Integration of Soft and Hard Laws: Profiling Legal Protection for “AI for All”

3.1.1 AI Technology for All

“AI technology for all” implies that AI should be both open and affordable to everyone. Openness suggests that AI should be made as open-source as possible and not monopolized by a small number of technology giants; at the same time, it necessitates improving the public’s understanding of AI. Affordability means that the cost of deploying and using AI should not be set so high that only a small elite can bear it. This approach is reflected in various initiatives. For example, the nonprofit organization AI4ALL focuses on educating and empowering underrepresented groups to ensure broad benefits of AI (AI4ALL 2025); Greece launched a similar “AI for All” initiative to train civil servants in both basic and advanced AI skills (Panagopoulos 2024). and the Bill & Melinda Gates Foundation has emphasized “ensuring access to AI for all,” particularly in low- and middle-income countries (Bill and Melinda Gates Foundation 2025). Furthermore, many nations’ policies and legal frameworks stress making AI technology open and affordable, reflecting the principle of “AI technology for all.” For example, the European Union has funded initiatives such as “AI4EU” to encourage the development of inclusive AI applications and, through its digital policies, has encouraged member states to strengthen AI-related education and training in order to narrow digital divides (European Commission 2019). Beyond EU-wide legislation, individual member states have referred to ensuring that “no one is left behind” in their national AI strategies, as exemplified by Finland’s free AI education program for its entire population (European Commission 2019). The Global Digital Compact likewise emphasizes open data, open AI models, open standards, and open content, ensuring that societies and individuals can use digital technologies for their developmental needs (United Nations 2024). The Asilomar AI Principles also underscore that AI technologies should benefit and empower as many people as possible (Future of Life Institute 2017).

3.1.2 AI Justice for All

“AI justice for all” underscores both formal justice and substantive justice. Formal justice requires that AI not discriminate against specific populations, demanding the prevention of bias and the fair treatment of all users, while substantive justice calls for AI not to exacerbate social inequality but instead to provide equitable opportunities for people of all social classes, races, genders, and economic backgrounds. This broader notion of substantive justice aligns with ideas from thinkers such as John Rawls, who emphasizes fairness in opportunity distribution (Rawls 1999). India’s 2018 National AI Strategy, spearheaded by the NITI Aayog, exemplifies this emphasis on substantive justice: its slogan “AI for All” aims to use AI in driving sustainable and inclusive growth that transcends class, geography, and economic constraints (NITI Aayog 2018b). Besides, many discussions similarly demonstrated the notion of “AI justice for all”. The 2019 “Beijing Consensus on AI and Education” encourages countries to ensure that the design of AI adheres to ethical norms, avoids discrimination, and remains fair, transparent, and auditable, thereby reflecting the significance of justice in AI for all (UNESCO 2019). In 2019, the OECD adopted AI governance principles, and were subsequently recognized and adopted by the G20. The very first principle is “inclusive growth, sustainable development, and well-being,” urging stakeholders to promote AI in a responsible manner that benefits all and highlights substantive justice (OECD 2019). IEEE’s “Global Initiative on Ethics of Autonomous and Intelligent Systems,” launched in 2016, published its first version of “Ethically Aligned Design” in 2019. This guide promotes human-centered, welfare-oriented AI design concepts and emphasizes fairness, transparency, accountability, and thus both include the notion of formal and substantive justice for “AI for All” (Institute of Electrical and Electronics Engineers 2019). In the same year, the EU High-Level Expert Group on AI proposed a “human-centric AI” approach that respects European values and human rights, its guidelines for “Trustworthy AI” outline seven key requirements, in which diversity and non-discrimination, as well as social well-being, echo “AI justice for all” (High-Level Expert Group on Artificial Intelligence 2019).

