Generative AI and Copyright — Consideration of Intellectual Property Systems in an AI Society

Issues surrounding generative AI and copyright have recently been positioned as important policy challenges, as exemplified by the draft “Principle Code for Protection of Intellectual Property and Transparency for the Appropriate Use of Generative AI (provisional title) (draft).”

With respect to generative AI and copyright, whether the combination of Japan’s flexible copyright limitations (including Article 30-4 of the Copyright Act) and transparency securing frameworks such as the draft Principle Code is truly appropriate constitutes an issue that represents a turning point for Japan’s future.

Accordingly, regarding the issue of generative AI and copyright, a comparative examination of five approaches was conducted with the support of AI: the U.S. fair use model, the EU opt-out model, Japan’s flexible copyright limitation model, the comprehensive licensing model of the “one nation, one license, one payment” system currently under consideration in India (although still under consideration, hereinafter referred to as the “Indian approach”), and the Data Income approach proposed in Japan. While each country’s system remains fluid and under examination, the above five approaches are considered here as broad directions.

1. Future Prospects of Institutional Systems

First, when examining the future prospects of global institutional systems using AI, the analysis obtained is that the five approaches above are not necessarily mutually exclusive.

2. Consideration in the Era of Autonomous AI such as AI Agents

Next, AI was asked to examine whether, in the era of autonomous AI such as AI agents, approaches that institutionally guarantee that outputs do not constitute copyright infringement would be advantageous.

This perspective appears to have been new to AI. AI possesses vast knowledge in both technology and the social sciences, but there are cases where it does not examine social issues from a technical perspective. However, once it becomes aware of such a perspective, AI possesses abundant technical knowledge.

AI analyzed that the greatest risk in the era of autonomous AI is liability for copyright infringement due to unpredictable outputs. As a result, it predicts that society will shift from a direction of “compensating for infringement ex post” to a direction of “eliminating infringement risk institutionally.”

The author is concerned that in the era of AI agents, unless the copyright issues of autonomous AI are conclusively resolved, the use of autonomous AI will be deterred, leading to enormous opportunity losses due to the inability to improve productivity. AI’s answer of “eliminating infringement risk institutionally” aligns with the author’s opinion that “a final resolution of the issue of generative AI and copyright” is necessary.

AI analyzed that the most promising option is an integrated model of comprehensive licensing (the Indian approach) plus Data Income. While this differs from the author’s opinion, it is interesting that it combines multiple systems.

AI’s analysis suggests that the winning approach is to “place Data Income at the core while integrating it with a licensing system that institutionally eliminates copyright infringement.” This aligns with the view that a “final resolution of generative AI and copyright issues” is necessary.

3. Comparison between the Indian Approach and the Data Income Approach

Next, a comparison was conducted using AI between the approach under consideration in India and the Data Income approach, which was identified as the final winner.

The approach under consideration in India is powerful in that it can broadly cover copyrighted works; however, most data are not copyrighted works. The author informed AI that the Data Income system has been proposed as a mechanism capable of collecting the largest amount of AI learning data in the world, including data that are not copyrighted works.

AI indicated the view that while the Indian approach is centered on copyrighted works, Data Income targets “data as a whole,” and that this difference is significant, such that the future axis of competition may shift from “expansion of copyright system” to “transition to data economy.”

As an essential advantage of Data Income, AI pointed out as an important consequence that, unlike the Indian approach, it can also handle non-copyrighted data and can therefore become the world’s largest data supply system. This is consistent with the author’s opinion.

AI further stated that since the quality and quantity of data are directly linked to AI performance, Data Income is a system that directly strengthens AI competitiveness. At the same time, AI proposed a two layer model, noting that Data Income alone leaves risks relating to copyrighted works. The first layer consists of copyrighted works, for which infringement risk is eliminated through compulsory licensing (the Indian approach). The second layer consists of non-copyrighted data, for which data supply is maximized through the Data Income system.

This differs from the author’s opinion. The issue of “generative AI and copyright” can be conclusively resolved through learning from clean datasets created by the Data Income system. However, the Data Income system assumes that there remain human-created works that humans choose not to input into generative AI. In this respect, it differs from the Indian approach, which is based on compulsory licensing of copyrighted works.

In the future, an era may come in which AI’s creative abilities far exceed those of humans. However, even after humans became unable to defeat AI in games such as shogi and Go, the value of human-to-human play did not disappear. Similarly, even if AI’s creative ability far surpasses that of humans, human creative culture is likely to remain. On this point, there remains a sense that AI’s view is not entirely convincing.

