<POINTS>

• While the content of policies tends to draw the most attention during policy discussions, the method of decision making is just as important as the content. In the era of AI, decision-making methods for policy are becoming an even more pressing concern.
• It is now possible for anyone to generate a sophisticated policy proposal, and this is precisely why we must direct our attention toward the difficulty of value judgements as to what constitutes “good policy” and be willing to invest the necessary time for consensus-building.
• The act of establishing a decision-making method itself falls within a key area of policy called “constituent policy.” HGPI’s June 2026 policy recommendations on Patient and Public Involvement (PPI) in R&D include proposals that fall within this category.
  

Introduction

Policy discussions are usually centered on the content of policies. However, methods of decision making for policy are just as important, if not more so. In particular, I feel that with the rapid advances in AI today, questions regarding the decision-making method have become even more important. In June 2026, Health and Global Policy Institute (HGPI) presented policy recommendations titled, “Expectations for a System that Spans Diseases, Ministries, and Agencies for the Effective Promotion of Patient and Public Involvement (PPI) in Research and Development.” In this column, rather than share the content of those recommendations, I would like to record my thoughts on questions related to its decision-making method, which is a topic the writing process for those recommendations led me to reexamine.

The “Three I’s” of Indivisibility in the Policy-Making Process

The policy-making process is extremely complicated by nature, and it contains many aspects that are completely indivisible. In political science, “idea, interest, and institutions,” or the “three I’s,” are widely referred to as the variables for explaining policy outcomes. In the “idea” plane, different types of beliefs like worldview, principled beliefs, and causal beliefs coexist in discussions on which policies are “correct.” In the “interest” plane, actors hold conflicting interests. For example, in response to a given policy, corporations might consider how it affects the pursuit of profit, politicians may use it to seek election, the government might focus on the evasion of responsibility, and citizens might consider its immediate impact on their household finances. Finally, in the “institutions” plane, existing laws or customs may encourage or even suppress policy change. As these three factors are interrelated, policies are not and cannot be decided easily. The policy-making process is a protracted endeavor wherein participants find common ground or points on which to compromise while acknowledging these difficulties.

What Makes “Good” Policy in the Age of AI?

Rapid progress in AI is transforming our perceptions of the difficulties described above. Anyone can now produce detailed policy ideas with unprecedented speed. With AI, it is now possible for someone to reference a vast amount of literature, identify case studies from developed countries, and write a logical and well-organized proposal in a single night. Some of the proposals generated this way may even actually be “good.”

However, I would like to take a moment to stop and consider. What, exactly, is “good”? This is something I ask myself daily over the course of my activities at HGPI. Who is it “good” for? Even if a proposal is “good” for various stakeholders who are alive today, if it leaves an excessive burden for future generations, can we really call it “good”? Taking a step further, even when a proposal looks like it will be “good” for all of society, if it was put together without the voices of the few who have been placed in disadvantageous positions, can it truly be considered “good”?

In an age where AI is becoming widespread and detailed policy ideas are mass-produced, the more such ideas proliferate, the more obscured the difficulty of these value judgments becomes. The high degree of logical completeness in these detailed proposals may end up obscuring conflicts in the “idea” plane—that is, the fundamental differences in how we view “correctness.” This is why I believe now is the time for us to reexamine the nature of “correctness” in policy.

Two Aspects of “Correctness”

It is easier to determine which policies are “correct” by categorizing them according to two aspects. The first is, “Is the content correct?” The second is, “Was it decided using the correct method?” In legal philosophy, the former is often referred to as “rightness” while the latter is called “legitimacy.” Each aspect has value in its own right, but a policy lacking either one is unlikely to be considered acceptable by society.

The emergence of AI and Large Language Models (LLMs) in particular has made it much easier for people to produce highly-detailed content, and policy proposals generated by AI can possess sufficient “rightness.” However, determining how to make decisions—that is, the act of building consensus—is still an area that demands time and effort from people. In the modern age, when we prioritize cost- and time-effectiveness, some may consider the time spent building consensus as a waste. However, consensus-building is not a process for quickly producing answers. Rather, it is a laborious process in which participants candidly express their interests to each other, learn to accept each other’s thoughts, and persistently seek out the best compromises to make. This “cumbersomeness,” which may appear inefficient through the lens of cost- and time-effectiveness, is precisely the cost we must bear to make decisions in a society with a diverse range of conflicting interests. It is what guarantees the other form of correctness in policy: “legitimacy.”

I would also like to add that this is the interesting part of the policy-making process. We sometimes retract or alter a proposal, or at times pretend we never even presented it to begin with. I believe that this very human behavior that makes the act appear futile at a glance is actually what gives it its meaning.

Establishing a Decision-Making Method is Policy, Too

Furthermore, establishing a decision-making method is policy in its own right. Policy does not only mean the acts of starting new projects or revising existing systems. Considering what to stop, as well as designing who decides, in what venue, and how, are also important policy areas. Political scientist Theodore J. Lowi defined four systems of policy: distributive, redistributive, regulatory, and constituent. Among these four categories, constituent policies refer to the policies that decide the very nature of government frameworks and procedures.

Pillar 3 of the aforementioned policy recommendations on PPI is, “Establish a promotion system that spans ministries and agencies with the Cabinet Office in mind,” and it falls within the area of constituent policy. Specifically, the recommendations call for placing a senior official in charge of PPI at the Cabinet Office and establishing a Working Group that spans ministries and agencies and that includes parties like researchers, patient advocacy groups, and PPI support organizations as members. As this recommendation does not dictate the content of research itself and aims to establish a forum and structure for making decisions about research, it can be considered an example of constituent policy.

Conclusion

In the policy-making process in the age of AI, I believe that instead of allowing ourselves to be lured by the seemingly convincing nature of its content or the speed at which that content is generated, we must be willing to devote the necessary time to decision-making. Continuing to ask ourselves how decisions should be made is just as important as asking what decisions should be made, and this will be an essential element in future policy-making. At HGPI, we will continue, as a non-profit and independent organization, to examine these questions together with a diverse range of stakeholders.

 

Authors

Shunichiro Kurita (Senior Manager, HGPI)