3.1.3 AI for All Countries

“AI for All countries,” especially developing countries, is equally critical. Global justice theorists like Thomas Pogge and Charles Beitz believed the need for equitable distribution of resources and technology across national boundaries to prevent new forms of deprivation in less developed regions (Pogge 2002). UNESCO’s vision of “AI for All” similarly includes preventing the exacerbation of technological disparities within and among nations (UNESCO 2024). One global organization has sought to ensure no country is left behind by establishing an AI for developing countries forum, illustrating that such initiatives aim to make AI accessible and beneficial on a worldwide scale (AI for Developing Countries Forum 2025). The notion of “AI for all countries” is also reflected in several areas. Emerging multilateral collaborations such as the Global Partnership on AI (GPAI), established by dozens of countries in 2020, aim to ensure that AI benefits different cultures and economies through cross-border cooperation (Global Partnership on Artificial Intelligence n.d.). The African Union has formulated a regional AI strategy framework to harness AI’s potential in education, agriculture, and other sectors for promoting inclusive development (African Union 2024). The Global Digital Compact likewise highlights the need to “build capacity, especially in developing countries, and facilitate access to, development of, and management of AI systems for sustainable development.” (United Nations 2024).

3.2.1 Law and Economics

Originating with Ronald Coase, Law and Economics also provides a legitimacy for “AI for All.” In the view of Judge Richard Posner, one of the leading scholars in this field, Law and Economics is a discipline that uses economic principles to explain legal issues (Posner 2014). In fact, the economic analysis of law primarily aims to improve market efficiency and reduce transaction costs. Yet, greater efficiency does not necessarily mean a better society (Sen 1992). Consequently, while the perspective of economic analysis is neither a sufficient nor a necessary condition for “AI for All,” this section argues that, even from within its own framework, “AI for All” has economic legitimacy because it can reduce market externalities and enhance market efficiency.

From the standpoint of Law and Economics, whether a particular service should be supplied by the market or regulated by the government depends on whether it constitutes a public or quasi-public good, and whether there is a market failure necessitating government intervention. For centuries, the market economy has worked wonders. Under Coase’s theorem, in a world with zero transaction costs, the market naturally arrives at an optimal allocation, and the law itself has little to do (Coase 1960). Therefore, absent compelling reasons, a service should generally be left to the market – unless it constitutes a public or quasi-public good or unless there is a market failure. The former rationale derives from Paul Samuelson’s paper, “The Pure Theory of Public Expenditure,” while the latter follows Pigou’s explanation in The Economics of Welfare, where government intervention may enhance overall social welfare in the presence of negative externalities that cause market failure (Medema 2020). Thus, one should first determine whether AI belongs to the category of public or quasi-public goods. If it does, then government intervention is warranted. If it does not, generative AI services should in principle be left to the market, unless there are market defects that hamper efficient supply. When the market cannot correct such supply inefficiencies and negative externalities on its own, the government should intervene to provide or regulate these services or products.

First, it is acknowledged that, at present, generative AI services do not possess the non-rivalrous and non-excludable attributes of public or quasi-public goods. Unlike typical public goods such as national defense or public healthcare systems, generative AI services are often provided by private entities and can be made exclusive through access controls, payment mechanisms, and similar measures. They may also be subject to congestion or degradation in quality if overused, reflecting their rivalrous and excludable nature. For example, certain OpenAI models require user registration and membership subscriptions, indicating that they are not non-rivalrous; server capacity is limited, so an overload of users could result in outages, showing that they are not non-excludable. Therefore, generative AI services do not fall under the category of public or quasi-public goods. In principle, these services tend to be supplied by the market. However, if the market’s pure pursuit of profit ignores the broader public interest, it may lead to inefficient supply or harm to social welfare.

Second, in the long run, failing to implement AI for All will exacerbate existing societal divisions and creating economic negative externalities, ultimately undermining the environment for efficiency and innovation. From an externality theory perspective, externalities arise when the production or consumption of a good or service generates benefits or costs for others that are not fully reflected in market transactions (Medema 2020). If generative AI services depend entirely on market supply based purely on ability to pay, certain social groups may be excluded or underserved. This exclusion, while potentially efficient from a narrow, short-term private market perspective, can contribute to negative externalities in the broader economy. Examples include widening the digital divide or marginalizing underdeveloped regions and vulnerable populations, which can lead to deeper social inequality (Ziewitz, Fourcade, and Boyd 2021). As structural inequalities intensify, this may compromise social stability. Such instability, can ultimately increase business operating costs, deter long-term investment, hinder overall productivity, and thus deteriorate the very market environment necessary for sustained efficiency and innovation. Market mechanisms alone often lack sufficient incentives to fully address these externalities, suggesting a potential need for government intervention, regulation, or other collective actions to mitigate these risks.