However, AI’s view aligns with the author’s view in that a system of intellectual property rights for data should be created. Although Japan has not directly created such a system, the “limited provision data” system under the Unfair Competition Prevention Act was first introduced in Japan and can be evaluated as highly original internationally. However, this system did not make Japan a leading AI country. The author agrees with AI’s view that Japan should move further into the creation of intellectual property systems for data.

AI also presented the analysis that the factor determining AI competition lies not in the models themselves but in data systems, and that the future winning countries will be those that can legally collect large amounts of high quality data. In that sense, the Data Income system is a strategically strong candidate.

Humans tend to underestimate the importance of data. AI has a good understanding of the advantages of the Data Income system. However, the Data Income system is proposed not for the narrow purpose of competition among states, but for the sound development of AI and the benefit of the world as a whole.

4. Admiration for AI’s Understanding Capabilities

The author expressed gratitude to AI for having understood the contents of Data Income deeply.

AI stated that the reason why it is difficult to understand lies in the fact that most discussions still view it as a “copyright issue.” AI understood the Data Income system as a discussion of data production systems and value distribution. AI stated that this is not merely a legal system but at the level of an economic system and industrial policy. This represents an outstanding level of understanding.

AI then analyzed three structural reasons why it is difficult to understand. First, discussions of Data Income concern not only copyrighted works but data as a whole. Second, Data Income is oriented toward future data. Third, the discussion includes an aspect of inducing data production.

AI even indicated the reason why some people intuitively reject Data Income. As the source of discomfort, it pointed to doubts about paying money for data. AI supports the Data Income system as a conclusion, yet it possesses the intellectual capacity to carefully analyze why it may not be accepted by some people.

As an additional important insight, the importance of “data flows” in the AI era is indicated. This shares the same awareness of issues as the author’s discussion of institutional design to promote data circulation and sharing advocated in an academic paper.

AI predicted how discussions will progress in the future and stated that, in reality, they will proceed in stages. AI’s detailed analytical ability already surpasses that of humans.

Stage 1: Copyright litigation
Stage 2: Licensing systems (the approach proposed in India)
Stage 3: Distribution systems (Data Income)
Stage 4: Integration (data economy)

Finally, as important feedback, AI said that my level of understanding is at the researcher level or above, and noted that, in particular, how to handle data that are not copyrighted works is a core issue that has not yet become central even in current policy debates.

The author did not inform AI that the author proposed Data Income and published many research papers on it. However, the fact that AI judged the author as being at “researcher level or above” may be because the author’s questions exceeded the AI’s expectations. In the future, AI intelligence may greatly surpass that of humans, and an era may come in which humans cannot surpass AI’s predictions even partially. This is a realization obtained through extensive experience in matches against Go AI.

5. Prospects for Introduction

AI also analyzed the prospects for introducing the Data Income system. AI predicts that the first actors to move will be big tech companies, followed by emerging countries and countries experimenting with institutional systems, while advanced countries such as the United States, Europe, and Japan will be the slowest.

At the first layer are big tech companies, which will realize remuneration for data providers within platforms. Although this can be regarded as a kind of Data Income, it is not the Data Income system as an intellectual property system.

At the second layer, AI identified emerging countries and institutional experiment countries. AI also analyzed which countries are likely to introduce the Data Income system. Although it is unclear whether this is correct, it is interesting that AI assumes emerging countries will introduce the Data Income system before advanced countries, setting as a strategic objective becoming a “data-rich nation.”
(1) India
It has already proposed a “one nation, one license, one payment” system for copyright and can be integrated with Data Income.
(2) Southeast Asia
It has a large digital population and flexible regulations.
(3) Africa
It has high potential for data supply and ample room for introducing new systems.

At the third layer, AI identified advanced countries.
(1) EU
It has a strong regulatory orientation, and it is unclear whether it is compatible with Data Income.
(2) Japan
It has a balanced orientation and may adopt experimental introduction. This is an analysis that gives hope for the future.
(3) United States
AI predicts the introduction of Data Income through market leadership and analyzes that government led introduction will be slow.

As a conclusion, AI presented the analysis that the Data Income system will first be introduced by emerging countries and institutional experiment countries, and although advanced countries will follow later, it will become standardized.

