away with its responsibility by saying, “We don’t know the reason because AI made the decision.” The transparency of the system must be ensured for both administrative staff and applicants.
In addition, generative AI such as ChatGPT is expected to improve the efficiency of government office operations, but the problem here is the relationship with information. To improve the accuracy of generative AI, we need to make AI learn data. However, it seems difficult for the government to determine how much information can be given at that time. Issuing too many restrictions without much consideration might result in ruining the convenient features of excellent systems.
Moreover, the digitalization of legal affairs, which the Digital Agency is currently promoting, is an issue for discussion. A road map, which shows the perspective of eventually introducing the use of digital technology to the administration and enforcement of laws, has been laid out. However, even there, the principle of administration by law will become an issue. I just introduced the classic idea of carrying out enforceable administrative actions. As we have acted in accordance with the idea, legal regulations on the aspect of a thin but broad and significant impact is not very advanced. I think the question needs to be asked again: “To what extent is law necessary?”
In the legal world, until now, legislators who are representatives of the electorate have established laws, and government has made rule-based decisions on the details that cannot be determined by law. From now on, AI is expected to assist increasingly in the decision-making that administrations have thus far undertaken. Even then, we must shift our laws and systems so that we can develop laws that balance safety and convenience.
The changes brought about by the spread of AI have the potential to change the way legal discipline works in all areas. In this regard, administrative law is discussing mainly human rights and organizations, and in particular, the administrative information law, which deals with issues such as information disclosure, personal information protection, and public records management, has been somewhat neglected. However, in an era where coexistence with AI is inevitable, I think the legal field dealing with information will become more important.
* The information contained herein is current as of March 2024.
* The contents of articles on Meiji.net are based on the personal ideas and opinions of the author and do not indicate the official opinion of Meiji University.
* I work to achieve SDGs related to the educational and research themes that I am currently engaged in.
Information noted in the articles and videos, such as positions and affiliations, are current at the time of production.