Finally, as Figure 1 illustrates, when initiatives promoting broader access and utilization of AI, aligning with the principle of “AI for All”, are implemented, the marginal social benefit (MSB) of AI usage likely exceeds the private benefit captured by market transactions alone. Consequently, the socially optimal level of AI adoption (where MSB intersects supply) is greater than the market equilibrium – reflecting uninternalized positive externalities. If service providers proactively work towards the goals implicit in “AI for All,” whether motivated by policy guidance or their own initiatives (for example, by making foundational aspects of the technology more open, keeping prices for basic access affordable, disseminating knowledge widely, or providing training), these actions can help realize and internalize positive externalities, generating substantial societal benefits beyond the direct users. When more people gain meaningful access to generative AI services, they can drive innovation in healthcare, education, agriculture, environmental protection, and other fields, which in turn yields widespread economic and societal gains.

Figure 1:

Positive externalities on AI for all.

3.2.2 Justice Theory

Justice theories provide a moral foundation for “AI for All,” insisting that AI development and use be fair and inclusive. Both formal justice (impartial and consistent rule application) and substantive justice (fair outcomes regarding distribution of benefits and burdens) are relevant. Philosophers like John Rawls and Amartya Sen offer principles to evaluate AI ethics and policy.

Rawls’ theory of justice as fairness emphasizes equitable distribution of social goods and opportunities. Two key Rawlsian ideas apply to AI ethics: fair equality of opportunity and the difference principle. Fair opportunity means positions and services (like AI-driven benefits) should be open to all under conditions of equality, which in AI translates to non-discrimination by algorithms and equal access to AI tools. The difference principle holds that inequalities are acceptable only if they benefit the least advantaged members of society (Westerstrand 2024). In an AI context, this implies AI systems and policies should be designed to improve outcomes for marginalized groups rather than worsen gaps. So, if AI increases productivity or wealth, Rawlsian justice would argue for mechanisms that ensure disadvantaged communities share in those gains. For example, AI decision systems should not only avoid bias but proactively help those worse off – echoing Rawls’ focus on uplifting the worst-off (Gabriel 2022).

Beyond Rawls, Amartya Sen and Martha Nussbaum’s capabilities approach enriches the conversation by stressing that real freedoms – capabilities – should be expanded across communities. The capabilities approach holds that true well-being lies not in what people have, but in people are able to do and capable of achieving (Robeyns 2017). This approach shifts the focus from abstract resources to people’s actual freedoms and abilities to achieve well-being. “AI for All” from Sen’s perspective means ensuring that AI technology expands people’s capabilities – e.g. the ability to get an education, communicate, or obtain healthcare – for everyone, not just the elite, to ensure people have the real capability to benefit from AI in their daily lives. Sen warns that pure economic growth or tech advancement can mask growing inequalities in capabilities. In the AI era, there’s a risk of a large underclass dominated by a techno-elite if access to AI tools and digital skills is uneven (Yoon 2021). We see this in concerns that advanced AI could primarily empower wealthy corporations or countries, leaving others behind. Capabilities approach in AI ethics thus advocates for policies that empower marginalized populations with AI. The capabilities approach in AI ethics therefore advocates for policies that empower marginalized populations with AI – by investing in AI literacy, multilingual AI assistants, or affordable AI services that enhance the practical reasoning of marginalized groups, expanding their ability to perceive, imagine, and think. As one scholar notes, Sen’s framework can guide us to establish social conditions where all persons can flourish in the age of AI (Yoon 2021). In summary, justice demands that AI development be assessed by how well it expands human freedoms and reduces capability deprivation.