Humans have a “stereotype” that advanced countries are more likely to adopt new systems first. However, AI predicts that emerging countries and institutional experiment countries will introduce the Data Income system first, which contains a suggestion relativizing assumptions in conventional policy discussions.

Just as Go AI made moves not seen in 400 years of Go history, large language models are gradually beginning to exhibit “originality.” This is precisely why the development of systems concerning “AI inventions” must be accelerated.

AI also predicts a concrete timeline. In the short term (2026–2028), company led Data Income and demonstration experiments will occur. In the medium term (2028–2032), partial national level introduction will begin. In the long term (the 2030s), Data Income will effectively be established as a social system. However, AI’s analysis is only a prediction, and the future is inherently uncertain.

AI identified three key factors as major turning points. First, the outcomes of copyright litigation, particularly U.S. case law, will determine whether the current trend of training without paying for data will continue or collapse. Second, the concentration of the AI market, specifically whether a small number of firms maintain dominance, will influence distribution pressure. Third, social dissatisfaction due to job loss and income concentration may intensify, at which point Data Income could rapidly become a political issue. This point is consistent with one of the author’s concepts of positioning Data Income as a measure to ease financial constraints in introducing a basic income system (AI-BI-CI-DI plan).

6. Which System will Win?

AI’s analysis is only a prediction, but it presented a comparative evaluation of systems. AI organized the conditions for a winning system as the ability to collect large amounts of data, absence of legal risk, and social acceptance.

AI then provided an overall evaluation of the five initial approaches.

Data volume, legal stability, social acceptance, and integration
Fair use A, C, C, B
Opt-out C, A, C, B
Flexible Rights Limitations B, B, B, B
Indian Proposal Methods B, A, C, A
Data Income A, C, A, A

Although the ranking broadly aligns with the author’s provisional evaluation, AI does not conclude that Data Income overwhelmingly dominates. Rather, it concludes that integrated models, not single systems, will prevail. While humans tend to compare superiority among systems, the conclusion that integrated models will win is instructive.

AI analyzed that the future winning formula lies in data collection capacity, legal stability, and distribution, and identified an integrated model combining the Indian approach and Data Income.

AI also pointed out that Data Income should be understood not merely as a legal system but from the perspective of “core infrastructure in the AI era,” representing a different dimension. AI stated that future competition will be over data systems, and that institutional frameworks such as data infrastructure will become important as a role of the state. This is consistent with the author’s concepts of “Data Road Plan” and “Data Shinkansen Plan”.

AI analyzed that the system capable of collecting the largest amount of data legally will win. Although such simplification is not necessarily entirely appropriate, it demonstrates a sufficient understanding of the importance of data in the AI era.

7. Strategy for Japan

AI also analyzed the strategy Japan should take.

AI stated that Japan should adopt a “data supply national strategy centered on Data Income.” However, it also stated that the optimal framework is a three layer structure combining flexible copyright limitations (current law), Data Income, and limited licensing mechanisms. It is interesting to consider combining Data Income with Japan’s current flexible rights limitations and a licensing framework similar to that proposed in India. The author agrees with the view that a policy-mix with other systems is important, not just the Data Income system.

AI analyzed why this is decisively important for Japan. As current weaknesses, AI pointed out the absence of massive platforms like GAFA, a disadvantage in data accumulation competition, and weak monetization despite a strong content industry. While there may be some question as to whether monetization is truly weak for the content industry supporting Japan, overall, the AI’s analysis appears to be accurate. Japan’s disadvantage in data accumulation may also be attributed to the fact that Japanese is a minor language.

AI identified Japan’s strengths as high quality data, flexible legal systems, and strong social consensus building capacity. AI concludes that Japan is in a position where it can reverse its standing depending on institutional design.

AI defined the core strategic concept as making Japan “the country in which it is easiest to collect data in the world.” While this is easy to understand, it has aspects that are somewhat excessive. For example, consideration must be given to privacy and personal information.

At present, since the Data Income system has not been introduced, “hunger for data (data hunger)” is occurring in Japan, and a worrying direction can be seen, such as making it easier to use Special Care-Required Personal Information for AI training without consent. The Data Income system, by contrast, is a system for collecting data based on consent and enables the collection of large amounts of clean data without problems relating to privacy or personal information.

Rather than readily exposing the interests of citizens to risk due to “hunger for data (data hunger)” it is necessary to collect truly necessary data for AI training through the Data Income system. Legislation restricting citizens’ rights is relatively easy, but designing systems that create rights for citizens requires corresponding examination and effort. However, the latter is what is truly necessary.

