The AI Act [Part 1] Recognizing the risks of entrusting our lives to AI — Interview with Chuo University Professor Hirano [Bridging the humanities and sciences]

AI (artificial intelligence) is gradually permeating into our lives. As Japan does not have laws that comprehensively regulate AI development and use, we must consider potential future challenges and the risks associated with entrusting important aspects of our lives to AI. Science Portal spoke with Professor Susumu Hirano at Chuo University's Faculty of Global Informatics, an expert in AI and law, who also holds a law license in New York State.

Why were job seeking students rejected by an AI interview?

— You serve as a committee member on AI at government expert meetings and similar events, right?

Having become a member of the Ministry of Internal Affairs and Communications' AI Networking Review Conference (formerly ICT Intelligentization Impact Assessment Review Conference), I have participated in government-sponsored meetings. I attended international meetings of the Organisation for Economic Co-operation and Development (OECD) as a representative of my country. I was fortunate to have opportunities to speak with various stakeholders such as companies and consumers, which broadened my perspective. Since establishing the Faculty of Global Informatics at Chuo University in April 2019, I have been teaching students here at the Ichigaya Tamachi Campus (Tokyo) rather than the Tama Campus.

— Did you feel a sense of crisis about how AI is being used while talking with your students?

Yes. The first cohort of students from my seminar told me that they had undergone AI interviews as part of their job search. When I asked about the outcome, they said that things hadn't worked out. When I asked, "Why do you think you were rejected?", they replied, "I don't know," or "There's nothing I can do about it." Of course, I was aware that companies generally do not explain why applicants are rejected.

Still, I could not help wondering: Why do my students not question a situation in which a machine is allowed to determine such an important aspect of their lives? Curious about this, I did some research into legal literature and learned that in the United States, there is a lot of criticism of the practice of entrusting hiring decisions to AI.

One reason for this criticism is that, for many years, numerous major companies have used AI systems to screen the large number of resumes they receive from job applicants in the name of efficiency. The problem, however, is that AI cannot fully understand context. For example, if an AI system is instructed to filter out applicants whose résumés show a leave of absence of more than six months, it may reject them without considering the circumstances behind that gap. For example, there may be a woman with high motivation and ability to work, who took a six-month leave for marriage or childbirth but is now able to return to the workplace. An AI system may fail to appreciate such circumstances and screen her out automatically

Other AI-driven hiring practices have also come under intense scrutiny. One example is the use of AI to evaluate and score recorded video interviews. Critics argue that such systems often lack a sound scientific basis, provide little transparency or accountability, and may produce discriminatory outcomes. In one notable case, a company that offered AI-based interview assessment services was the subject of a complaint filed with the U.S. Federal Trade Commission (FTC) by a civil rights organization. As a result, the company stopped offering the service.

Different approaches among Japan, the United States, and Europe

—The United States seems to be a place where AI is advancing with the growth of tech companies, and issues are indeed arising.

That's right. Because AI and algorithmic decision-making have been adopted more rapidly and extensively in the United States than in Japan, various issues have become apparent earlier there sooner. My personal view is that Japan should regard these issues as "lessons to learn from" and use AI more wisely. I have adopted this approach in my research.

By the way, I find it interesting that the three powers—Japan, the United States, and Europe—take different approaches to various issues inherent to AI (referred to as "AI governance").

First, the European Union (EU) has adopted a highly regulatory approach. Through the AI Act, a regulation that applies directly across EU member states, the EU imposes legal requirements on the development and use of AI. The EU AI Act classifies AI systems into four categories according to risk: unacceptable risk (prohibited), high risk (subject to strict requirements), transparency risk (subject to transparency obligations), and minimal or no risk (generally exempt from specific obligations). Only the lowest level has no restriction, and the rules also include penalties, making it quite strict. The EU takes a 'precautionary principle' stance, imposing regulations before any problem arises.

On the other hand, the United States, which stands at the opposite end to the EU, dislikes the precautionary principle and has taken a "permissionless innovation" stance. This trend has only intensified since President Trump returned to power.

At the same time, it is important to remember that the United States operates under a federal system, in which individual states retain substantial legislative authority. As a result, even if a federal administration favors a less restrictive approach to AI regulation, individual states may pursue different policies. Moreover, future changes in political leadership could also lead to shifts in federal policy. For this reason, it is important not to focus solely on the actions of the federal government. States such as Illinois, for example, have enacted and implemented laws regulating certain uses of AI.