Taken together, justice theory validates the moral imperatives behind “AI for All”: not only should AI be non-discriminatory, it should actively work to redress inequalities across diverse social, economic, and cultural contexts.

3.2.3 World-Systems Theory

While legal economics and justice theory mainly shed light on domestic contexts, world-systems theory places “AI for All” within a global and historical framework. Originating from the work of Immanuel Wallerstein, world-systems theory dividing the world into core, semi-periphery, and periphery nations (Wallerstein 2004). It provides a macro-level perspective on AI accessibility and technological divides between developed and developing countries. World-systems theory exposes how core nations unjustly maintain their advantages to extract benefits, highlighting the necessity of “AI for All Countries”. The meaning of “AI for All Countries” emphasizes the responsibility of AI-leading core nations toward peripheral and semi-peripheral countries and serves as a crucial idea for breaking the cyclical inequalities embedded in the global AI system.

In world-systems terms, AI increasingly become a new mechanism of unequal exchange between core and periphery, reinforcing dependency. Wallerstein’s analysis sees wealthy core countries dominating high-tech innovation and benefiting disproportionately, while poorer periphery countries provide raw materials or cheap labor and lag in development (Hopkins and Wallerstein 1982). This dynamic is evident in AI today. The US concentrate AI research labs, cutting-edge tech companies, and infrastructure, exporting AI products and services, whereas many developing nations mainly consume AI technologies or supply data and cheap data annotators. A study by the World Economic Forum points out that the economic and social benefits of AI are geographically concentrated primarily in the Global North (Yu, Rosenfeld, and Gupta 2023). Another study provides empirical evidence of this disparity, noting that AI capital stock from 1995 to 2020 is heavily concentrated in developed regions, exacerbating wealth inequality (Smith and Lee 2024). Thus, the “AI divide” mirrors the digital divide: core countries “dominate in … technological advancement, while periphery countries are often exploited for their resources” (including data and cheap labor) (Couldry and Mejias 2019).

Giovanni Arrighi’s work on long-term economic cycles and hegemonic powers can be used to explain how the AI revolution could be seen as part of a new cycle of accumulation: today’s leading powers (the U.S.) compete for dominance in AI, while other nations risk falling behind (Arrighi 1990). However, Arrighi also noted opportunities for semi-periphery states to rise by leveraging new technology (Arrighi 2007). Nevertheless, world-systems theorists would caution that without deliberate intervention, AI’s benefits will concentrate in existing centers of power. The flow of AI talent and capital is mostly from periphery to core. Global income and knowledge gaps may widen as AI increases productivity primarily in tech-savvy economies. Arrighi might interpret initiatives like China’s massive AI investments as attempts to restructure the system, potentially creating a more multipolar tech world (Arrighi 2007). But for many low-income countries, the worry is becoming even more peripheral in an AI-driven global economy.

4.1.1 Price Cost Eroding “AI for all”

Over the past few years, Generative AI has advanced significantly, marking a paradigm shift in both capability and social impact. Unlike earlier AI systems, which were limited to specific tasks such as facial recognition or recommendation algorithms, Gen AI can perform generative tasks, such as writing novels and generating code. This leap in generality and autonomy is exemplified by models such as OpenAI’s GPT-4.5 and O3, Anthropic’s Claude 3.7, and Google’s Gemini 2.5 Pro, which demonstrate reasoning, creativity, and decision-making capabilities that rival or surpass human expertise. Moreover, state-of-the-art models exhibit remarkable proficiency in complex tasks; for example, OpenAI’s “deep research” function autonomously conducts comprehensive internet research, achieving a 26.6 % accuracy rate on the “Humanity’s Last Exam” benchmark, comparable to human performance (TechRadar 2025). As Gen AI capabilities continue to evolve rapidly, even more advanced systems are under development.