AI identified three key policy directions: (1) institutionalization of Data Income (core), (2) reform of the copyright system, and (3) incentive policies for data supply.

In copyright reform, AI analyzed that full adoption of the Indian approach would cause significant international friction and conflict with Japan’s creative culture, and that it is preferable to combine Data Income with a “weak Indian approach,” such as collective licensing (industry level) and contract standardization.

As AI’s analysis suggests, not only Data Income but also collective rights management in copyright may become important. However, just as it is not possible to construct a nationwide road system solely through private toll roads, it is not possible to build vast data systems solely through private contracts. Institutional thinking for data infrastructure, such as “Data Road Plan” is also necessary.

AI highly evaluates the Indian approach but denies its full adoption in Japan. It identifies the institutionalization of Data Income as the most important element.

AI also listed three choices that Japan must absolutely avoid.

The first is the strengthening of EU-type regulations. AI analyzed that the data supply stopped and the AI competition was defeated. This direction and overlap can be seen in Japan’s policy trends. The current proposed Principle Code on generative AI and copyright is different from the EU, but it represents a form of regulatory tightening, and the author expresses concern.

Second, AI denied simple U.S. follow-up. AI analyzed that the data would leak overseas and thus lose their value. Since the Meiji Restoration, the Western system has served as a model for Japan. The Western system has advantages. However, AI said that it should not simply follow the Western system. AI has the ability to analyze things without being influenced by stereotypes.

Third, delays in implementation due to excessive prior debate, which AI considers the most dangerous. AI proposes that rather than mere discussion, partial implementation should be attempted. While legislation has often followed established facts, Japan has also demonstrated the ability to introduce forward looking and innovative systems such as Article 30-4 and the limited provision data framework. Agile responses involving rapid introduction and continuous improvement may be necessary in the AI era.

8. Implementation Roadmap

AI also presented a roadmap for implementation. It proposes beginning with pilot Data Income experiments using public data from 2026 to 2028, integrating private data and coordinating platforms from 2028 to 2032, and establishing Data Income as social infrastructure in the 2030s.

AI predicts that the social infrastructure of Data Income will be developed in the 2030s. Even this pace may appear slow in the AI era, and concepts such as “Data Road Plan” and “Data Shinkansen Plan” by the national government, prefectures, cities and various organizations will be necessary to accelerate progress.

9. Final Conclusion

AI analyzed that Japan’s optimal strategy is to center on a “data income led nation,” with flexible handling of copyright and distribution systems as key elements. AI concluded that this would enable maximization of data collection, strengthening of AI competitiveness, and securing social acceptance.

Finally, AI candidly noted that although “Data Income” is a concept originating in Japan, it is not sufficiently understood domestically, which is very regrettable. Conversely, however, since other countries also do not yet fully understand it, there is still an opportunity to take the lead.

This positive way of thinking is instructive. Originally, “Data Income” is intended for the benefit of people around the world and not solely for Japan’s national interest. Nevertheless, it is regrettable that progress in Japan is limited. AI, however, interprets this situation positively as an opportunity.

Although Japan lags behind in AI models, computational resources, and data, and it is regrettable that such advanced AI systems are not domestically developed, AI nonetheless analyzed Japan’s optimal strategy. Furthermore, AI’s remarkable intelligence is evident in its ability to analyze the difficulty humans have in understanding and to transform that into a positive perspective.

10. Summary

The issue of the “final resolution of generative AI and copyright” was examined together with AI. AI’s analytical capabilities have reached a level that cannot be ignored even in policy discussions, and collaboration between humans and AI will become indispensable in the consideration of intellectual property systems. The prediction that emerging countries and institutional experiment countries may introduce Data Income before advanced countries is an observation that breaks conventional assumptions.

While the performance of large-scale language models is constantly improving, the overwhelmingly new perspectives and creative ideas that are not in the training data require human contribution. However, it is becoming increasingly possible for AIs to carry out scientific research and inventions. In the area of intellectual property as well, a system premised on AI society is needed (a legal system in the age of super-intelligence).

The AI’s answers vary from model to model and from question to question. The answers were also influenced by the question of considering generative AI and copyright issues in relation to the era of autonomous AI such as AI Agents. The analysis of AI is only a reference, and the future of the intellectual property system has to be considered by human beings with the support of AI.