Japan's AI Act: A "Semi-Hard Law" and the challenges of managing risk

— How did Japan come to enact the AI Act?

For many years, Japan has relied on a "soft law" approach to AI governance. To avoid discouraging research, development, and practical applications of AI, the government has issued non-binding guidelines and policy recommendations through expert committees, while encouraging businesses and other stakeholders to comply with them voluntarily.

However, in subsequent public opinion polls, many people answered that they were "worried" about AI. In response, the Cabinet Office established the AI System Study Group in the summer of 2024 to examine the possibility of introducing legislation on AI governance. I participated as a member of the group and contributed to its discussions.

At the first meeting of the study group, then-Prime Minister Fumio Kishida attended and called on the group to develop policies that would strike a balance between managing AI-related risks and promoting innovation. In its Interim Report, released at the group's sixth meeting later that year, the Study Group recommended legislation designed to meet that objective. At that meeting, then-Prime Minister Shigeru Ishiba attended and instructed then-Minister of State for Special Missions Minoru Kiuchi and others to expedite the submission of a bill based on the proposal. Following these recommendations, a bill was submitted to the Diet in 2025 and was enacted in May of that year as the Act on the Promotion of Research, Development and Utilization of Artificial Intelligence-Related Technologies.

The AI Act can be characterized as a form of "semi-hard law," occupying a middle ground between binding legislation backed by legal sanctions ("hard law") and non-binding instruments such as guidelines and policy recommendations ("soft law"). One reason for this approach is concern that overly restrictive regulations could discourage AI research, development, and adoption, thereby undermining the goal of balancing risk management with innovation. Another reason is that meaningful regulation generally requires a sufficient factual foundation demonstrating the need for legislative intervention. Many of the risks associated with AI remain speculative or have not yet fully materialized.

Thus, it was pointed out that drafting a bill of strict regulations was practically difficult at this stage. Furthermore, there is concern that the risks of AI have not been adequately addressed under the current laws in the first place. It was therefore argued within the Study Group that existing laws should first be applied and enforced more effectively. If, through that process, genuine legal gaps become apparent, the factual basis necessary for new regulatory legislation would become clearer.

Reflecting these discussions, the AI Act adopts an approach that emphasizes the implementation and enforcement of existing laws and guidelines rather than introducing comprehensive new regulatory obligations. The AI Act does not impose criminal or administrative penalties. In parliamentary deliberations, the government explained that the measures available under the Act are limited to actions such as investigations, consideration of countermeasures, and the provision of guidance, and that the Act should not be understood as a traditional regulatory statute. AI users and businesses are expected to cooperate with government initiatives; however, the government explained during Diet deliberations that this obligation is merely a best-efforts obligation, rather than a legally enforceable duty.

Enactment of regulatory law is possible if current laws alone are insufficient

— Please tell us specifically about the legislative trends in Japan.

During the Diet deliberation on the AI bill, there was criticism that a best-efforts obligation and the responsibility to cooperate were insufficient. However, as mentioned earlier, there are practical difficulties in stipulating prerequisites for imposing a penalty against a 'potential' issue that has not yet arisen. Rather, the urgent task is to properly enforce the current laws, and this will help us find a potential "gap in the law."

Additionally, the "Artificial Intelligence Strategy Headquarters," newly established under the AI Act with the Prime Minister as the head and all cabinet ministers as members, also serves to support the enforcement of current laws on various AI issues. This Headquarters may act as the command center for various issues caused by AI, so for example, when there is a risk of violations of current laws overseen by each ministry/agency, the Headquarters may support law enforcement.

In addition, by using the government's function of AI-related investigations as granted by the AI Act, the government may monitor trends and progress in AI research and development from now on and prepare for emergency situations. For example, we can prepare for the emergence of "general-purpose AI," "strong AI," or "singularity" that may arise in the future, where AI capabilities could surpass humans and pose a threat to humanity.

In the first place, there is a tendency in Japan to dislike sudden enactment of strict regulatory legislation that lacks legislative facts. For example, the Equal Employment Opportunity Act was not initially a regulatory law before its amendment. However, it was revised to a regulatory law in response to social changes as time evolves. The AI Act also adopted a style of soft law-type legislation as a preliminary step to regulatory law.

In the future, however, as time progresses and AI technology and its applications evolve and become widely spread, various issues will become apparent. Once it is clear that applying current laws and guidelines alone is insufficient, a regulatory law may be enacted, for example, in the form of an amendment of business laws tailored to specific risks.