However, the price of access to most advanced generative AI tools has also undergone a significant leap. While some advanced generative AI models remain open-source or freely accessible to the public – among which DeepSeek R1 stands out prominently – venture capitalist Marc Andreessen described its emergence as the “Sputnik moment” in the AI field. Nevertheless, DeepSeek R1 still trails behind the most cutting-edge large models (Mozur 2025). The most cutting-edge models are unaffordable for the majority of people, posing a substantial challenge to the vision of AI technology for all. The advanced capabilities of Gen AI models – such as OpenAI’s GPT-4.5 and O1 pro – have been matched by steeply tiered pricing that entrenches inequality. OpenAI’s 200/month “Pro” tier, for instance, grants 120 “deep research” queries, while free users are limited to two – a disparity that extends to API costs, where high-volume access to models like ChatGPT 4.5, 150 dollars per million output tokens. Even mid-tier subscriptions (e.g., Grok at 30/month, ChatGPT plus at 20/mouth) remain unaffordable for billions globally, particularly in regions where 20 represents a substantial share of income. And OpenAI is exploring the pro max subscription-2000/month level. As showed in Table 1, even the mid-tier subscriptions divided about 2–15 % of their mouth income. As for “pro” lever, occupied 20%–150 % for developing countries’ people’s income, it’s impossible for most developing countries people to afford.

As a result, the leap in capability of advanced Gen AI also brings about a leap for the person who uses it (Cheng and Gong 2024), significantly impacting the ideal of “AI for all”. When it comes to “AI justice for all,” this trend deviates from the principle of substantive justice: creating a capability caste system, wealthy users and corporations harness Gen AI for innovation, education, and market dominance, while marginalized groups – students, small businesses, and low-income nations – are relegated to slower, less intelligent, outdated, or restricted versions. Individuals with access to the most advanced Gen AI can transcend conventional skill boundaries and even evolve into “super-individuals.” For example, OpenAI’s Deep Research function can effectively replace a research assistant, enabling users to complete literature reviews or data analyses within minutes – far outpacing the productivity of human researchers (Reuters 2025). According to a now-outdated McKinsey study, earlier versions of generative AI could already boost information retrieval efficiency by 400 % (McKinsey Global Institute 2023). Plus, research also indicates that Gen AI significantly enhances learning capacity, and managerial skills, widening capability gaps exponentially rather than incrementally (Cornell University Center for Teaching Innovation 2024). This exponential advantage consolidates existing social hierarchies, reflecting the broader dynamics of digital private power concentration (Sun and Xiao 2024). A wealthy student paying 200 dollars per month for AI-based assistance can produce work once requiring a professor’s guidance, while an entrepreneur uses AI to generate patent-level solutions – turning technical privilege into capital gains. These “super-individuals” are not created by innate talent or diligence but by the ability to pay, stifling social mobility and reinforcing a cycle in which “capital empowerment” fuels “technology empowerment,” leading to “capability enhancement” and ultimately more “capital accumulation.”

4.1.2 Literacy Divide and Infrastructure Condition Undermining “AI for all”

A critical yet underrecognized obstacle to “AI for all” lies in cognitive disparities – the gap between those who understand Gen AI’s potential and those who remain unaware of its existence or utility. A study by the St. Louis Fed shows that high-education workers use depth 3.2 times more than low-education groups, leading to a 15 % increase in the productivity gap (Federal Reserve Bank of St. Louis 2025). Gen AI is forging a new digital divide concerning not only who can log on or possesses technical digital skills but also who can comprehend, create, or control AI technologies (Hendawy 2024). This knowledge gap is compounded by AI literacy: While tech-literate populations leverage AI for tasks ranging from coding to career advancement, marginalized group often lack even basic awareness of these tools. Research have shown, the most vulnerable groups with the lowest levels of AI knowledge and AI skills were mostly older, with lower levels of education and privacy protection skills, than the average users (Wang, Boerman, Kroon, Mller, and H de Vreese 2024). Without targeted literacy programs, this cognitive barrier entrenches a hierarchy of opportunity, where the technologically initiated reap AI’s benefits, leaving others further behind in education, employment, and civic participation.

Plus, the promise of “AI for all” collapses where basic infrastructure is absent. Globally, 2.6 billion people remain offline (International Telecommunication Union 2023), – preconditions for using Gen AI. Even when devices are available, computational demands pose hurdles: running models like GPT-4.5 requires stable high-speed connectivity, which remains inaccessible in areas with bandwidth limitations. Without universal infrastructure investment, Gen AI’s transformative potential remains confined to urban, affluent enclaves, deepening global inequities.

In short, traditional digital divides – concerning both AI literacy and digital infrastructure – are simultaneously creating and exacerbating a new “AI divide,” making previously disadvantaged groups even more vulnerable. This significantly undermines the three core ideals behind “AI for all”: Firstly, individuals remain unaware of AI technology and have no means to access it, let alone benefit from “AI for all.” Secondly, marginalized communities become further disadvantaged, making substantive justice impossible. Thirdly, at present, it’s more accurate to describe the situation as “AI for certain countries” rather than “AI for all countries.”

5.2.1 Measures for Implementing Hard Law

Currently, generative AI services are neither be seen as “public goods” nor “quasi-public goods” – markets remain the primary drivers of innovation and supply. However, “AI for All” carries significant public value, and overreliance on market forces risks exacerbating structural inequalities and other negative externalities. Consequently, the critical role of hard law lies in two dimensions: first, clarifying the boundaries of governmental public responsibility to avoid oversimplifying AI as a “free public good,” and second, deploying incentive-based legal mechanisms to correct market failures, social inequalities, and other negative externalities, thereby promoting broader and more affordable access to AI services for the public.

Regarding AI price barriers, multiple hard law measures could be considered. Firstly, incorporating AI services into government-guided pricing should be considered. Various countries have historically employed price regulations for essential public commodities and services. For instance, in the U.S., the Hepburn Act (1906) established maximum railroad freight rates. In China, the Pricing Law mandates governmental pricing or guided pricing for critical commodities affecting national welfare, including medicines, energy, and transportation. Similarly, Japan’s Price Control Act permits governmental regulation of commodity and service pricing during emergencies. Other nations such as Germany, France, and the UK also regulate pricing in sectors like energy, telecommunications, railways, and healthcare to varying extents. For AI services, if their societal and economic significance continues to grow, governments may consider subjecting them – in whole or in part – to guided pricing. Specific measures could include setting price ranges based on per capita income levels or AI’s impact on public services and safety, thereby preventing exclusionary costs from barring public access to core AI applications. Secondly, promoting “AI for All” through government procurement could be implemented. Most countries have government procurement laws regulating public purchasing behaviors. AI procurement agreements could help negotiate better pricing through economies of scale (McGee 2025). By procuring advanced large-model APIs and then providing them to education, healthcare, and small and medium-sized enterprises, this approach would exert less market impact compared to government-mandated guided pricing.

To enhance citizens’ AI literacy and bridge the AI divide, multiple hard law approaches exist. Firstly, in public service assurance law, free AI services could be provided through public cultural institutions. Various national legal frameworks regulate public service facilities like museums and libraries, covering their construction, management, operations, funding, and public accessibility. For example, the U.S. Museum and Library Services Act of 1996 outlines funding and operational guidelines to ensure public access. Under the “AI for All” framework, countries could revise or reinterpret such laws to empower public institutions – using fiscal funds – to procure commercial AI services and offer free or low-cost access to the public. This approach would alleviate the financial burden on low-income or marginalized groups, enabling them to utilize advanced AI models through public facilities without requiring expensive personal devices or subscriptions. Secondly, in education law, revising or interpreting educational legislation to formally integrate “AI literacy” and “digital skills” within the scope of the “right to education” is advisable. Educational laws should explicitly incorporate “understanding and applying generative AI” into defined educational objectives. For general adult education and lifelong learning frameworks, community colleges, vocational training centers, and online platforms should be encouraged to provide AI literacy programs, ensuring ongoing learning opportunities across individuals’ life stages. For example, California’s recent law requiring schools to teach students about AI signifies a growing recognition of the importance of equipping young people with these essential skills (Jones 2024). Furthermore, legislation governing educational facilities should include broadband networks, smart terminal devices, and cloud computing resources as essential educational infrastructure, clarifying spending proportions or minimum standards in terms of network coverage rates and per capita device ownership.

5.2.2 Differentiated National Pathways: From Core to Peripheral AI Technology States

The effectiveness of hard law in realizing “AI for All” depends on a nation’s possession of AI technological capacity. In other words, universal access to AI presupposes the existence of domestic capacity to provide “AI for all”. If advanced large language models reside outside a nation’s borders, and domestic markets and industries lack influence, efforts toward “AI for all” will falter. For this reason, while outlining potential hard-law institutional approaches, this section also categorizes four types of nations based on their distinct inclusivity challenges and corresponding response strategies. Drawing on world-systems theory and the current global division of labor in AI technology, nations can be categorized into Core, Near-Core, Semi-Core, Semi-Peripheral, and Peripheral AI technology states. Countries at different tiers face stark disparities in technological sovereignty, market scale, international competitiveness, and regulatory capacity when advancing “AI for All” through hard law, necessitating distinct legislative priorities (Table 3).

Firstly, for core AI technology state, The U.S. possesses the world’s most robust AI ecosystem, leading in large-scale model development, chip design and manufacturing, foundational research, and industrial applications, with full AI sovereignty. It can effectively implement “AI for All” through domestic hard law – for example, by adopting government-guided pricing, implementing targeted subsidies, strengthening antitrust and data governance, and mandating AI literacy in education laws. The legislative challenge lies in balancing two goals: maintaining incentives for innovation and global competitiveness in the tech sector while preventing the monopolization of technological dividends by a few corporations or privileged groups, which would undermine societal equity.

Secondly, China, the Near-Core AI Technology State, ranks second globally in foundational model development, domestic AI chip R&D, academic output, and application scale, with partial AI sovereignty. However, it remains dependent on foreign suppliers for certain high-end chips and core algorithms. Domestic hard law can play a significant role in shaping national AI capacity and improving accessibility, particularly in balancing innovation incentives with inclusive access.

Thirdly, these semi-core AI Technology States nations (e.g., South Korea, Japan, Germany) exhibit active R&D and industrial ecosystems, moderate capabilities in large-scale model design and chip manufacturing. While excelling in specific dimensions, none matches the U.S. or China in comprehensive strength across models, chips, computing power, research, applications, and policy investment. For “AI for All” legislation, reliance on foreign core technologies and models complicates efforts to enforce pricing controls or subsidies, as decisions by multinational corporations often dictate outcomes. Nonetheless, domestic antitrust laws, data governance frameworks, and labor regulations can still promote fairness and affordability in AI adoption, albeit with less impact than in Core or Near-Core states.

Fourthly, for semi-peripheral AI Technology States, these nations (e.g., India, Brazil, Indonesia) lack advanced large-scale model design capabilities and are typically large emerging economies or developing countries actively deploying AI. While they possess some technical and talent infrastructure, their capacity for high-end chip production, cutting-edge research, or large-scale industrial AI applications remains limited. Hard law can regulate domestic AI pricing, public service access, and educational guarantees, but challenges persist due to reliance on foreign technology. For example, “long-arm jurisdiction” struggles to influence pricing or platform strategies of foreign firms in local markets. Thus, international cooperation – via regional or multilateral platforms – is critical to facilitate technology transfer, public AI infrastructure development, and global/regional coordination on AI education and training.

Finally, most developing nations fall into peripheral category, lacking advanced research institutions, AI industrial bases, or capabilities in foundational model development and chip manufacturing. Their AI adoption is limited to off-the-shelf solutions provided by foreign entities. Domestic hard law has minimal leverage to counter transnational monopolies or high-cost services, as they lack control over core technologies. The solution does not lie in hard law, but rather depends heavily on international organizations, multilateral aid, or treaties to integrate “AI for All” into global development agendas. This includes securing preferential or free access to AI tools for low-income nations, coupled with infrastructure and educational support. At the same time, efforts should also be made to expand electricity and internet access and provide training in AI skills (Nugroho and Cammelli 2024).